Hacker News Reader

Comment by benatkin | Hacker News Reader

It's worse than virtually any other way to learn a language. Saying that you can't expect it to be perfect doesn't justify it being really bad.

DuoLingo isn't optimized to learn a language, it's optimized to give people the feeling they're learning a language.

Flash cards are better, books are better, CDs are better, Spotify is better, YouTube videos are a lot better, and learning in person is way better.

View on HN

But is that actually true? When I see people say Duolingo is bad, they're always comparing it to some highly optimized language learning strategy, where you weave together Anki and suchandsuch grammar resource and a strict regimen of 30 minutes of target language media every day.

I'd agree that Duolingo significantly compromises effectiveness for the sake of engagement, but to most people engagement is the hardest part of learning a language!

Yes, it's true. Using Duolingo is significantly worse than just about all your other options, including picking up a $30 Living Language textbook/workbook/CD set of whatever language you want. Living Language isn't aimed at university students or the most motivated people either. It's just one of the cheaper options.

This is not just a throwaway comment, either. I've built and run a language school for four years (with two partners), taught thousands of children, worked at a language learning startup, and have read a lot L2 acquisition papers and know this space pretty well. I also contributed to Anki long ago, but I'm not a huge fan of using SRS for learning vocabulary. Extensive reading (and listening once you can) is a far better strategy. (Anki is pretty good for memorizing discrete pieces info though, such as how to write each Chinese character or all the atomic weights and numbers on the periodic table).

As for engagement, I think Duolingo is pretty boring compared to making new friends, talking with people or even just reading a book about something I'm curious about. If you're learning any popular language (e.g. English, French, Spanish, Mandarin, Japanese, etc), you'll be able to find some pretty good graded readers aimed at beginners that let you start actually reading in your first few months of studying.

> I'm not a huge fan of using SRS for learning vocabulary. Extensive reading (and listening once you can) is a far better strategy

Can you explain your logic and how you measure success? FWIW I'm ~8 languages deep and started using SRS with my 5th language. For the last couple languages I've tracked my metrics extensively (i.e. words learned and retained per day). Also with the SRS tool I use I read sentences along side the words I'm learning so I get some context.

There's a lot of L2 acquisition research on extensive reading and it's definitely worth checking out.

I covered some of my reasoning about the limitations of SRS in the 2nd half of this podcast: https://alchemist.camp/learning-machine/spaced-repetition-sy...

Seems like you pretty much contrast:

[1] SRS flashcard apps as an exclusive language learning system

[2] Extensive reading

While (2) is clearly better if those are your only choices, as you point out yourself SRS can easily be integrated into other curriculum or used as a supplemental tool. I'm not sure how many languages deep you are but I find getting through the first 1000-3000 words a real slog in the language and it is also quite hard to read in a language when you are at sub-3000 words unless you are reading especially prepared texts in a textbook for language learners.

My preferred method is to do the basic grammar of a language and first 2000 words as quickly as possible with a cram method and then work on speaking and transitioning from passive to active vocabulary. SRS is a nice tool, esp. if the SRS tool you are using also has included sentences in context (I mostly use learnwitholiver.com). Then usually I work to train for the state run test of the language that I'm studying at a reasonably high level.

Turns out Gwern has a very well researched article on SRS: https://www.gwern.net/Spaced-repetition

That's odd. Your strategy of rushing through basic grammar and a small number of words quickly is actually a use case I said SRS is useful for.

In the link you're responding to, I said:

> "The second situation where spaced repetition works really well is when you're studying something more complex, but you can use spaced repetition as a scaffold to get through the initial phases a bit faster. Learning vocabulary is like this. Though there are the issues that pointed out earlier about context and collocations and word boundaries, spaced repetition can still be a huge win. And that's because you don't have to use it to learn and remember everything. You can use it to get a scaffold, to get started with whatever foreign language you're learning or medical vocabulary or whatever kind of vocabulary. And after having memorized, say, five hundred or a thousand or even five thousand words in this new area that you're studying, you can then do a lot of reading and use those words and encounter them in many different contexts and actually acquire a solid grasp of that topic."

I would definitely start with training my ear to the sounds in a language at the very beginning, but flashcards can also be a useful supplement early on. It's funny you mention Oliver! I was in contact with him when he first launched his site (just for Chinese, initially). For what it does, it's made well. LingQ is probably my favorite for getting the initial base down, though.

FWIW, I don't think I've every encountered anyone who saw SRS as an exclusive language learning system, except maybe a few hopeful app developers who have realized how easy it is to make a flashcard app.

The real question is how much of your study time do you want to spend on decontextualized sentences or words vs time spent on sustained reading, listening or conversation?

