Sensense manin and listing to japanse

Book Clubs
#1

I’ve been using AI as part of my Japanese reviews, and I’m curious what other people are doing.

One of my main goals with Japanese is to reach the point where I can engage with media: books, audiobooks, anime, games, and so on. I’ve repeatedly been humbled by how difficult Japanese is. I honestly cannot overstate how hard I’ve found it to think in the language.

I’m now getting to a point where I can recognize more sentences and get a general idea of the meaning, but I still do not know enough vocabulary to fully engage with native material comfortably.

What I’ve been doing recently is buying Japanese books and using AI to help me identify grammar points and sentences that I may not know yet. For example, I’ll take a book, run parts of it through AI, and ask it to point out grammar structures based on what I have already studied or what I am likely to struggle with.

In one of my recent books, I ended up with around 700 selected sentences/grammar points from the book. I’m going through those as quickly as I can. I still have a lot of Bunpro sentences and Anki reviews, so the book sentences are more of a background project for now, while Bunpro remains my main priority.

My next goal is to start reading these books while also listening to the Audible version at the same time. I don’t think I’m anywhere near being able to understand anime naturally yet, but I hope that reading and listening together will help bridge the gap.

One of my biggest problems with Japanese is that my reading, and writing ability is much stronger than my listening and speaking. I can sometimes make sense of written sentences, but when people speak to me, I often cannot understand them quickly enough. even with words I know when I heard them how sounds like just random noices And when I try to speak, it takes me a long time to form what I want to say because Japanese feels so different from the languages I already know.

So I’m curious: what are you doing to learn Japanese and engage more with the language?

Are any of you using AI for sentence mining, grammar review, reading support, or listening practice? What has been useful, and what has not?

1 Like
#2

At my current level (should pass N2 @ 120-130 points), Japanese Output feels like it is 10 to 20 times harder than input, and from what I’ve gathered there’s no real shortcut there, practicing output is the only way to get good…

If you are comparatively struggling with listening but are a good reader, try to find subtitled videos on Youtube. I was in the same boat last year but watching 1000 hours of voiceroid 実況 content puts me at the level where only ojisans talking super fast and slurring their words filters me. There’s probably other content you might like with japanese subs, jimaku.cc has a good selection. Sometimes the subtitles are targeted for deaf people and will have a lot of text to describe the situation and background noise.

The thing I do the most with AI is taking a sentence I am not sure of the meaning, or where I have the meaning but I am not sure exactly why it has this meaning, and ask every question and follow-up questions. It is very good at that. I think you do want very intelligent models. My recommandations would be Claude 5 Fable, which was insanely good at that for the 3 days we had it, but GPT 5.5 or Claude Opus 4.8 does the job very well. I would avoid the Gemini models since they really have a tendency to tunnel vision on one problem they think you have and have a hard time staying coherent after four or five turns of conversation.
I also do it when I confuse bunpro grammar points or when I don’t understand why another pattern couldn’t work (usually because I’ve not clearly understood the nuance, bunpro is almost never wrong).
I’ve heard stories about using AI for mining uncommon words from a book, but I’ve never done it.
I’ve used AI to script a few changes to my anki cards, or even straight up making decks with them, mainly through AnkiConnect. You can really salvage old decks easily now.

1 Like
#3

It depends on how quickly you want/need to output.
I can only argue allegorically but I personally have learned English for 7 years and, while I have barely ever spoken it, I am completely fluent when speaking nonetheless. At some point you will just automatically remember phrases since you have heard them thousands of times. Although its of course never a bad thing to practice, I generally wouldnt actively worry about output unless you know you need to be able to speak soon. Enough input will eventually lead to output if you are willing to play the long game.

Edit: In order for this to work you of course must first do the rigorous work to get the grammar down and build an intuitive foundation. Of course English and Japanese are different, but you will be able to get fluent in either of them one day, and once you fluently understand a language, you will usually start speaking and thinking in that language (so outputting) automatically.

