I'm not sure if it would be helpful for your project, but you might be interested in the concept of the grammar of graphics (https://towardsdatascience.com/a-comprehensive-guide-to-the-grammar-of-graphics-for-effective-visualization-of-multi-dimensional-1f92b4ed4149).
I use a version of this in R as implemented in ggplot2. What is nice about it is that you can use standard text descriptions to build an arbitrary graph.
I think this is why LLMs work, and some research backs this up. Humans actually don't create new phrases for unique situations very frequently. Much or even most of what we say is existing word chunks stuck together.
For example, look at this sentence. It communicated what I intended, but it is just a small idea conveyed with a standard text requiring no thought to generate. It could have easily been "Peregrine, with self reflection," or something. Memes are a more obvious example of this.
At some point in that imaginary culture maybe they just abandoned the original language since they could adequately communicate using only shared story references. Whether that part is realistic for an advanced technology culture maybe requires suspension of disbelief.
For a more sophisticated take on this, there is a similar story inside of the Citadel of the Autarch by Gene Wolfe that asks some interesting questions. In that case, the language is specifically limited to ideologically-approved tracts in order to limit what the populace can think about, so as to be easier to control. However, the story told might be subversive.