I track my time via 22 minute segments and typically do ~3 per day of vocab vs 9 per day of structured language learning and another 6 per day of reading and writing. So of the structured time I spent language learning it is a bit less than 20% of the time I spend. My target is about 100 words per day but my actual best time over an extended period is about 58 words per day on tracked words. Oliver is good but there are a bunch of features I'd add if I could. Will check out LingQ as well.

One of your criticisms of spaced repetition is the difficulty of representing associations between words on flashcards. I agree that that's difficult (though you can mitigate it by including example sentences in your deck). However, I think that that criticism expects too much of spaced repetition. It's a very powerful technique for learning and holding onto new vocabulary, but it doesn't replace reading, listening and speaking. You really can't learn nuances of word meaning with flashcards alone, but you can build up and retain a large vocabulary. You can deepen your understanding of the subtleties of the language in other ways.

I would say that Duolingo does the distinctly American thing of optimizing for making you “feel good” about your progress, even if that progress has no correlation with actual language ability. It’s more of a self-esteem simulator by way of language-learning than it is a useful tool for foreign language instruction.

The point is that if you want to learn a language, you either follow an intensive regimen like the one you've described, or you don't and you make essentially no progress. Learning languages is not rocket science. It's mostly just work and dedication.

What do you mean by worse? I’ve spent a lot of time on Duolingo as well as in face-to-face language classes, in high school and university. I’ve also spent a smaller amount of time self-studying from books, CDs and software that predated Duolingo. I’m not fluent in any of the languages I’ve studied because of a lack of immersion, but I learned a lot more from Duolingo than any other source because it was so easy to establish a daily habit and stick with it without having to pay anything. I’ve had basic conversations as a tourist in France after learning French exclusively from Duolingo. I’m sure I would have learned more if I spent that time in class instead, but I didn’t because it’s very expensive and inconvenient compared to spending 30 minutes a day on Duolingo while I’m commuting.

It is inefficient. The vocabulary lessons on Spotify run through words a lot faster. You don't have free reign of what category to study things in. It teaches vocabulary when you should be learning grammar. It teaches pronunciation just by having you repeat words, which is inadequate - the specifics of pronunciation (including emphasis of syllables) must be explained.

I don’t doubt that it’s less efficient than learning from a human teacher, if you just look at the time spent by the student and ignore the “efficiency” of software that interacts with millions of students at the same time. I’ve never heard of anyone learning a language by passively listening to Spotify, so I have a hard time seeing it as a Duolingo alternative, but I’m sure just like Duolingo and other language tapes, it would be a valuable complement to any other language-learning tool. None of them will make you fluent without immersion.

What makes you say that explicit instruction in grammar (which does exist in Duolingo, although you’re right that the emphasis is on vocabulary) and pronunciation is necessary? That’s not how people learn their native languages. Considering that anyone can learn any language with enough exposure, and most people fail to achieve their language learning goals due to giving up after less than a thousand hours, I’m skeptical of this kind of pedagogical criticism.

Language acquisition after the developmental window is fundamentally different. Native language learning is not a good model for second language learning.

The evidence for that “is limited, and support stems largely from theoretical arguments and analogies to other critical periods in biology such as visual development”: https://en.wikipedia.org/wiki/Critical_period_hypothesis

I think it's not so much that second-language acquisition is harder than first-language acquisition, but that first-language acquisition is horribly slow. After five years of full-time study, do you want to reach the level of a five-year old or do you want to be able to handle any communication you could handle in your first language? Not to mention that you probably don't have the time to study a language full-time anymore.

Babies spend a lot of time babbling, basically brute-forcing their way to a correct pronunciation. But what if you could make use of phonological research and resources like https://www.seeingspeech.ac.uk/ipa-charts/ to help you get there faster?

Parents spend a lot of time playing "look at the X" games to help them learn vocabulary from examples. But what if you just look up unknown words in a dictionary and use spaced repetition to efficiently retain them?

Second-language learners have already bootstrapped their way to being able to communicate with other humans. So why not just make use of that, rather than repeating the same slow process of trial and error for another language?

>do you want to reach the level of a five-year old or do you want to be able to handle any communication you could handle in your first language

In terms of grammar and vocabulary, five year olds have a much higher level than all but the best L2 speakers. I would regard the ability to speak as well as a 5 year old native speaker as an excellent outcome for five years of full time study of a language. Very few people achieve such a high level within that time.

> In terms of grammar and vocabulary, five year olds have a much higher level than all but the best L2 speakers.

I don't know about grammar, but I found a study [0] that estimates children in kindergarten (age 6) to have a mean vocabulary size of about 3000 root words. If you wanted to beat that in 5 years using spaced repetition, you'd only need to add less than 2 new flashcards per day. In my experience, 10+ new flashcards per day is doable. (When I was adding that many, I spent an hour a day on reviewing flashcards, which eventually made me scale back because I was only studying for fun in my free time.)

