Duolingo should be on borrowed time, as a method to learn a language it’s inefficient to the point of being ineffective - it seems to serve as an acceptable reason for fiddling with one’s phone, like a Guilt-Eze pill.
I spent 18 months on it and the learning was soundly thrashed by three weeks in a classroom, and for all the pupils, not just me, primed for success by 18 months with Duolingo.
Unless anyone has had a different experience, Duolingo is an exercise in self-deception. No doubt there will be people in difficult situations that have managed just fine with Duolingo, or in spite of it, but it doesn’t feel like a growing market.
One thing that I deem a falacy though, is comparing it with e.g. an expensive paid for classroom or personal lessons. It may well fail in such direct comparison, and no surprise!
People then use that comparison to claim "it's worthless", and I disagree with that.
If the claim is "30min of structured learning with duolingo will get you no closer to learning a language than 30 min of mindlessly scrolling facebook cat videos", I don't believe that's true. And THAT is the correct comparison :)
One thing to note though, is how you use it - note I used the term "structured learning" in previous sentence, not "doing match madness for points" :)
Personally, I spent first few months learning on duolingo, and it was wonderful; had some small chats with people who speak that language at work, occasionally wrote a sentence or two in that language in emails, etc. Then I spent next few months preserving the streak. I did the challenges and matches and got thousands of points and first places... but learned nothing. In fact, my knowledge deteriorated.
Then after 200 days, I lost my streak and... breathed a sigh of relief. I am now making sure I do the lessons and practice and learn, and purposely avoid the competition part of it or activities which give many repetitive points but don't increase my knowledge. My familiarity with language is increasing again and I can make more and more sentences on my own.
> I have a 297 day streak right now and I just feel like slave to the owl. I’m going to get to 365 and then purposely break the streak to free myself.
Why do you even care? It's an artificial count by a for-profit company with an Owl logo that doesn't even know you exist outside of a few database rows.
You've just self-described it as slavery, why would you keep doing this to yourself?
Sure, the count is tracked by the company Duolingo, but that's irrelevant. The count is also tracked by the user, and that's what's important. Even more importantly, the commitment to keep to the count is made by the user. It's their commitment, not Duolingo's. Therefore, it's not relevant that Duolingo encouraged that commitment and benefits from it. The user is still going to be subject to the guilt and shame that comes when humans break our own commitments, and that's what makes quitting easier said than done.
It's kind of like telling a smoker to just quit because it's the smoking company who benefits. Yeah, sure, but the addiction is internal. They're not just selflessly smoking to help Philip Morris.
It never ceases to amaze me how much people discount the existence of things because they "only" exist in the minds of those who believe in them.
Nobody has yet shown me a flawless proof that they themselves exist, or anything else, and yet somehow, I manage to get through my days just fine based on the versions of things in my mind that correspond to the empirical phenomena that I experience. And occasionally those mental constructs change with new empirical data even.
One of the things that enables civilization and not dying of disease living in your own filth is a certain vague appreciation for order that comes out in certain ways for different people… like the semiobsession with streaks which plenty of businesses take advantage of.
Lots of people harness this for their own good with streaks for things that are really actually valuable for them.
One thing I've heard here in HN, is that people LOVE to complain and bemoan their situation, and willingly abrogate their power and choices to some inanimate policies and corporate choices.
The GP's comment was that they were effectively 'slave' to a skinner box, and even recognized it. And yet, refusal to do anything about it.
I see it again and again. The answers are there, and a bit harder, but yeah.
If I make it to a full year I'll at least feel like maintaining the streak as long as I did was in service of a specific milestone goal.
After a year the milestones become less meaningful (400 days? 500? 700? meh) yet harder and harder to break the longer you go. A friend recently hit 1000 days in Duolingo--really hard to give up once you've gone that far.
Devices can't literally own a person, so it's not the correct use.
I don't like being nitpicky, but if we're going to argue about the "correct usage" and not accept colloquial usage has perverted the term, I may as well set the record straight:
>a person who is forced to work for and obey another and is considered to be their property; an enslaved person.
fair enough, I thought that definition was unofficial or marked as slang.
>No such thing has happened, words can have both literal and figurative meaning.
yes, like "literally" having the definition of "figuratively". I would indeed call that a perversion:
>to divert to a wrong end or purpose
most of the time that is at best marked as slang (such as use of a double negative to mean a negative, and not a soft positive), but I guess that's not my call to make.
Hmmm. Personally when the non-disambiguated wikipedia page of a specific word starts with, "$WORD is the ownership of a person as property, especially in regards to their labor." and includes a photo of a physically mutilated other person, then maybe there are reasons to be a bit more careful since it's very easy to pass the wrong message.
If you get daily language exposure some other way then why not. But if the owl is the only thing ties you to your daily exposure then you'll have nothing.
That's how human brain works through constant exposure. If you don't care learning the language and honest with yourself about it then it's fine too. But this topic is mostly about the people who do want to learn the language.
to learn another language it's critical you use it as often as possible, in an ideal world you want to completely swap out your native language for the one you're learning, that's the reason for the streak, to keep you engaged in the target language to help you learn, it's not like a snapchat streak which is completely pointless
I've had the same issue with the Fitness app on my smart watch (and others like Readwise). I wanted to keep the streak alive for the sake of it, but I was starting to get hurt. For Readwise, it was just becoming mindless daily review.
Thankfully I got out of it and am much more consciously breaking streaks.
The downside is that I've clearly been exercising a lot less since I stopped focusing on the streak, so it works in some ways…
We should have a way to adjust what we consider a streak: maybe a daily streak is important to you, but maybe 3 times a week is what matters to you. Having the option for each app would be great.
Yeah, daily streaks make no sense for strenuous exercise. Everyone needs a rest day. Makes more sense for food, though. Since you usually never skip a day of consumption.
I'm not entirely sure if I feel that I have to since the streak doesn't really correspond to a calendar year so much as it's "a year", but I think I will indeed go to 366 anyway just for good measure, unless Duolingo specifically recognizes 365 as a milestone.
> If the claim is "30min of structured learning with duolingo will get you no closer to learning a language than 30 min of mindlessly scrolling facebook cat videos", I don't believe that's true. And THAT is the correct comparison :)
The correct comparison would include the effect size on language learning of using Duolingo instead of scrolling through cat videos. The effect size appears to be close to zero. And without considering in the analysis the long-term negative effects on the self-perceived ability to learn a language after spending hundreds of hours using Duolingo and being able to simply say "Hola" in Spanish, which may make the use of Duolingo a net negative on lifetime language learning.
But at that point a person has to take some responsibility. I don't think one can spend a hundred hours learning on Duolingo and not get beyond Hola.
Can a person find a way to waste hundred hours on Duolingo? Sure. I got perilously close :). But after a tenth of that I could generate sentences for friendly emails and at beginning meetings etc when I actually tried to learn.
Has the "100 hrs and not gotten beyond hola" been your personal experience or somebody you know? Trying to understand if I'm arguing with a strawman / straw scenario, or actual occurrence. If latter I'd be genuinely extremely curious to hear a lot more! What were the 100hrs spent on?
But among the people (about 20) who have told me they have used Duolingo to (try to) learn my native language, I have yet to find one who can say more than a few disconnected words. And among the people I have seen who are fluent in my native language or the other languages I speak, I have yet to find anyone who said they used Duolingo to learn that language. Am I extrapolating too much? I think the sample size was pretty good.
I don't use this app and I get the general sentiment but I can tell you one thing about effect size: the effect size of a comment that says you only learn "hello" after 100s of hours is 0!
That's not how effect sizes, in statistics, are estimated.
But, yes, I used a hyperbole, it is a classic tool of rhetoric. I could have said that they just learn how to ask for a coffee, or the directions for the Eiffel Tower. The meaning is the same.
