NEW Story Bot for AI generated stories

[Steve]

Premium users will now see this button:

Clicking it will open the Story Bot, allowing you to generate stories, biographies, etc… using AI.

e.g. Choose “Classic Fiction”, type the name of a story, “The Jungle Book” say, click Generate, and in less than a minute, voila…

Give it a try and let me know what you think! If you come up with any particularly good or bad results, please share them here. (note that all stories generated with this tool will be automatically shared on the public library too)

[Marsel_Basholli]

WOW, I tried it out and it’s brilliant. I didn’t even know I wanted something like that. But it’s so useful, especially when you’re starting a new language.

[WarsawWill_RandomIde]

Hi Steve,

You’re really on a roll, so much added in just the last couple of weeks. I’ve tried it out in all categories in a few romance languages, and it’s an excellent addition. I imagine it could even get quite addictive, especially in the “Original stories” category. These stories have a certain similarity of literary style (and final moral), familiar from other AI generated stories I’ve seen or heard. And in mine at least, they often feature a young woman called Carla, usually carrying a notebook.

The challenge I think here will be to think up prompts to elicit a fun story featuring any particular areas of vocabulary we might be interested in.

[Steve]

@WarsawWill_RandomIde

You’re really on a roll, so much added in just the last couple of weeks. I’ve tried it out in all categories in a few romance languages, and it’s an excellent addition. I imagine it could even get quite addictive, especially in the “Original stories” category. These stories have a certain similarity of literary style (and final moral), familiar from other AI generated stories I’ve seen or heard. And in mine at least, they often feature a young woman called Carla, usually carrying a notebook.

Yeah, I’ve found that the original stories can be a bit bland and repetitive. I prefer the other genres where the story will be something I’m more interested in.

If anyone has tips for making better original stories with LLMs, please share! I haven’t gone far down that particular rabbit hole.

[WarsawWill_RandomIde]

You can add, for example, language points you want covered. I just asked for a story in Portuguese, titled “Planning a trip to a rock festival”, and then added in brackets “please include some future forms, vocabulary for equipment needed, and some ‘if’ structures”, and it more or less came up with a lesson on conditionals, including all three structures, and lots of Future Subjunctive - just what I wanted.

I think you can also specify characters and plot lines, etc.

[Phil]

Thank you for the tips WarsawWill; I didn’t know I could add extra things like that.

[Marius]

This feature has great potential. Is there a way we can make the AI generate stories from the word list?

The potential of this technology as I see it is not to replace human written storries, but to replace flashcards. The idea would be for it to use the space repetition algorithm used by the flash card system and generate stories using those words.

If I read the story without clicking on anything, the algo could assume that I know them. If I click on a word, it would reset the space repetition.

If it’s just random text, I’d much rather read real stories written by people or articles that have some content I’m interested in. The only instance I would chose GenAI content over human generated is if it accelerates my learning. If it doesn’t generate content based on some sort of personalization, based on my vocabulary level, then it’s not really superior to a text with similar difficulty.

As I wrote in another post, I think think the best way to improve one’s vocabulary the fastest is to choose conent based on one’s level, which would require a system to keep track of learned words, count the unique words of a new text, compare it with the vocabulary one knows and rank it relative to what one knows. That would allow one to keep a balance between new vocabulary and comprehension, which in turn helps with vocabulary acquisition. The advantage of this over GenAI content is that one reads with purpose. One reads what one is interested in. One read for both content and vocabulary acquisition. The drawback of GenAI content is that one reads it primairly to improve one’s language knowledge and not so much for content.

[Steve]

Yes! I’ve been thinking of doing something along these lines to generate texts using the words you’re learning. I plan to experiment with this (no firm promises though).

[Marius]

Honestly, I think that some sort of way to rank texts based on one’s vocabulary knowledge would be a much more useful feature than even this. With the ability to easily import content to Readlang from the web, I can save a lot of stuff, hundreds of articles, blog posts, stories, etc. So, what should I read? In what order? I’d much rather read the content that interests me than some random texts. A GenAI text is just a shel with no actual meaning. Why read GenAI if I can read real content with real meaning? Answer: to practice. Further question, why not practice with real text with real meaning that also engages me and draws me in? If I read real text with real meaning that interests me, then I’m reading. I’m not practicing, I’m not doing flashcards, I’m not reading some meaningless fake text. I’m immersed in the act of reading, which I’m enjoying. And I could do this 14 hours per day. I couldn’t do flashcards or read GenAI text 14 hours per day. To me this is the big problem that nobody has solved yet. This would be truly revolutionary.

[Marius]

This would be analogous to grader readers. But it would be customized and personalized based on my interests. It would “grade” or rank the texts in my library.

Graded reader

There are plenty of graded readers in English, but not as many in other languages. And if there are, they aren’t available electronically. And if they are, they aren’t as interseting to me, as the content I’m interested in.

[Marius]

Another idea is to actually add some meanint to it. So how about generating summaries of the texts in my library. As opposed to just generate a random text, why not go to an article I saved and generate a “rewrite”. Ask the AI to rewrite the article using my vocabulary level and the words that are on my list. This would significantly improve the attractiveness of this tool.

I already use summarization to summarize the news. I use Reader Readwise as an RSS feed aggregators. I get 300-500 headlines from various newspapers. Readwise generates an automatic summary. I skim the headline. If I’m interested, I read the summary. If I’m still interested I read the actual text. For me, this technology has been transformative.

