Learning a language is such a large, long term undertaking that I appreciate how Duolingo tries to use a few tricks to keep people on-track. It's also one of those areas where interests and incentives (maximising the time on app; regular usage) are rather aligned.

However after getting halfway into their Chinese course I feel quite disillusioned with their approach and actual content. You'd think an app with their market presence would have some amazing teaching strategies... but they don't. You can get through half of the course and still not know how to count past four. There's also lots of cultural context and finer points that are simply missing.

Anyway, I'd be curious to see how a more community-driven approach could play out, any whether it would lead to better content.

I find Duolingo is pretty good for vocabulary in a "slow" context.

The trouble is, that slow context is already better served by translation apps.

Duolingo is really bad at developing verbal fluency, which is the thing you actually need in today's world of translation apps.

No, the best way to learn a language is comprehensible input. Every other language acquisition method is bootstrapping that eventually needs to segue into actually using your target language to read or listen to things, unaided.

What these bootstrapping exercises are doing is not unlike, say, what early expert systems or Cyc did with AI. They aren't so much building an understanding of language as much as they're handing you a bunch of logical rules to parse out into sentence constructions. The problem is, that's not how human language actually works. In fact, it's not even how humans use programming languages, even though those do have formal specifications.

If you want an "open source Rosetta Stone" what you want is Anki and a flashcard deck for it. But even then, that's limited to vocabulary memorization, which is just bootstrapping. Personally, if you wanted to build a good language acquisition app, you almost certainly would want to have some kind of large language model in there powering it.

Duolingo is sorta like flashcards and I think it makes a good easy entry into learning

But flashcards that connect words and concepts are still much better than flashcards where you merely translate.

I think you're mistaken?

The grammar translation method is seem as obsolete, but Duo isn't that. You don't learn rules formally (e.g. memorise explicit and formal rules on how to conjugate a verb in the past continuous tense, and what all these rerms mean) then apply them.

If anything, people constantly complain about how Duolingo just gives them sentences and doesn't give long explanations about the grammar, you just have to pick it up. Very modern.

People also complain about how duolingo has "nonsense" sentences, because it deliberately drip feeds vocab in similar categories which is actually the right way. You learn one fruit, one colour, one body part, etc at a time; so yeah occasionally you might get something like "tom has a purple apple on his nose" but there's a reason for this.

The only real faults with Duolingo is that it focuses on listening and reading, so you need to practice speaking and writing elsewhere. It does have an AI chat, but it's... kind of bad IMO.

And that most courses only cover a year or two of learning. And that there's very few languages. But if you want to learn enough to get started in more immersive learning, IMO it's fine.

And there's people who complain that they spend so much time metagaming to try to win the weekly leaderboard that they actually hurt their learning, but if you really need a cartoon owl to give you a cartoon gold medal then maybe you shouldn't blame the app ...

duolingo doesnt do grammar but it does translation. Unless you want to become a translator then theres no point in learning how to translate from language A to B. What most people want is to understand and speak which is a different skill than translation

So, like, learning to figure out what someone said, or say what they want in another language. Sounds a bit similar to translation. Saying it doesn't transfer can be claimed about anything. Talking 1:1 with a tutor isn't talking 1:1 with someone else, but it's a question of how much transfers.

Depends on the context whether it's a good task. Duo seems to aim for low friction and beginner level (A1 ish), high adherence / engagement. It's an app, they can't just say "read this text and think about it", feedback and assessment matter. They ab test like crazy, and have multiple task types, I'm guessing translation works in the real world for their purposes (low friction, app, beginner).

The received wisdom that translation bad comes from the pendulum swinging away from grammar translation.