DuoLingo is so annoying with the gamification. It just won't shut up. I found that if I limit my use, it stops reminding me what league I'm in, probably because it realized I don't care.
I just hope I'm not learning wrong things from the new AI-powered checker. It's bad enough that I can't always find an explanation why I got something wrong. (I've actually used ChatGPT 3.0 to explain some of my mistakes, and then was able to verify its explanation because of the keywords it used, and then googling those.)
I don't know if previous versions of DuoLingo had different approach, but on iOS, Android and Windows, I just said "No" to every option when first using it, and I don't get ANY of those. I declined notifications, widgets, reminders, everything.
Not sure about the AI checker yet; we'll see! I've used ChatGPT myself though as a wonderful language tutor to complement Duolingo, so I'm trying to not be completely skeptical / cynical :-)
Sure, I can see that; doesn't bother me personally in the least though as I just said "No" when it offered reminders/widgets/notifications/whatevers, so I don't think I ever heard of it as said or missing me or anything when I broke my streak. My wife even finds it cute, so it's an eye of the beholder thing :)
If that’s how you want to spend your time, then that’s fine. But, it’s kind of delusional to think you’ve been learning to speak another language. At best you’ve been playing a language-learning inspired game.
I know people who have done Duo Lingo for years. I was shocked when I saw how bad their language skills were.
I mean, lead a horse to water. I had 3 years of enforced language learning in high school. I barely learned any conversational Spanish. These sorts of apps are supplemental, otherwise paying some $10 subscription to learn a new language would be a steal.
It's best treated as reading some motivational quote in the morning. Gets you out of bed and into the mood but it won't carry you through life alone. Is that worth $10/month? Depends on the person, but it would be for me if I was being serious.
When I started learning French (probably 8 years ago), I started with Duolingo. Pairing Duolingo with some other tools/methods, I made good progress, and before long, I outgrew it.
Last year, I decided I'd start working on Japanese, so I did the same thing I did before: hit Duolingo hard, listen to audiobooks to build up my phoneme recognition, etc. I was shocked at how slow progress was - not because I wasn't putting in the effort but because it had me redoing over and over and over again the same damn words and phrases. Gohan, mizu, sushi, gohan, gohan, mizu, gohan, sushi.
I was a proponent of the site/app for quite a while - even after I outgrew it - because I legitimately believed it helped me. Now that I've seen what it's become, I can't recommend it to anyone unless you strictly do it for entertainment value with no expectation of useful education.
It really needs something like Anki has where you can say "I know the word for rice confidently, don't ask me for a long time" with the space between repetitions extending based on your feedback.
Some months back, one of the updates "reset/readjusted" my progress.
And it feels like that was the nexus point between it feeling like progress, and feeling like learning by rote. Even after getting past where i was pre-reset, it still was slow, rote memorization.
I got to 150 days, and the news about layoffs and AI was enough to have me reevaluate it and realise it has just become a pattern matching dopamine box.
That's another thing: the 'enshittification' is in full-force with really terrible ads plus the not-bad-but-annoying Super Duolingo ads. I could probably deal with that if the platform was good, but as I mentioned, it's just not providing any education value (to me, at least; YMMV).
I've tried classroom instruction, and teacher-driven courses, multiple times in my life. They never really stuck for me.
Duolingo has worked for me where other techniques haven't. I'm now at a stage where I can read a newspaper article in Spanish and understand ~90% of the content.
(5 year streak, but that's 5 years of 5-15 minutes a day which turns out to be exactly the commitment that works for me.)
For comparison, I taught myself Spanish with paper flashcards I wrote, and a grammar book and I was at a similar stage after 5 months. 3 hours effort per week approx so a bit more than you.
If I had the discipline to do it I'm sure I could get further with other techniques, but for me the biggest feature of Duolingo is the way you don't need very much discipline at all to make progress with it.
"You don't need very much discipline" and "5 year streak". Sorry, but is my definition of discipline that off? You managed to have 5 years of discipline. That's the reason you managed to learn a language, not because the duolingo gamishittification does miracles.
The gamification is what helped me make it this far. Remembering to pop open an app for a few minutes on your phone takes so much less discipline than any other form of language learning I've tried in the past.
I'm sure it's not the best language learning tool around, but it has taught me enough of three different languages to visit a foreign country, ask for directions, order things, etc, with a number of conversations never falling back into English. And all for the price of free.
It all depends on one's expectations. I have no expectation of speaking a new language correctly or with a decent accent (I'm probably too old). I've being using Duolingo for some time and I complement it with Youtube videos and random texts. My French spelling has improved and I can read Russian news. I'd call it a success. Of course it won't replace a proper course and teacher, but it's not worthless.
Learning a language at an older age is anyway difficult and will only succeed with a lot of effort. As someone that's learned a foreign language at almost 30 in a classroom, I can say that Duolingo does some things well, mostly by forcing or motivating you to repeat to memorize things. I say I learned it in a classroom, but objectively I never would've made it if I was not in the country whose official language I was learning. So the classroom gives you some basics and then you need to practice a lot until it becomes something you don't need to think about. As they say, internalize it. This takes years, but also depends on the language, some are more difficult or have different learning curves. Duolingo is again good at the practice part.
Most people try to use logic and comparisons to their native language, but this is imho not a good way. Kids don't learn like that. It's argued that grownups lose this ability, but personally I sometimes say thing where I have no idea why I said that and if it's really correct. But then it turns out it is. It's somewhere in my head and it got there by listening to other people speak or reading. I know a lot of people that got nowhere by going to class because they had their group of friends from e.g. Spain, China or Korea around them. I met a person from the US who was disappointed with the little progress they made and said "I don't understand that and why it's like that". My only answer was really "because it just is". So I think Duolingo has a place, as a sort of a virtual native speaker you can interact with.
I’m bilingual from birth and have never found languages difficult, but finding the time to put into them does get more challenging over the years.
Maybe that’s why I went with Duolingo initially. I’m not finding the language learning particularly hard even though I have twenty years on you, but I also have the benefit of having to be in a classroom four hours a day, five days a week. That is making the difference, apart from a very good teacher.
I know that in theory languages get tougher as you get older, but the teaching of languages has come leaps and bounds since I did French and Spanish at school, and my motivation has shifted too, so I’m finding things easier now than ever before.
Duolingo has the same incentives problem as dating apps - once someone is sufficiently fluent in the language they want to learn, they stop using the app. So instead of efficiently teaching the language, Duolingo makes the process as slow as possible while tricking the user into feeling that they're quickly learning.
Eh, I don't think that's accurate since there's no "monogamy" equivalent to language. It takes a long time to learn a language deeply, and if that were made easier, many people would just want to learn more of them. The nature of (the vast majority of) relationships is that when you find one you're done looking for more, that is not the nature of any type of knowledge.
Learning even one new language when you are already an adult is hard (or even almost impossible if complete fluency is the goal) for an average person. It's not just acquiring some knowledge, it's hard, and I'd say in general people are much more likely to f up their relationships just for the thrills than to learn new languages just for the thrills.
It does take a long time, and also "real life" immersion, to learn a language deeply. It's the immersion part that Duolingo doesn't do well if at all. So it has to artificially slow down those things it does do in order to keep subscribers.
> I spent 18 months on it and the learning was soundly thrashed by three weeks in a classroom, and for all the pupils, not just me, primed for success by 18 months with Duolingo.
I wonder how well you would have done if it hadn't been for that 18 months. Because 'three weeks in a classroom' sounds suspiciously to how I learned Polish: I seemed to be making zero progress for years and then in a very short amount of time (a couple of weeks at most) it suddenly 'clicked' and I started to make sense of what I heard. I probably was learning at some level that wasn't accessible all the while thinking I wasn't making any progress at all.