If you know full-time language students who don't have as large a vocabulary, I wonder how full their time actually was.

[0] https://www.researchgate.net/publication/232518465_Estimatin...

That's an estimate of expressive vocabulary (i.e. the range of words typically used). It's very obviously not the case that a typical five year old knows only 3000 words of English.

The methods section of the study I linked begins as follows:

Testing reported herein was conducted in Mayand June of 1999. Testing was done individually for children in kindergarten through Grade 2. The investigator introduced herself and said

I’m going to ask you what some words mean. Later, I'll ask you to read the same words and some others. Some of these words are common and some aren’t. If you want to stop at any time, you can. I’m going to read you a sentence and then ask you what a word means in that sentence. You can use words, pointing, or acting to explain the meaning of that word.

We then gave three examples, including, “ ‘I bought a new car.’ In this sentence, what does car mean?” We continued with “ ‘Johnny fell and broke his arm.’ What does arm mean?” If the child had difficulty with these examples, we would go over them and explain how to answer. (In the arm example, pointing to one’s arm would do.) A final example involved a hard word: “ ‘The material was translucent.’ In this sentence, what does translucent mean?” This example was given so that we could explain that there would be words children did not know, and that it was alright to say, “I don’t know.” (No children actually knew translucent.)

This seems to me like a test of which words children know, not the range of words typically used.

I can't find the term "expressive vocabulary" anywhere in the paper. Did you assume that's what they were estimating because the number seems low? Actually, 3000 words is quite a lot. The Simple English Wikipedia tries to use only the most common 1500 or so words, for example. Most concepts can probably be expressed in words a five-year old is familiar with. It's just that not everyone takes care to express themselves that clearly, so having a larger vocabulary is still beneficial.

>Did you assume that's what they were estimating because the number seems low?

Yes, look for any estimate of 5 year olds' vocabulary and it's much higher than this (unless it's an estimate of expressive vocabulary).

You don't really need to look at studies, though. Just talk to some 5 year olds.

This is a good point. I did take a class on phonetics at uni, and it helps a lot with learning the basics of a new language. Maybe Duolingo should do a phonetics course which teaches IPA using examples from a range of different languages!

You can't trust Wikipedia on linguistics.

> I’ve never heard of anyone learning a language by passively listening to Spotify

Maybe not, but learning a language by passively listening to podcasts is a popular strategy.

I'll plug our own mostly free-to-use project here, I think many will find it useful: http://languagelearningwithnetflix.com (bilingual subs for Netflix and more)

I find a very good way to learn languages by Netflix is to watch content in foreign language plus foreign subs. This is really good for English, because virtually all content is subbed with closed captions. French on the other hand is nearly impossible since virtually all content has mismatches between the audio and the caption (i.e. they always have subtitles, never closed captions). I've yet to find way to search for closed captions (vs subtitles) on Netflix.

Way ahead of you :-)

http://languagelearningwithnetflix.com/catalogue.html#langua...

(should update the catalogue data though, bit old now.)

I was sceptical, but this does look neat. I am German and trying to learn Spanish.

- Can one disable the voice when looking up words in the dictionary?

- Does it only work with Netflix or do you have plans for Amazon/Disney+?

- Do you have large costs (hosting?) and plan paid pro features?

-- No, currently TTS always plays on opening dict. It would be simple to put an option, just maybe don't want to overload the user with options. If the issue is that the voice isn't very pleasent, maybe it's better to address that.

-- Just Netflix for now. We'll release a beta version for Youtube in about two weeks. Most of our users are in East Asia (75k in Korea), so, we'd look which services are popular there.

-- Hosting costs, well, not too bad. We use firebase.. outgoing data is 10c per Gb, but, we might add a few nginx catches on a digital ocean VM (in different regions) for some endpoints to get this down, and reduce latency for dictionary. Machine translations are the main cost (it's actually a paid feature already).

Oh, wait. There is a way to open dict without TTS. In the settings, look at 'On Left Click'.

At first glance this is really well done. Subtitles and translations for Japanese are tricky due to furigana so good job!

Thanks.

How many people go through these other better methods for learning a language? In that sense, it’s a great way to learn a language if you wouldn’t otherwise be learning a language.

I think most go through a bunch of them. I have, and I've used DuoLingo, and it helped some. I might not be giving it enough credit, as it did help me get started, but since exploring other methods I've stopped using it.

Using an ineffective language learning tool to maybe memorize a handful of terms is not in real terms more useful than never getting to or dropping other methods. If anything, it’s worse because it instills people with a false sense of their progress at a real goal (language acquisition) by simulating it with a fake one (accomplishing tasks on Duolingo).

Reply on news.ycombinator.com