Lately I’m seeing a lot of social-media ads for Japanese-learning apps based on AI: ones that listen to your speech and fix mistakes, ones that use movie/cartoon clips and go through them for you, etc. Has anyone tried any of these? If yes: can you recommend one (or warn me off one)? I’m curious what works—and what doesn’t.
I like to keep my language learning simple currently using WK for kanji and vocab and pimsleur/youtube for speaking.
I have not used any AI learning app, but as a general rule anything that comes from an LLM AI cannot be taken at face value. These tools will speak with confidence, regardless of whether the things being said are true or not. This problem is pervasive in every popular usage (chat bots, search engines), but would be particularly dangerous in a learning app. Since the whole point of using a learning app is to learn something you don’t already know, you would have no way of verifying which statements made by the AI are accurate and which are fabricated nonsense that just sounds plausible.
My girlfriend tried such an app (for a language that wasn’t Japanese) and was thoroughly unimpressed.
I think language models can be a useful tool in your arsenal for language learning, for instance to assist with translation or break down complicated sentences, but I absolutely wouldn’t trust an AI-centric approach. It’s just not reliable enough.
I would actually argue that AI tools work best as an intermediate-to-advanced learner, when you can critically appraise the output and judge if it sounds right or if something is amiss. I often use ChatGPT to correct my output in various foreign languages for instance. Sometimes I use its corrections, sometimes I keep my version, sometimes I use a hybrid of both.
For beginners there are so many resources built by humans and thoroughly vetted, I don’t see the point of using AI tools that are going to get it wrong 10% of the time.
I use the AIs to help me figure out some grammar in some textbooks and it usually does a good job. That said, LLMs can’t be trusted. You always always always have to take their output with a massive grain of salt because they are confidently wrong. They are more confident than they are correct and will happily follow a broken stream of logic.
Their primary objective is to keep you coming back and staying hooked to the nice friendly bot and some people have even spiraled out into madness because they’ve fooled themselves into thinking its real sometimes with exceptionally disastrous consequences.
I used one for Korean and it was incredibly frustrating because they just don’t account for user’s speed, voice patterns, etc (I don’t have any more things to list but it sounds more extensive when I put etc. right?)
I think AI can do a lot of really basic things really well but anything that requires nuance or expertise, it fails spectacularly, even when you account for proper prompt “engineering” which most people don’t do.
You can always try them yourself and maybe they’ll work for you. They should (if they don’t just stay away) have a trial mode or something to get a feel for it without paying money. But be warned that it’ll probably have you doing basics and will be become less reliable as topics become more complicated an not so easily digestible for the AI.