Duolingo is 'infotainment', it gets you from zero to a little bit above zero and it may give you some useful number of words to recognize and maybe even to use actively. But it won't help you to speak a language, for that you need to progress beyond it. This is pretty much the same story with every other language, nobody learns to speak and use a language actively by reading the dictionary. But it certainly doesn't hurt to have a basic 200 or 300 word vocabulary on which to build.
Personally I use Anki for such learning rather than commercial services, it allows for far more customization and you won't be sucked into silly gamification for the sake of extending your subscription because there is none.
That everyone in the class demonstrated the same progress showed that Duolingo provided little advantage, which is what I was attempting to say in the quote.
I gave up Duolingo around 18 months before starting the course, which is for complete beginners with zero knowledge.
>everyone in the class demonstrated the same progress s
Did they really? Did they all ace the tests and do all the assigned homework? I find that even harder to believe than Duolingo being the primary resource to teach you a language.
It's not like it's impossible, depending on the age of the class, money spent, if the school had a selection process, etc. But most adult classes aren't really that selective.
We're only 5 weeks in and people have pretty much gotten to the same level of basic progress, which is far further than Duolingo in 18 months, regardless of homework, no tests - and eventually everyone will have to take a B1 exam anyway. B1 is required by the state, so I expected most people would have a slightly higher level of motivation, but people are of a much wider range of skill than I've previously experienced in a class and some of far less motivation.
The thing is that their whole marketing and selling point is mostly to people who don't have real strong motivation or focus to learn a language or don't have the time. E.g. to the tone of "you'll need to do this for just two minutes a day", "it'll feel more like a game", " much easier than in a traditional classroom setting ", etc.
But the truth is that it doesn't really work that way; someone with focus and motivation learning grammatical rules and more in a classroom will always be either more effective or progressing at least at the same pace (or at least be just as ineffective).
Their business is more about selling the idea of learning a language and that it is actually easy, partly by presenting it in a gamified way. It kind of works because people fall for it and buy a subscription for ~3 months but probably eventually stop. They then need new subscribers and have to double down on that marketing/selling of the idea.
In reality, there is no shortcut to learning something. It's just about having the curiosity, motivation and discipline to do so. And many people may think they have that or are enticed by the idea of learning a new language but their life doesn't depend on it enough to have the aforementioned qualities to do it.
I don't think that deeply about it and suspect you're misinterpreting the goals of their target audience.
I don't use Duolingo to learn a language as effectively as possible. I use it because it's a super easy and an entertaining way to stay motivated while I learn a language. I would never go out of my way to take a class or watch videos to learn the language.
I don't think people are "falling for it". It does exactly what they want it to do for them.
The language I was learning had no tuition available for adults in my country and I practiced with native speakers when I could find them and consumed media from the community of the target language.
All while on public transport, when I would otherwise be on my phone. All for the price of nothing (though their model has grow much more user hostile).
This is not to advocate for Duolingo as a company or even the app as a method but I found it effective and useful as a tool.
Duolingo isn’t going to learn a language for you. You have to internalise the grammar and vocabulary.
Could I have done the same with a book? No. I could have done something different. Could I have done the same with a class? No, no classes existed for me.
Nothing is a substitute for immersion, and learning a language is so incredibly hard that it's fair to question the motivation of the learner and ask whether they intend to live in the place that speaks the language.
If you intend to move, then whether you practice Chinese for 4 years in a classroom vs. with Duolingo, any differences are going to melt away very quickly and what will emerge as the forefront factor of success will be one's innate ability to push yourself in a foreign environment; i.e., don't be safe and only interact with people who speak your native language.
Learning a language is not that hard. Everybody does it, so it is not be that hard.
I learned French and Spanish natively because of my family, but I speak other languages like German or Mandarin, although I only know the most common Chinese symbols I can understand people talking and communicate myself effectively.
The most important thing is not forgetting what you learn, specially when the neural connections are weak. For this I used Anki and my own tools.
You do not need to push yourself so much. In fact I do not recommend pushing yourself into a foreign environment without basic understanding, specially If you are an adult and need to function in that enviroment.
I traveled to Germany only after studying it for years 30 min a day. It was much better experience than in China in witch it was similar to your "pushing yourself" method where very few people speak English.
In China the rate at which I could learn was not better than if I had started learning that in Europe. It overwhelmed my capabilities.
Also, China is a different culture, and you have to learn it the hard way with painful misunderstandings.
We quickly learn how vulnerable we are without a local language or Lingua Franca.
In my experience working class immigrants who have immigrated to places that have no large established language community in their new home are the best demonstration of this. They often start with nothing but end up with great skills whereas others will just stay and interact with their native language community and be barely conversational in the majority language.
The first time I used DuoLingo, I was just trying to get through the tree, without really learning, and that was useless. Now I have a different technique: I’ll only listen (never read), and if I can’t comprehend and respond immediately, I’ll fail that exercise deliberately, so I’ll be re-tested later. For me, I only understand the language if I can understand some one speaking to me and respond in real time.
I started doing the listen-only as well! By this, I mean I just don't look at the screen, I'd love to know if there is a way to setup a listen-only option.
I've also used Jumpspeak, which is an AI conversation partner. It works ok, but the speech recognition is... not great. But you can have a conversation and practice listening and responding. I was able to treat the AI as an uber driver and ask about places to go in Peru, and how to get there, and why they were nice.
This might be a cynical take, but I think that for most users of Duolingo, it functions more as a game than as a tool which produces substantial gains in language knowledge. I'm a computational linguist, and whenever on hearing this people ask me what I think about Duolingo, I like to say it's language-flavored entertainment.
Is it necessary to specify TCP/IP? Is it not also applicable to books? Reading piles of pulp fiction is subtly different than language-flavored entertainment.
The gold standard is immersion, everyone agrees on that. Most people can't have that, so a mix of language instruction, conversation with native speakers, and exposure to appropriately challenging texts with memorization tools like Duolingo or Anki as a _supplement_ is the next best thing.
I know of some people who have learned languages through Duolingo. I also know some people who took classroom courses and accompanied those with Duolingo – and they believe Duolingo was helpful in that sense. I know people who learned foreign language by a combination of Duolingo + staying in a foreign country at the same time.
I myself have used Duolingo to learn some rudimentary basics of a foreign language and I haven't forgotten those.
I doubt Duolingo is the most efficient, but that doesn't mean it's useless.
You can't do it with just Duolingo. But I truly believe that there is no method except for full immersion which can make you learn a language alone. It's always: in person classes, reading, watching medias, changing your phone's language, going out to restaurants, language learning apps, making new friends.
You learn the vocabulary, you learn pronunciations and you learn how to listen and comprehend. I don’t see what’s so bad about it. It’s never going to be able to compete with a native speaker just speaking to you in that language but nothing will.
The problem lies in the perceived benefit that the user believes. The app, through its marketing and content, leads you to believe that you can and in fact, will, learn the language of your choosing.
Anecdotally, this has not been the case for me.
Similar to the parent, I have accelerated my learning tenfold by ditching D**** in favor of reading, watching, and listening to content in my target language as well as speaking to a tutor I found online.
I've wasted a tremendous amount of time going through and ultimately completing a tree only to find myself crossing the finish line not being comfortable saying anything in that language. Is that a criticism of me? Perhaps. But, the big D once fooled me into thinking the answer to that question was that it is. Which is to say, I don't believe it's a criticism of me.
My anecdotal experience has been that it taught me enough words in Spanish to start talking to people at bars and then once I started doing that my conversational ability exploded.
I’ve used duolingo several times on and off throughout my life for different languages (Spanish, German, French). Sometimes paid, sometimes free.
In the beginning I thought it was excellent and that’s why I kept coming back.
I tried it again the last 12 months and the app wasted so much of my time making me skip through promotions and do-nothing animations. Opening chests, collecting gems, all the gamified stuff.
I didn’t need it. I was already self motivated to learn without a bunch of time sinking doodads that you have to click through.