I could ask any GenAI model to take a text and summarize it in the target language using easier text. In fact, I could write a prompt in Readwise asking it to summarize the Spanish content I read in easier language. But the big problem is that it’s not customized. It doesn’t take into account what I already know. And it is not graded. It doesn’t increase the difficulty level.

Ideally I would save down stuff from the web, then go to my library and generate a graded re-write. And the AI would rewrite the text based on the spaced repetition algorithm and based on my current level. It would have to track the vocabulary I already know and I acquired. But essentially, every day the text would get a bit more difficult, it would add new words. And, very important, it wouldn’t appear more difficult to me. Because the AI can find the optimal difficulty level, It would find the level that’s just right. And because I would be getting better every day, I wouldn’t feel like the texts are getting harder every day.

Maybe one input would be to be able to nudge it, to make the text a bit more difficult or easier.

Another input could be to be ale to take a text in any language and rewrite it in any other language. For example, most of my reading is in English. But I could read some of the content in Spanish, which is the language I’m learning.

It would be like having my own personalized language tuttor writing texts every day especially for me.

Honestly, I think that this is the future of language learning.

The only improvement to this method would be to add a realistic native voice, so I would listen and read at the same time to internalized the sounds of a language.

I trully believe that this would significantly speed up the language learning process significantly. The main reason is that it would not seem like work or learning. It’s just reading. I’m reading things I’m interested in and in the process I acquire a new language. Effortlessly.

I only have a small amount of time in a day I can dedicate to language larning. But if I could convert some of the stuff I have to read to the language I’m learning, then I would read more in the garget language.

[WarsawWill_RandomIde]

Hi Steve,

I have to say I’m really enjoying the original stories feature, and I think it has great potential.

I’ve been using an AI generated podcast called Fluent Fiction, mainly because it is available in many languages, like Romanian and Catalan, that are not served so well elsewhere, and because it has new episodes every day.

But your Readlang Story Bot has much more potential. For a start we have a choice of level, which makes it much more likely that we can get something that fulfulls the conditions of “comprehensive input”, like the graded readers Marius mentioned. Secondly we can choose subjects that are interesting to us. And by giving it enough additional prompts, I’m getting almost exactly what I want in terms of both vocabulary and structures, while getting enough new vocabulary to be useful. And in fact, perhaps for these reasons, I’m finding the stories I’m getting from Readlang’s story bot more useful than those of Fluent Fiction.

Taking a feature from listening to podcasts, I’m now using the Readlang story as a listening exercise before I read it. As soon as the text is generated, I copy the whole thing from the Edit page, and paste it into Google Translate, and just listen (also recording it to an mp3 with Audacity) without looking. Only after that do I read the text in Readlang, with its look up and explanations features.

I haven’t tried giving extra prompts to the other categories, but I think that the general approach I’ve described could be really interesting. And in the Original stories category, you have total freedom to chose the context, whether you want to learn how to order a meal in a restaurant, or go on a mystery tour of Machu Picchu. It just needs a little thought.

[George_Stroud]

Hi Steve,

Just a little bit of feedback - I’ve just tried generating some Arabic stories and it’s included the English version in the body of the text. I think it generates in English and then maybe translates into Arabic?

Thanks

[Steve]

@George_Stroud

Just a little bit of feedback - I’ve just tried generating some Arabic stories and it’s included the English version in the body of the text. I think it generates in English and then maybe translates into Arabic?

Thanks for reporting! The AI is a bit unpredictable so it’s not that surprising that this might happen occasionally. If you notice it’s happening regularly please let me know and I’ll investigate!

[WarsawWill_RandomIde]

Just to say that you can generate stories from vocabulary lists. I already do this directly on ChatGPT, so I was interested to see if it would work on Readlang’s bot, and it does. My prompt is "Please write a short story in (target language) including as many of the following words as possible. The form of the word may be changed - " + list of words. I’ve tried a comma-separated list, and copying from a column, and both seem to work. Generally, on ChatGPT at least, it manages to use the whole list. At the moment I’m trying it from vocab lists I’ve completed for Duolingo lessens. Here are some examples (Beginner selected):

Stories based on Duolingo Dutch units

Here’s one from my last 25 French words from Readlang, using Readlang’s own export words function, which takes seconds (not all used, but the vast majority are there)

tendre
couler
relever
raccourci
rapprocher
cric
bâche
insensées
prouesse
désaffectée
goulot
s’allégea
impérissables
chaudrons
chouannerie
métissages
haubans
épurées
élancées
désenclaver
toussota
remémorèrent
s’embourba
coffrage

Les chaudrons des souvenirs (Advanced)

And here’s one based on the following Spanish words (almost all used in one form or another):

granero,
a raya,
tirón,
advertir,
vendados,
picor,
chanchullo,
axila,
galimatías,
pueden irse al traste,
volvió a las andadas,
no cabía en sí de felicidad,
presumió,
sobaco,
rebosaba,
esculpidos,
alentadores,
desenrolló,
purpurina,
entretejido,
trasquiladas,
patinada,
se toparon con,
se ceban
pletórico de amor

El granero de los secretos en San Esteban (Advanced)

Incidentally, although I haven’t tried it with these vocab-based stories yet, I usually finish my prompt with:

“The story should be amusing and include lots of dialogues, using colloquial language and idiomatic expressions.”

And often this results in some interesting expressions.