Here is an experiment I have been doing. I am Beginner2 approaching 2k words learned in my TL. I have been having ChatGPT write a short story in my TL and asking it to include 20+ words I want to work on. I find I “learn” a word in one piece of content then run across it in another and don’t recognize it. Thought this would be a way to work on seeing known words in different contexts.
I landed on 20 because experimenting with 100 showed ChatGPT could not handle that volume and was dropping words. Consistently it includes the 20 when I ask it.
What comes back from ChatGPT is a short story with 50% known words, 25% words that are still Lingqs and 25% new words. I like that ratio. The stories are well formed, but sometimes the elements make no sense. A group of soldiers celebrate a victory in war by eating ice cream. Things like that. However the context is still there enough to get meanings of words in the setting, even if silly. Its a lot of work rather than cruising Lincq’s generic content, but it is fun so keeps me engaged.
yeah, when you will have chatgpt integrated, we can ask to create endless mini stories with our preferred genre and at the level we have. That would be awesome. Tons of content without wasting much time in finding it.
I heard that you can tell it to continue the story when it stops but I haven’t tried it out yet. With Korean it just stops after one paragraph. I like weird stories as long as they’re not so weird that I can’t understand them.
This is exactly the same strategy as I thought of. I think it will increase retention by a decent amount which at the end of the day is what it’s all about.
You’re right: it’s hilarious the wierdness of the stories. It’s like something a little kid would write. But as you say, I don’t care as long as it’s providing legitimate content.
One thing I will say that I noticed since I’ve been doing it the last couple days:
My head hurts when I’m reading the stories.
What I mean by my head hurts is the feeling I get is the same feeling I got way back at the beginning when I was learning Spanish. I was more or less high beginner. I took a class for a couple of weeks which was all in Spanish at the same time as I was speaking with native speakers a few hours a week. What I noticed was that my head hurt during and after the class and also during and after my sessions with native speakers because I was concentrating so much. I think I leveled up a decent chunk after that period of time.
It is possible IMHO that it is therefore working. We’ll see. After a couple months I’ll see my stats and I will know if my retention has jumped.
just type “continue”.
It doesn’t actually continue the story though. It does some kind of a riff on the original story which is kind of discontinuous from where it was at. Like two similar but not the same stories just cut off arbitrarily and joined together.
Still works to get exposure to the language but it’s wierd.
Yes, you’re head hurts when you’re doing too much effort. The same as if you are doing to much physical exercise: your body is going to hurt. Our mind tends to go to trance or avoid too much exposure as a safe measure probably. But with chatGPT you can ask to simplify the language, you can ask for A1 stories, A2 and so on. It doesn’t always retain it but it’s quite good. It depends also on the language. They say Claude or Sage are better in other languages, I believe you have to try with different bots in one specific language to see what suits you best.
When I was in France, even after studying French for months, I started a job as hotel receptionist and my manager left me there disappearing all day. (she was great though)
The constant exposure to the telephone with multiple calls and people arriving was devastating, at the end of the day my head was always hurting. But my mind was integrating very fast the “passive” language in active and after a week I was confident on the telephone and didn’t have any headache anymore. Impressive how fast we adapt if we have no choice.
Awesome. What it feels like is different than when I’m reading on lingQ. If it’s too easy it’s like nothing. If it’s too difficult, what’s happening is that I’m decoding it instead of reading it.
With the chatGPT stories written with words that I’m learning in anki, I just about recognize them so I believe it’s literally exactly at the specific point which is barely “comprehensible input”.
I’m really hoping the sore head is a symptom of material that is at exactly the sweet spot.
In any case, I will be able to report back on the stats in a couple months. If my retention rate has jumped then it is working.
Btw, in one way or another there is something here that could be a great strategy. When these bots will improve, it would be incredible.
But actually, this could be a great strategy for LINGQ as well. Are you reading this Lingq Team?
LingQ could exploit the vocabulary sitting there to create stories with ChatGPT. For example, we could ask to write stories with a percentage of known words and words at level 3. So it would be easier to level them up.
Or we could ask to write stories with a mix of known words and a specific TAG. This could be great to review collocations, that we can tag while reading.
You know what? A story chatgpt just concocted involved two engineers working on a tunnel and one of them decided to take a nap in the middle of a job on a park bench with a blanket.
Because park bench and blanket were two of the words in my word list.
I have the picture of the engineer napping with a blanket in the middle of a worksite with a park in it now burned into my head. Plus I now know for sure the words for blanket and park bench. Ha!
you know what? Now that you’re talking about pictures, and seen that you are using ANKI, you can actually create pictures with midjourney or stable diffusion for some of the words or concepts you would like to have an image too. I’ve just thought about it right now. It would be a lot less time consuming than searching the web.
That’s actually a really good idea because sometimes it’s hard to find an exact image that corresponds to the image that I have in my head as opposed to the idea that someone else might have for a particular concept. For example “dog” in my head is some kind of bigger dog somewhere between a german shepherd and a labrador. For others, it might be a small dog. Likewise “skillet” means specifically a square cast iron pan with grooves that is dark in color. I believe (have not much proof) that something that most closely resembles your own idea of the core conceptual picture you have in your head for any given concept will be easier to recall than something that is kinda sorta.
You can steer it a little. Rather say “continue” the story, but maybe give it some additional context to follow. This has worked for me when I played around with a number of weeks ago. ymmv
Since we last spoke a couple days ago I discovered you can say
“conclude the last story you just wrote in three lines” and instead of riffing on some variation it will actually conclude it.
Interesting. I used ChatGPT to write a story as well as solve some coding problems a few months ago, and just prompting it with a “continue” always did the trick it’d continue right where it left off (I checked). It would even start the next answer with an 's if the last word of the previous answer was missing it.
yes, usually it does, it keeps track also if it says (when asked) that it doesn’t remember what it wrote before. I think this feature is a work in progress. Sometimes it loses track. It depends on the initial prompt and after a few answers we should ask to remember the prompt. But it doesn’t always work at all.
(she was great though)