I guess duolingo is good if you are struggling with motivation.
Recently I used a paid app called Pimsleur that I liked. There’s also a lot of good free resources on YT
Your point is a good one, although I have to point out that Duolingo does at least let you skip through the gamified stuff very quickly. When I'm doing lessons I reflexively double tap the "Ok" button to skip over the little whoo-hoo animation.
It would be awful if I was forced to wait for the animation to complete, which fortunately, they do not. The same applies to basically all the animations except for the opening screen.
> Duolingo should be on borrowed time, as a method to learn a language it’s inefficient to the point of being ineffective
A lot of Duolingo's questions ask a user to pick and choose words to complete a sentence. That gives a false sense of achievement when in reality it simply does not challenge one's mind enough to learn a language effectively. Besides, a simple sentence does not give me enough context to build my intuition. I personally find the idea of Comprehensible Input works much better: reading extensively, watching videos extensively, using a language increasingly frequently, and etc.
The way to think about Duolingo is like another mobile phone game. Except the "universe" of the game is language learning instead of a random farming or middle ages theme. And so you learn about the language in the same way you learn about farming or middle ages from those games. Meaning if you never read anything about farming or the middle ages now you know of some things, but you'd really be much faster ahead by picking up a proper book on the subject. So it is with language learning apps.
The vocab is definitely useful in the real world, and audio comprehension is improved. Foreign language textbooks also have vocab sections, and Duolingo makes that part easier.
I have found it helpful, even as the central foundation for learning languages, but you have to remember at the end of the day it is *you* who is learning the language and you absolutely must mentally engage yourself, it can definitely lure you into a path of scoring points without making real progress.
I will pause my lesson and jump over to Google or ChatGPT, and usually end a session with a pen and paper writing out any phrases I was less familiar with. You also need to make sure you’re being challenged, with some languages I get into a phase where it is really easy for a long time, when that happens, you need to skip ahead because just scoring points is meaningless. I still think the gamification is useful and unique and can help keep your overall learning efforts progressing. Also I always supplement with Pimsleur, an audio-only program that is great for listening while you’re at the gym or biking or otherwise unable to read or interact with a screen.
My finding as well - how I engage with DuoLingo had massive impact on my progress.
Focusing on streak, leagues, and XP/points was detrimental. I'd do the repetitive high-points exercises instead of lessons and feel "I've accomplished my goal" (of having a high score).
Once focusing on my goal of "I want to learn a language" rather than points, it was WAY better. I also create my own cheat-sheets for conjugation / adjectives / etc, and find ChatGPT a phenomenally patient tutor for my incessant "WHY, French? WHY is it thus?!?!?" questions :->
My experience was different. I was able to learn enough Spanish to survive in Peru for a month with just 6-7 hours in Duolingo. Different people may have different learning styles. When you're comparing to classroom, are you normalizing the number of hours spent? Of course live interaction will be several times more effective, but it's not something everyone can afford to do. It's also about goals. Duolingo isn't for you if you really want to become a proficient native speaker or be able to understand books. But it's great to learn a language casually enough to converse with locals.
I'm about 18 months in as well and feel like I don't know crap. I just can't memorize thousands of new words. I honestly have no idea how anyone is bilingual.
No one source of language learning material will bring you to fluency. There's a great community of language learners over at the Language Learners Forum [1] (note: it's been down the past few days but seems to be back up, although slow).
One resource that seems to be universally loved over there is Assimil [2]. I've been working through one of their courses and it's been great. I augment it with youtubers that make comprehensible input videos at my ability level, as well as with iTalki lessons.
I tried Duolingo once many years ago, but there's no way that memorizing words is going to get you all the way there. You need reading, writing, listening, and speaking practice in real world situations.
> I tried Duolingo once many years ago, but there's no way that memorizing words is going to get you all the way there. You need reading, writing, listening, and speaking practice in real world situations
I’m not disagreeing that multiple sources is necessary, but describing Duolingo as just a “memorising words” app is wrong.
It has reading, writing, listening, speaking, and conversation exercises.
That’s totally fair, I haven’t tried to use it in more than 5 years and it’s probably a lot more than I remember. It didn’t work for me though and I just wish I had found Assimil a long time ago.
If people find it a useful tool and it has helped them gain fluency though that’s great!
Go live, date, and work in a country that speaks the language you are studying. You will find that you don't need to start from ground zero and you will gain speaking and listening skills quickly.
And that's pretty much the thing. Pretty much everyone I know says immersion is the thing. You can probably learn to read academically. And I'll probably do some more language courses before a more immersive situation. But for the vast majority of people "I've spent a month with this online course" fluency isn't realistic.
Just subscribe to ChatGPT and tell it to teach you whatever language you want to master including meaning of each sentence in your own language. There are plugins for voice input/output as well so you can learn while doing a workout etc.
It's not, it's just a different style, "full immersion" where you have no clue what is going on and they think at some point you'll understand. I don't think it works well but whatever. Best is to pay for a trip to some remote area speaking only the target language and force yourself to understand it quickly, swim or sink.
No idea, if you can survive with VR goggles for hours, and you have reasonable VR partners not trying to teach you all kinds of slang and swearwords, then maybe.
> I spent 18 months on it and the learning was soundly thrashed by three weeks in a classroom, and for all the pupils, not just me, primed for success by 18 months with Duolingo.
Yeah, but if you have time for 3 weeks in class, why do Duolingo in the first place? Those serve different markets. Duolingo is for low effort, low commitment, but interested enough to do it a bit in their spare time kind of people.
I found it helpful for reacquainting myself with vocabulary from a language I learned decades ago. I think it would also be helpful in learning the basics of a new language adjacent to that language (romance, in my case) before going on a trip. Beyond that, I’m not sure how useful it is on its own.
It took me a few times to understand their sentence correctly.
What I think OP meant to say:
All pupils found success. My success could be explained by being primed for success with 18 months of duolingo, but all pupils had similar success.
As written, it reads like "all pupils [...] primed for success by duolingo" which is perhaps why you tried to look for irony to match it with their conclusion.
Studying Duolingo for 18 months should have primed them for success, but after just three weeks of an in-person class, it was evident that it had not. It's implied (or assumed) that the class was for 2nd year students or less, so Duolingo must be less effective than traditional instruction.
It seems like AI, deployed against a few hundred learners, ought to be able to discern what the most efficient path to learning a new language is. Then you can just create a new app that does exactly that
At least in the grassroots language learner community, there seems to be a growing sense that the most efficient way to achieve real fluency in a second language is: regular long-term effort to read or listen to text or media in the language that is just a little above your level of understanding. Notably different from how duolingo works.
I suggest that the power of these new language models is that instead of finding “the path of success“ the models can adapt, and learn to accommodate individuals with every respective path to success. Chat, GPT never has the same conversation with two different individuals, roughly speaking. If Duolingo can pivot to deliver automated, personalized learning experiences, then more power to them. That would be fantastic in my opinion.
That's the nice thing about ai though, it could be tailored to your experience and learn so you do. For example if you are progressing on pace with similar learners it could apply a different learning program or vary course materials until it sees improvement
Immersion will absolutely get you speaking and listening fluency (with enough time). Reading and writing? That's another matter. If you're coming from a Latin alphabet language and trying to learn to read and write Chinese or Japanese then you've got a lot of study and practice ahead of you. Acquiring the knowledge of several thousand Chinese characters (needed for both languages) can only be done through deliberate study.
I canceled Duolingo subscription in favor of just using ChatGPT. Duolingo wanted $30 (a $20 increase) for Max, which is as far as I can tell less flexible and capable than ChatGPT anyway.
As someone else has said in response, my vocabulary, pronunciation, and understanding of grammar rules are solid. I believe I've already learned 3000 words of Spanish at this point. I will need immersion to gain the speaking and listening skills, but I was never deceived about that. I figured this would be a small daily investment for a few years that I would eventually need to actualize.