2 Likes
#4

I’ve had the same thing with English but starting from my mother tongue, English is trivial relative to Japanese. I’ve not yet reached the point where it becomes totally natural, but yes, at some point with enough input, your brain starts to try to output too. I think that at some point you kind of want to worry about output, since you will get situations where you have to output and never trying to output at all was a big error I made last year.

Generally I agree that input is all you need, but the fastest way to be fluent is to do some output practice (or even straight up do output in daily life, work, travel, etc etc). The long game can be very, very, very, very, very long. I still think you should avoid too much output at a lower level of understanding but at some point it can be also very good practice to help understand input better too. Most of the advanced learners I know IRL basically only focus on output nowadays, “no real progress being made with only input past a certain point” one told me last month.

2 Likes
#5

I’ve heard positive things about essentially babbling out your inner monologue in Japanese as you go about your day (internally or spoken). It’s something you’ll need to force yourself to do, though, especially with newer words and grammar, and it doesn’t necessarily help with listening skills.

For listening get yourself a lower level podcast and play it slow. Work your way up. Or YouTube videos. You need to get comfortable with not understanding everything, or even much of anything, and hammer new content in as much as possible. Your brain already has the base structure of the language, it just needs to learn to recognise it in speech.

All easier said than done of course. Unfortunately when I don’t understand a lot, I get so bored.

Another option is comprehensible input Japanese videos. They can work really super well and give you a lot more confidence (but I also find them dull at lower levels, and you need to watch a lot of them).

1 Like
#6

Me personally, I do have a native Japanese tutor led lesson (mostly in Japanese unless I don’t understand a word or don’t have the words to explain something and I confuse him when I try) where they correct my language in free talk, correcting pronunciation in reading out loud as well as doing mock conversations and asking me comprehension questions after reading passages out loud.

Alongside this, I do a regular language exchange with a native Japanese person every week where we discuss different topics and he also helps me to phrase things more naturally.

Both of these speak native speed unless I ask them to repeat something, in which case they will repeat it more slowly.

Input wise I have recently (this year) gone back to textbook study to refresh my understanding of things I’ve done previously but are struggling with, I use bunpro for grammar SRS, WK (through Kakehashi) for kanji/ vocab SRS as well as graded articles, stories and crosswords for recollection of readings.
I use reading native manga and children’s books/ tsubasa bunko books (or similar) for reading and watching anime or listening to podcasts, audiobooks or such.

I do feel that my biggest leaps in understanding were caused by comprehensible input through reading stories or watching Comprehensible input videos. And my biggest leaps in picking up the sounds of the language were from listening constantly over extended periods of time (several hours in one sitting - passive listening) to either audiobooks, podcasts or even dramas/ films in the background while doing other things.

Saying that though, I feel now more than ever that I need active listening and intensive reading to understand and improve my understanding than I need passive listening and extensive reading.

2 Likes
#7

Every single person I know who tried to use AI for things like you are describing eventually realized how often they hallucinate doing this sort of task. You don’t really know what variables you’re throwing in when translating from japanese to english, translation can never be exact due to the different levels of context required by the two languages. In general, mixing languages is a known problem for LLMs, and I would be careful.

5 Likes
#8

This, I wanted to interract with this thread but I tend to give up when I see people using AI to learn a language.
They generally will argue it’s really good and that you also need to use AI or you are missing out.

Then I read that and was like ” ho my god, you can’t even use basic free models and you need to try multiples ones? That’s so much hastle to probably have hallucination”.
But I don’t know if OP want his thread to become another pro AI or anti AI thread.

So as OP guessed, I don’t use AI and try to find hinative or wanikani or reddit thread on any point I struggle with, though as I am still in mid N5 it’s easy to find answers compare to complex grammar points.

I think watching anime is a totaly different skill than reading with the audio book on so you will have to jump into it and be ready to struggle the first couple of episodes. Start with easy anime such as pokemon, digimon, yugioh etc and mine some of the words in anki to slowly familiarize with the vocab.

