Let's create a new language together!

[turbid lintel]

Hey fellow ChatGPT prompt engineers,

I'm seeking collaborators to help me explore new methods for creating a made-up language called Kasumi. While I've already established some basic grammar rules, the project is still ongoing and open to new ideas.

The goal is to create the (perhaps) first fully functioning AI-generated language with its own grammar rules, and not just a copy of an already established language with replaced words.

To develop the basics of Kasumi, I followed a semi-structured approach:
Firstly, I sang a random melody and used absolute nonsense words to establish an intuitive foundation for the language.
Secondly, I utilized ChatGPT to generate similar words and assist in brainstorming potential names for the language.
After selecting "Kasumi" (although still considering "Kasiumaya"), I asked ChatGPT to create a table of contents for a grammar book to provide an overview of the necessary components for the language.
Using ChatGPT's help, I explored various language systems to customize Kasumi into something easy to understand and intuitive.
After an extensive and exhaustive process, I created a set of basic rules that I believe works for the language.

Creating a new language from scratch has proven to be a challenging but exciting process, and I'd love to get some feedback and suggestions from other prompt engineers and language enthusiasts.

You can check out my progress so far in this Google document: [apparently, I cannot send links. Please send me a DM if you are interested in the project].

I am eager to receive feedback and suggestions from other prompt engineers and language enthusiasts to continue developing Kasumi. It would be great to make this into a community project, and I would be happy to share ownership with anyone who make some substantial contributions to the creation of the language.

I'm looking forward to hearing from you and working together to push the boundaries of AI-generated language!

[rough zephyr]

many-to-many language zero-shot translation neural networks supposedly contain an abstract representation of language in the middle layer. In those engines, i think, the input language first gets encoded into the abstract "universal language", and then decoded into the target language.

This abstract universal language is not programmed in, but inferred from the training set.

Kind of, if you would want to develop a useful and effective universal language, then studying this middle layer might be valuable.

[turbid lintel]

Thank you, @rough zephyr That’s very interesting! I’ll read up on it!

[urban falcon]

@turbid lintel I've been trying to attempt a similar task with gpt3.

I'll have to try this.