Maybe because your brain made new connections, but was "primed" by what it knew from Duolingo.
A bit like you plan a holiday, read the guidebook, then you go to the destination. Your knowledge of the place was thrashed vs. reading the guide book. But the guidebook helped with some foundational knowledge and was still useful.
Thanks for posting your experience. I haven’t tried duolingo. Do you think it’s possible that the 18 months helped prep you for classroom learning , vocabulary etc?
How intense was your classroom learning and how much time per week?
I took years of French in school and would like to learn Spanish.
Duolingo is basically the app equivalent of flashcards. It's a little smarter than that, but not much. If you're serious about learning a language, it absolutely has a place. But Duolingo __alone__ is not going to make you competent, much less fluent.
Personally, I'm happy with it. I took years of Spanish and started with Duolingo in French with zero prior French, and I can now recognize phrases and hold some basic conversation; for 5 minutes a day with no significant continuous time commitment that's not bad!
I've only recently started the classroom learning, week 5 just began, and I'm in there five days a week, four hours a day, so that's really great and the outcome required is, I think, B1 level. Duolingo over 18 months contained so little vocabulary and grammar, plus I stopped it maybe 18 months ago too, so its impact was restricted to about one day's worth of class.
I did five years of French at school and three years of Spanish and definitely enjoyed the Spanish more.
Duolingo works for me because it keeps me engaged in a way other language learning resources wouldn’t. I might be making super slow progress but I’m making some progress versus none at all
No Japanese. I get it, but it does seem like choosing any Logo graphic language limits what services there are. At least if your base language is phonetic.
I prefer Mango Languages, which I get for free through my public library. The progression feels better to me, it's not as gamified (sorry for using that word), and there is some (though not a ton of) actual lesson content in there. For example, learning Spanish, it gives some context on the different usages of "Ser" vs. "Estar" for "to be", and it includes discussion on gendered nouns. DuoLingo may have gotten to some brief lesson content at some point, but for me it was too grating to continue with and I quit before I made much progress with it.
I used Duolingo in 2022 for the 35 days leading up to a holiday in Greece. I went from knowing 0 greek to being able to read the alphabet and words, pronounce them and do things like order in a restaurant, explain allergies and convey our room number.
Around the beginning of 2023 they did a couple of things that ruined it a bit for me. They completely changed the track system so now you have to follow one set path, and they got rid of the classes feature which was awesome because for around $5 I could have a zoom lesson with a native teacher and other people learning the language.
Also the sentences it teaches you are a bit less silly now and that was a cool gimmick.
Wasn't one of their initial monetization ideas that crowdsourced translations - from amateurs, not even paid contractors - would outperform AI translation? The translation thing never came close to panning out, from what I can tell.
Back in the day you used to be able to see a mini discussion thread about any exercise. The comments in these threads - from unpaid users - were frequently much more helpful than the official teaching notes. But they got rid of those a while back, too. At the time people suggested it was because they wanted to push people into some sort of paid tutoring offering that they'd announced. But I don't see any hint of that around. Now I'd guess it's because the questions are no longer hand crafted by people, and are now autogenerated by AI, leading to an inability to have discussions about specific questions.
This seems to be a scaling issue for all of these online language/SRS learning systems. Every single one of them scales and then starts dialing back/deleting/hiding all of the user-generated content that helped the platform get to where it is today.
Almost a decade ago "smart.fm" was a thing and the best thing about it was all of the user-generated content (blogs) offering discussions and explanations for so many various languages. From grammatical concepts being explained to people asking questions and having users provide answers in public Q&A style format. Then smart.fm got rid of all of that and became "iknow.jp" and deleted everyone's blogs and hard work built to create the smart.fm community (many of whom then moved to Memrise).
Memrise was originally about using user-submitted mnemonics and it had a vast library of them being created. Nowadays, as far as I can tell, none of the mnemonics are around anymore and they're no longer the focus. The focus is on monetization and gamification of their Pro user stats (of which I have lifetime membership until the year 9999 due to my help/work during their Beta testing years).
Because mnemonics were the focus - words across all of their courses had to be combined so that the mnemonics between courses would carry over. Myself and a number of (unpaid) volunteers spent months combining all of the words for every popularly used language (I helped with the Japanese dictionary) across all existing courses at that time.
Hosting (and moderating) user-generated content is an issue at scale. At the start when you have mostly good faith actors and few trolls it works quite well. But after a certain scale moderation becomes a massive issue.
At least I got a cool staff-only T-shirts out of it. Ben is an awesome dude and when I donated to the Memrise bus tour godfundme I asked if I could have one of their staff shirts that I knew they had - and they actually sent me one instead of one of the bus tour shirts!
The better funded of these sites seem to last a bit longer/scale a bit larger but it seems the death of user-generated content is inevitable after a certain point.
Wait, CAPTCHAs don't generate or capture any data. It prevents computer automated systems, which were a huge pain at the time. Google then came around and released reCAPTCHA which was using originally books for Google Translate but later Street view photos for Google Maps. That was when dara collection and learning was introduced.
Yeah, you used to be able to dive into really interesting/useful discussions (pretty much a forum) about nuances of certain phrases/words, usually involving people who have grown up speaking the respective language. It was really helpful to get that extra detail and context on weird little quirks that may not have been obvious by the content built into Duolingo itself.
It was fairly recent, maybe in the last 3-4 months. I know for much of 2023 I would get the mini threads for some (not all) questions, but now I don't get them for any question. I agree it's a shame, because they really were the most educational thing on the app.
Such a large part of making lifelong connections in this world are from hiring people who you need to make something work. I made some wonderful connections in my previous jobs that I never would have otherwise if they could do everything themselves with AI or if I could do everything myself without them.
The fact is, automation can make life better but beyond a certain point, if hardly anyone needs anyone else, then all those wonderful connections of working together vanish. We can stil make those connections theoretically, but the lack of needing someone else removes the actuality and genuineness of the process.
I wish that everyone who created AI and who welcomes it would think about this longer-time societal impact before becoming blinded with its immediate appeal.
Sorry to say but I think making AI in general is one of the biggest mistakes we ever made.
A company that was before 400 people, can now be 200 people. But people will rarely make meaningful bonds outside of those 5 people closest to them (Dunbar's number).
It's also not unique to AI. One person with a tractor can outharvest the amount of corn it used to take a whole village to harvest. This feels just like a dressed up luddite complaint.
Well, in my example of a company being reduced from 400 to 200, you can still fill out a couple of more layers ;) – I think you are good in that company with meaningful bonds.
On a team that was 12, but is now 6, the odds you find people you relate to is effectively halved. It’s rare entire workforces interact often enough to bond in the first place.
This is such an engineer's perspective. The point is, it is not just the number of connections, but the type of connection: diversity will be removed, since you only really need certain skillsets. The people with design skills and other types of skills will be removed.
This numerical accounting perspective so widespread with engineers and computer scientists is the sort of reasoning that is bringing us to disaster.
That's good to think about. If AI tools are genuinely helpful at work than it should change the people we meet there. Instead of bonding on lower level details we'll bond on what we choose to create or our larger contributions to larger creations. So I think that we'll still have these relationships.
There might also be companies where a few people work together doing different things, that used to each be a department. They won't have their jobs in common, but they'll bond over working in a strange new type of business and the mental agility required to be responsible for so much output.
There is an analogue to this in manufacturing as well. We lost something when we moved all of our manufacturing to whatever the race to bottom place was of the moment instead of understanding how to make our things from top to bottom.
Yeah. And that brought a whole slew of other problems (such as supply chain security and systemic fragility induced by centralization) that we just accept as worth it.