Listening is also another skill and you will to practise it with podcast or maybe just audiobook that you just read as you would know the story.

Speaking is the same though there are two schools for it, either you actively try to speak as soon as possible with vocab and grammar you just learned or you will start forming sentence after a long time of immersion as you will have learned tons of vocab and it will come naturally.

For immersion I personnaly do extensive reading on the free tadoku greaded reader site and Jrpg sakura then use comprehensible japanese for video/audio ( I try not to watch easier video to practise my listening).I do between 30 min to one hour of each daily.
Also I can only highly recommend Satori reader that will give you even more grammar points with great story, there is a free version if you want to try it out.

Dont forget learning a language is a marathon so try multiple learning method and see which one works for you.

3 Likes
#9

Thats why you use deep-thinking-mode when doing more complex tasks, and tell it to cite its sources. Hallucinations can be significantly reduced due to this.
Especially when using modern models like Claude Fable (if the U.S Government allowed it lol). Although for stuff like identifying grammar (which OP said he was using AI for, not tranlating) it should work out without worrying too much as long as the grammar isnt very niche or antique/classical.

1 Like
#10

To add to it, you should only be caring for the vibe, not the literal translation or output word for word.

#11

OP asked for an AI-related discussion in their last paragraph though lol.
Anways, yeah, if you dont want AI to hallucinate you have to put some effort into it, aka go through that “hastle,” and also pay for good models. Thats usually the case if you want quality of something.

I think there will be more and more “Pro-AI” people as AI will get better, and since the general sentiment will inevitebly change the better it becomes (and because the next generation will most likely grow up using it), discussions about it will also increase. I don’t think “giving up” is the best thing to do. At first, like two years ago, I was also very strictly anti AI, but when I noticed that it genuinly gets better and isnt just another tech-bubble, I tried to find ways in which I can actually use it productively, and learn as much as possible about it in order to mitigate the downsides.

1 Like
#12

Would joining a Japanese book club be motivating for you? (I realize they do not use AI, but they can help improve comprehension). On the Wanikani forum, there are many levels of book clubs, and people help each other with the meaning and context. You don’t have to use the Wanikani platform to participate in forums.
I find the groups motivating, and I think the AI would be overwhelming for me! As well, I use a textbook and have a tutor every week, plus practice outside of class,

1 Like
#13

Oops missed that!

Ho when I said I was giving up, it was talking with AI enthusiast.
I will never trust AI, I use it from time to time to see if it got better and it always give me wrong command/line of code (even halucinates it for discord bots) or completely mixes info on a subject (lastly on a statue in a french museum). Though you will say it’s because I use the free chatgpt/ Gemini that uses an old model.
I will never pay for AI and I think you will be surprise, for me AI is just the next NFT and it will fizzle out as it will be difficult to find a group of people willing to pay the real cost for it(price increase is coming again).
AI psychosis is a proof that most people can’t use AI correctly.
Though if you like AI and put a lot of effort into it to get what you want out of it, go for it!
That’s great that you found a use case that works for you and that you know all the problems it can generate.
I don’t have the patience to learn how to prompt or double check every AI answer so I rather directly post my questions online after searching for it on the no AI duck duck go.

1 Like
#14

Yes, paid and larger, newer models are better than free and older ones. There’s a lot of progress even in 3 to 6 months. Use larger models if you want more real world content, by definition smaller and cheaper (ie free models) have less knowledge since they have less space to encode information, and/or have been ruined with too much RL in order to benchmaxx (I see you Google, I see you random Chinese lab n°10384)

You can use most of the same tools you use when you want to check whether normal, real people are saying the truth. I feel like they don’t hallucinate so much as they bullshit. You really don’t need to prompt in any specific way, most of my prompts are simple natural languages (looking at my previous requests 90% are a variation of “what’s the difference/nuance between 〇〇 and 〇〇？”). Rarely any issues when building or learning on what the reply is.

If you want I can run that request through a larger model, but the one I am the most confident would work just got banned by the feds lol