In the Venn diagram of (1) vocabulary you know and (2) reading material that is available, there is a looooooong stretch where that overlap is very small and hard to find. Basically, it’s whatever your training program (Duolingo, Renshuu, etc.) serves to you and, eventually, Tadoku books and other graded reading material. I’ve thought a lot about how to increase that overlap, so that if you are grinding through vocabulary decks separately from any other course or computer-based learning program, you can get material that matches your abilities.
So, this is what I did yesterday:
1. Downloaded the JSON of my JPDB data.
2. Used this site to convert it to a CSV, so I can open it and get my total vocab list: [https://www.convertcsv.com/json-to-csv.htm](https://www.convertcsv.com/json-to-csv.htm) (You’ll have to figure out yourselves how to export Anki data into a useful format.)
3. Opened a fresh instance of ChatGPT (3.5, which is free)
4. Wrote, “Using only the following Japanese vocabulary, generate a set of short stories and practice sentences that do not exceed N5 grammar. \[PASTE VOCAB LIST FROM CSV\]”
For context, just crested 1000 words, having completed the N5 Vocab List along with my mining decks.
It replied with:
>
I wrote, “Regenerate them without English translations” and it recreated all 15 sentences without the translation. All of the sentences were made up of vocabulary in my list, and mostly used grammar concepts I knew, and the rest I worked out with my tools.
Then I wrote, “Using the same parameters, write a short story.” And it produced this:
>お姉さんは、毎日学校から帰ると、箱の中にいくつかの本を入れています。彼女は本が大好きで、家の一番静かな部屋に本棚を持っています。ある日、彼女が学校から帰ると、家の前で小さな子供が転んで泣いていました。彼女は子供を助け、入口の近くに座って話をしました。子供は笑顔で彼女に感謝し、彼女は笑顔でお礼を言いました。
ChatGPT struggles with adhering to specific parameters, and in the past, when I’ve tried things like, “Write a short story without using any words longer than 3 letters,” it just can’t. (There are a ton of hilarious results on r/ChatGPT from people trying to play word games like hangman with it.) When I’ve tried to use it as a Japanese conversation partner, the training that drives it to respond with certain ticks overcomes my request for using basic vocabulary and grammar, or perhaps I haven’t found the prompt that defines what I want. It surpasses my skills immediately, and isn’t very pleasant. However, *I think there were only 2 kanji in this short story that I didn’t know!*
This is as far as I’ve gone, but I wanted to post the results in case it helped anyone else, or if anyone has tried this and has any tips to get good results. I’m not recommending everyone switch over to this method. If you can read Visual Novels and are making progress, I salute you. But, for the large cohort of beginners who are in that desert, just grinding through vocab and computer lessons, hungry for material that doesn’t require a bunch of mining in advance, I think this is worth a try. I’ll definitely keep playing with it and check back in with updates.
I would love to hear from native and advanced readers if there is anything in this story that catches your eye! Am I going to learn bad Japanese from it because it tried to cram some of my vocab list into the story in awkward ways?
EDIT:
1. If you have never used ChatGPT, and don’t have any experience on how it handles Japanese, but you want to say that it’s not a good idea to use it, I don’t think that it constructively contributes to this project. If you have particular warnings about the text I included in the post, or from your experience using ChatGPT in Japanese, I would love to hear it! A lot of people seem to just think “that’s not a good idea” based on..????
2. If you think there is plenty of reading material available for N5 students, drop links. I already know about Tadoku and Satori Reader. Unless someone can unlock some treasure trove of material that I haven’t found, I’m pretty convinced that most of those who hold this opinion were able to climb the cliff and grind their way to native material, and forgot how difficult and frustrating it is to find material that matches early reading levels, and how bewildering it can all be.
3. Some people seem to think this project is about having material that makes a student fluent. It is not. It is about having material that can help bridge a gap into native content. It’s ok if the material sounds like a beginner wrote it. You don’t have to worry about people getting stuck there forever. The goal is to get words outside of a vocab deck and into some sentences, modified by particles and conjugations, to help with memory and see it in context. Unless the text I posted has plain errors, if it just sounds stilted and boring, but uses basic N5 grammar concepts, that would be a win for a lot of students. Like, Duolingo doesn’t even do short stories in Japanese until the later units!
by mountains_till_i_die