Curious to know if you believe it is truly lost because I would argue that what could be considered lost (artisan, soulful goods) still exists. Perhaps I also misunderstand "what" you believe is to be lost.
From my experience in Central Europe in no particular order:
Woodworking, shoemaking, metallurgy, even services like plumbing or an electrician are getting harder and harder to find. Building and repairing electrical machines. Refinery of oil-based products. Manufacturing of computers and circuitboard beyond the most basic.
There is no manufacturing sector, anymore. It’s all gone.
Tangential anecdotal point: I used to have an antique "home improvement" book from 1910 or so. The exquisite wood joinery that was shown there (as the achievable standard for the motivated man-of-the-house) does not exist anymore, it’s not being taught.
If meaningful connections at work go away, then clubs (books, chess, sports etc) and social hobbies will pick up the slack. Work doesn’t have a monopoly on meaningful relationships.
The machine generated content was already bad. I don't know if this AI transition will make it better or worse but I swear Duolingo was among the worst of apps I tried. People say it's ok to use supplementary when you have a 10 or 15 minute break but I think sentence mining or watching YouTube videos with subtitles is a better use of those minutes. I avoid telling giving this advice to people in real life because it seems dismissive of the time they are spending.
I realize lots of people have learned languages by watching movies in the target language and reading subtitles in their native language, but I have found that to be highly difficult. I guess everyone is different, but given that the vast majority of people who read something while listening to something else will either comprehend something from one and nothing from the other, or worst case, get nothing from either, it seems likely that doing so to learn a language is not optimum.
I could perhaps see this method as more effective in cases where the 2 languages are similar enough that there is some "lining up" between what you read and hear, but where the words are completely dissimilar and even sentence structure is completely different, I just don't see it as any better than a brute force method of learning.
In my opinion, watching movies is ineffective until you can read the subtitles in the target language. This is at the B1/B2 level. Before that you're just focusing on the native language.
Once you hit the B levels, things get easier. But it's uphill until then and tutors or courses are your best bet to get there.
You've got it the wrong way around there. The spoken language should be the language you're trying to learn, the subtitles should be in your native language. Of course that'll not help you learn to read a foreign language.
I agree with the above poster that it’s a better use of 10-15 minutes. But not native shows/tv. Most languages have all sorts of content targeted at learners, sometimes even broken out by level. The Dreaming Spanish YouTube channel is an example.
But the challenge is you end up spending those 10-15 minutes researching content to use for practice! (I’m exaggerating of course, as once you find a few that’s usually hours and hours of material).
The technique here is called “comprehensible input”.
I had a friend who was a translator for Nintendo and could translate across 12 different languages. He said after he had the first few learned, he would learn new languages by re-watching his favorite shows with both audio and subtitles in the language of interest.
This is sadly unsurprising. I have a foreign language degree and a bit over a decade ago, I spent some time trying to get a translation side-job off the ground.
Even then it was clear that , except for a select few languages, it was a race to the bottom with machine translation software as the driving force. I gave up on that career idea quickly.
Now I use ChatGPT and is does a great job. My experience with DuoLingo is it’s barely better than flash cards for language learning, and it’s no wonder they’re having a hard time making any money offering a product thats competition is cheaper and better.
ChatGPT is so good at translation, it would be ridiculous for Duolingo to not leverage it. I've been using ChatGPT to learn Italian, and I've noticed while playing around with it that it can even perfectly translate endangered languages like Occitan, which is far too obscure for Duolingo. I wouldn't be surprised if Duolingo is leaning in on AI content because they see AI as a threat, not a tool, and they don't want customers to realize that they could do better with ChatGPT.
> it can even perfectly translate endangered languages like Occitan
I’m continually surprised by the ever increasing hype surrounding AI. How could this statement possibly be true? Does ChatGPT have any Occitan texts in its training corpus? Are you fluent in this language to be someone qualified to make such a determination?
Every comment I’ve read from someone employed as a translator says it still doesn’t get the human nuances of languages. Yet, it perfectly knows this language?
Wikipedia even says “Occitan does not have a single written standard form”[1]. If that context is true, how could any thing possibly translate it perfectly?
I have no clue with Occitan specifically but "It does not get the human nuances of languages" would be an absurd conclusion for GPT-4's Translation competence.
Sorry, I think my comment probably came off more harsh? than i intended.
It's just 4 is so ahead of the pack. Korean Passage Example i cooked up recently from the excerpt from a novel and the official translations amongst other translators for comparison - https://imgur.com/a/XtH7dZt
> Every comment I’ve read from someone employed as a translator says it still doesn’t get the human nuances of languages. Yet, it perfectly knows this language?
Wouldn't it be fun to make a bind test on these people? I would be surprised if they can statistically distinguish machine translations from human transactions.
I used ChatGPT for Turkish and it got it really wrong. Wrong enough that even my limited Turkish knowledge could see it was wrong. That said Duolingo was pretty bad for Turkish as well (Babbel was far superior).
I guess maybe ChatGPT is trained mostly on Western/Indo-European languages?
I am learning Polish, and find that ChatGPT provides a much better tutor than any human. Ie. I would rather use ChatGPT than a human tutor, indifferent to the price.
I am also developing my own app to learn languages, and I can get to Duolingo levels in roughly a weekend worth of side project.
I project that a certain category of apps will die off really fast.
You never actually know that, and most people are. That's why there's so much "nous" use in French classes and so little "on" and so on. But the learn rate is so fast, that you get sufficient coverage to fix it up in actual conversation. My wife and I did a combo of another AI product (https://www.speak.com/) and ChatGPT to learn sufficient Spanish to get by in Chilean Patagonia and it did a pretty decent job (we just learned pronunciation through mimicry).
We're not experts as a result, obviously, but it helped us get around and stuff which is fun. It was pretty much a crash course started as we were landing in Santiago (which is where we entered the country).
> But the learn rate is so fast, that you get sufficient coverage to fix it up in actual conversation.
This should really be emphasised. Correctness might not be entirely on every time. But learning pace and engagement is so much higher, that it more than makes up for it.
Especially engagement is a point where ChatGPT wins over all human teachers.
How do you know a human tutor is not confidently incorrect?
I know it sounds like a glib answer, but in reality the answer to both questions is you don't know for certain in either case. Honestly I would say both are incorrect for at least some portion of their responses, but you seem to be insinuating that humans are closer to 0% while the LLM's percent is higher. In my estimation the LLM is probably less often incorrect but that's just my opinion. I am sure a more rigorous study could be conducted though.
It's true that you can't know with a human tutor either. But ChatGPT is known to be confidently incorrect. If a human tutor had earned a similar reputation, one would be wise to distrust him as well.
This is also my own experience. In particular, ChatGPT has deeper knowledge on complex grammatical structures (like the 7 cases polish have) than I can expect from a regular tutor – my friend was not able to explain complex consonant clusters, while ChatGPT happily babbles out resources and explains just what I need for a particular case.
I know that as well as with any human tutor. The main difference is that I am motivated to check the answer, eg. by asking the woman who is my main reason to learn Polish.
Also, the value of a tutor that in and encouraging way corrects my verb conjugation of być (to be) for the 500th time has a lot more value than the risk of it being wrong on a corner case.
A human tutor would semi-arrogantly have told be off and asked me to look at the dictionary (Yes, I have studied English, German, Spanish, and French in school, I have had more language teacher than anything else)
Curious, when you went back to that woman and checked an answer with her, did she ever say "No that's definitely the wrong way to say it" or anything like that?
I'm not arguing in any way I'm legitimately curious, as someone who has dabbled in language learning, as well as trying to use ChatGPT as a tutor, how its track record has been. Hallucinations was definitely one of my concerns when asking it about language stuff.
She corrected usage. Eg. when to use mama and matka where it is super hard to intellectually learn how formal one should be.
But again, in english, when is it appropriate to use mom and mother? It really depends on the context, culture, and relations.
Also hallucinations are probably a good thing for language learning. especially when you have an interface, chat, where you can interrogate – and an AI that does not get offended when you question it.
Any tutor makes mistakes, it is left for the learner to figure out whether it is BS later.
Truth to be told, human WILL never match ChatGPT's patience and coverage when it comes to explain most insignificant details of one particular language, any time anywhere you wanted. It takes out the fear/embarrassment factor when it comes to a human tutor.
If you are learning Polish, how do you know that ChatGPT is the better tutor of Polish?
I guess you could ask the same question of a human tutor, however this seems to me the sort of thing a chatbot could very confidently give BS or incorrect information, and you would have no way of knowing unless a human who already knows Polish pointed it out.
No human person has the stamina to spend 3 hours fixing basic conjugation tables and still encourage me.
As a European I have received extensive language tutoring and know how people teach. I also know enough Polish people that I can certify that they are just regular people without super-human tutoring skills.
Unless you consider all people you ever talk to for tutors, I think that statement is a, well, stretch.
In language learning you rarely learn alone from you tutor. The same is the case here: I also watch videos, movies, listen to music, practice pronunciation on friends, etc. Justlike you would with a human tutor.
i’ve been using it (with a stt->prompt->response->tts loop) for chinese practice and it is very good. being able to say “role play a market” and drop into it and switch mid sentence between languages when i get stuck and it just works it out and helps is amazing. i still think a human tutor is required for foundations but gpt is very helpful for practice.
I see a lot of complaining about Duolingo being worthless but I beg to differ.
First I have not used the app but IIRC it uses a lot of "Choose the correct word among those four" but it's quite bad for reinforcement.
It's much better to reinforce what you learned to have to type a sentence from scratch. I recommend using it on a computer.
Also, I don't know if that exercise is still ther, but having to pronounce a sentence is also good. The problem was it was too easy to pass though.
The other thing is that it does not replace learning the language, that is reading a grammar in order to know what are the general rules of the language.
But it does replace conversation if you are not able to speak the language and provide a a good source of daily reinforcement, and vocabulary.
It's an effective tool when used correctly in my opinion.
They removed typing in answers for source language (i.e. English) translations. Was a feature, but now it's gone. You have to pick from the words they give you
When it was a feature it was often frustratingly broken because their questions would have an ambiguity but they would only accept one possible answer as valid. So you would read the sentence, translate it, and then get told you were wrong because whoever wrote the question didn't recognize the ambiguity.
Removing the feature altogether was probably cheaper than fixing all of their questions.
AI translation falls flat on its face with gendered languages, when it is given text in English and has to translate it into French, German, or Polish. How is that even a consideration for a service that's supposed to be teaching people languages?
Can you give an example of this? In my experience (learning Polish) ChatGPT can usually derive the gender when the subject is a common name:
Sure, here are the translations of your sentences into Polish:
"You are beautiful Natalia" translates to "Jesteś piękna Natalia."
"You are beautiful Piotr" translates to "Jesteś piękny Piotr."
This publication is from february 2022 and can probably mostly be invalidated by modern software.
First try on a translation:
"Je suis étudiant(e). J'aime lire des livres et je suis très joli(e) en robe."
(Note: "étudiant" is masculine, and "étudiante" is feminine. Similarly, "joli" is masculine, and "jolie" is feminine, so you can choose based on the context.)
(A human translator would probably have inferred feminine by that fact that "I" looks good in a dress, ChatGPT has less bias)
Second try:
"Je suis étudiante. J'adore les longues promenades en forêt et lire des choses intéressantes. En particulier, j'aime lire Shakespeare. Oh, je me considère aussi comme une femme."
ChatGPT infers from the last part of the sentence that "I" identify as a woman and corrects gendering in the entire sentence.
There is no confusion. I am just challenging that machine translation is incapable of inferring genders.
Danish is also a gendered language. The genders are difficult to learn because there rarely are any rules to the genders. Ie, you can not derive it from the word, but only from the context it is used in or idiomatically.
This is an area neural networks perform well in.
Sure, here are the translations of your sentences into Polish:
"This is a beautiful book" translates to "To jest piękna książka."
"This is a beautiful table" translates to "To jest piękny stół."
Speaking of Danish (and Norwegian), ChatGPT has quite large issues separating the two. Often it will insert random Danish words into the text when I write the prompt in Norwegian. The languages are quite similar when written, but it does this even when I specify that it should be in Norwegian.
I'd be really interested in an example. Preferably a link to the actual conversation with GPT4 on ChatGPT or equivalent on another service, but even just typing out the sentence here with the translation you think is correct vs the translation you expect to receive would be helpful.
But that's not a realistic scenario for a learner who doesn't know what the translation ought to look like, because he/she is still learning the language.
Why would I use duolingo if it has about the same product as opening google translate, moving the target language side off the screen, typing in some inane sentence, guessing what it says, and revealing.
You take a product that had the USP of “quality” and decided you’d rather play in the “AI sludge content” thunderdome where anyone with a vague understanding of web development and openAI api access can compete with you for functionally the same product.
Well, these days, Google Translate is pretty bad as machine translation services go... it's much better to ask an instruction tuned multilingual LLM to translate for you.
> Why would I use duolingo if it has about the same product as opening google translate, moving the target language side off the screen, typing in some inane sentence, guessing what it says, and revealing.
That seems like a way worse UX to me than what duolingo offers
I wouldn’t say it’s exceptionally worse considering it’s free vs begging you for microtransactions.
Gamification is easy enough on your own, get some ice cream!
If you need accountability, ask a friend.
You could alternatively fill up millions of flash cards in Anki with random phrases out of chatGPT and really hone in on the Duolingo simplicity.
Silly examples, but I just feel like “AI phrases with TTS audio displayed along side 1 of 20 PNGs of a cartoon character” isn’t really a defendable product in a market that churns out multiple copies of that idea a month.
They used to be a course that was designed by employees/volunteers interested in language (in theory), now it’s a randomized. Eh.
I see no problem, fundamentally, with leveraging the benefits of AI to generate content for whatever thing. I see the problem as those who contributed the content upon which the AI was trained not getting paid for the use of their content.
Race to the bottom propelled by the current fad here we come.
Duolingo has been completely fucking useless for me compared to spending a few hours in the country in question though. It's pretty much HOW DO I GO LEFT AT THE BAKERY level of teaching.
it is impossible -even for native speakers - to pronounce certain words and in particular numbers I think in a way that makes duolingo accept them. This seems to be true for more than one language (think I have heard / read about it from both a Norwegian testing Norwegian and a Ukrainian testing Ukrainian language.
If Duolingo wanted to invent a new word it is probable IMO that they would be able to introduce and popularize it, in spite of no one else having ever thought of that word.
For example, if Duolingo wanted to invent “schmibbidibbi” as a word, they could give it a definition and start using it in their material. Pretty soon people would start using the word IRL.
I wonder if there will ever be any words invented by AI that Duolingo accidentally teaches the world, which did initially not exist outside of the Duolingo app.
that's an interesting idea, but I dunno if "language learners" comprise a sufficiently influential cultural group to influence language in this way.
If I learned a new word through duolingo, used it while speaking to native speakers of the language, and no one knew what it meant, I probably would be embarrassed and confused and not use the word again.
But maybe it'll happen! And along that line, I wonder if advanced language models will be able to identify "missing words" that we're lacking to discuss concepts that we haven't yet conceptualized.
The trick is to teach French learners that this word means something in French and at the same time to teach English learners that it means something in English.
So you travel to France and practice your newly-learned French. You run across a French person who has been learning English. You use the word thinking that it’s French, and the person you speak to has already heard it as an English word and thinks that you don’t know the word in French but they know what you mean thinking it’s an English word.
schmibbidibbi is clearly a distinct word, different from skibidi. But also near enough that conceivably an AI could invent the word schmibbidibbi if it knew skibidi. Depending on how it tokenizes words.
More and more AI-powered language tools and applications are appearing.
I have noticed this with Spanish: the translations from English are wonderful and skilful about 70% of the time, they could be improved about 25% of the time, and they are completely wrong and misleading the other 5%.
This is a rough breakdown, sentence by sentence.
The problem is that the 5% of incorrect sentences reduce the usefulness of the other 95%, so that the final translation is perceived as imprecise and lacking in coherence and tone.
I coincidentally saw a video recently about how Duolingo's AI model was getting some French things wrong. I use Duolingo, but it's a vocab tool more than anything.
it’s been going down in quality as an actual teaching aid for a while, beyond maybe sentence structure memorisation, so it’s sad to see this accelerate further
Duolingo should be seen as an "instead of scrolling Instagram", not instead of formal education.
Change your browser's default language to your new language https://developer.mozilla.org/en-US/docs/Web/API/Navigator/l... . Change your streaming services' default language. Watch your favorite shows in said language with subtitles! Greet your friends in your new language. You know you're doing it right when everyone around you knows you're learning that language. Chip away.
When forced, even the old dog brain will learn new tricks.
I'm using Duolingo heavily and I've run already at least twice to a situation that:
1) my incorrect answer was accepted
2) incorrect English translation was provided by the app
When you think about it, this is why using AI-generated text is no better than using a dictionary. I want to learn from someone who actually thought about the text and whether it makes sense in context, not just whether it's free from errors.
In order to translate like a human, AI needs to be AGI.
I've seen some weird sentences in the French course. I have the impression they were written by a French speaker who needs to improve their English a bit.
For me to have a foreign spouse requires that person to be able to sit around a dinner table in my home country and not force the entire tables to speak English.
These technologies allow me to get started on languages so fast, that nationality becomes less of a hassle when dating bilingually.
Again, I think one needs to have a dinner table with non-english speaker to really appreciate this :)
Which AI services does Duolingo use for the translations? It seems like they're satisfied with the outcome. Curious about the cost of the AI service as well.
My experience is that GPT-4 is quite good at most languages, including hard ones like Chinese - especially when paired with a “verification” step, where you copy something it wrote and start a new session and ask, “Is this sentence well-written <language>?”
It has a surprisingly deep understanding of nuance and context. You can copy some Chinese text that Google Translate completely fails to translate, paste it into GPT-4 and ask “Please explain this text including any nuanced meaning that a foreigner would miss“ and get a detailed, coherent explanation.
I used Duolingo for ~650 days trying to learn Japanese (top 2% supposedly), but once you get that deep into the language, the robotic voices really start to get annoying. It skips over words, pronounces words wrong, and expects you to know new words before it even teaches them to you, and eventually I gave up.
I might give the other language apps a try, but long story short, Duolingo doesn't need ai, they need a better speech engine.
Not used it for a year or two (got to a 365-day streak then decided to stop).
What annoyed me most was they became one of "those" companies that goes to extreme lengths to make themselves impossible to contact (except for billing issues) - they closed all their forums, and you can endlessly search for a contact form, email address, phone number etc. and you'll find nothing.
How about other online learning sites like Babbel.com?
Has anyone around used it and can comment on how good, effective, or otherwise it is?
On thinking about learning a new European language, I had thought about using Duolingo for the first months to start getting some vocabulary, then jumping maybe onto Babbel for more serious study. But not sure if that works too well as far as online methods go.
How does DuoLingo compare with Rosetta Stone? Is the latter better because it is expensive, or did Duolingo beat Rosetta by offering the same thing at a lower price? If Rosetta Stone is still better because it uses humans, I'd consider switching over, but it is like 3x the price of DL.
I'm not into the game aspects of the app and it's really only been good for vocabulary for me. I've found reword on Android really good. Spaced repetition.
And, in terms of AI, ChatGPT has been good because it lets you converse and respond with a real-sounding person. It's revolutionary.
There isn't really a need for human translators in language learning. Machine translation is good enough at translating into one's one native language and the translation quality doesn't really matter as you already know your native language.
I don't want it to translate to my native language.
We happened to have a teacher who was an old Esperantist with some quirky ideas, so I learned English with the so-called direct method. That means that after the first class, the teacher and everyone spoke only English in class. The textbook relied heavily on pictures.
Lots of people don't like the direct method, and I can understand why it's unpleasant, but it worked really well for me. I had German lessons in school for a long time, but never gained anything like the fluency I got in English - the German teachers varied from some very archaic grammar-translation stuff ("durchfürgegenohneum", "ausbeimitnachseitvonzu" etc.) to the ones who seemed to have no plan at all.
LLMs should be able to teach me with the direct method, though fine-tuning would help.
> the German teachers varied from some very archaic grammar-translation stuff ("durchfürgegenohneum", "ausbeimitnachseitvonzu" etc.) to the ones who seemed to have no plan at all.
Native German speaker here: such grammatical structures are really how I think about my mother tongue, thus I do believe these are good exercises: always have the grammar tree in the back of your mind when you speak German "and then speak this grammar tree". :-)
In opposite to English, in German grammar is quite important.
It would be at least three different translations of the same sentence from English to any language with a grammatical gender. I hope you would experience how bad it is by yourself.
Anecdotally, one of my parents' friends told me her daughter, a Frenchwoman living in Mexico and who works as a translator, has seen all her work vanish other than reviewing AI translations.
More translation work for customers, perhaps, but reviewing is grinding work she has no interest in pursuing, and has the luxury of being in a position to do so.
Duolingo is actually a great usecase for AI. It doesn't teach you a language in any meaningful sense, but is a fun side activity to practice. AI is great for apps like these.
Yesterday I was in the tech support discord for a large yc-backed company and after 25 minutes had plenty of replies but no useful, actionable info. Out of curiosity, I copy/pasted the question into ChatGPT. GPT's response was similar to the human's (~"look at the docs and do xyz") so I followed up with "did all those, now what?" and it immediately gave an accurate, actionable answer, and even mentioned an aspect I (and the company rep) never considered.
Tl;dr AI gave the answer (plus bonus material) 2 orders of magnitude faster than an experienced human. It enabled me to go and try those things, reducing frustration caused by incomplete docs, product shortcomings, and slow (i.e. 'human') cycle times in replying.
AI gets things wrong and hallucinates, but what it's replacing (humans) aren't exactly perfect either.
I have a subscription to both Duolingo and Memrise. Memrise I always struggled to get into even I initially felt it was useful. since their latest relase they've gone totally into the whole AI thing and the app is asking me to watch free youtube videos as part of the lesson that it selects for me. Totally useless.
As for Duolingo it has invested too much in gamification / addiction. If you compete in tournaments you can spend around 1-3 hours every day "competing" yet your skills won't improve much. That means spending money to waste your time. In that case it's better to just stick with the regular lessons. Duolingo definitely suffers from "enshitification" and next year subscription I'll downgrade to free-tier because it's just stupid.
IRL I signed up for 2 hour lessons twice a week. They are group activities that are not very intense but coupled with my homework bring me quicker to where I need to be. Plus they're a lot of fun, and I've made contacts IRL too.
Duolingo was one of the first companies to partner with OpenAI on leveraging their language models. I think they were using GPT-3 before ChatGPT was even popularized.
LLMs are the masters of bilingual bullshit, so really Duolingo has no choice but embrace them. The LLM stuff embedded in their existing product is challenging in a different and better way.
I spent 18 months on it and the learning was soundly thrashed by three weeks in a classroom, and for all the pupils, not just me, primed for success by 18 months with Duolingo.
Unless anyone has had a different experience, Duolingo is an exercise in self-deception. No doubt there will be people in difficult situations that have managed just fine with Duolingo, or in spite of it, but it doesn’t feel like a growing market.