About the generation of similar words

[quimqu]

Anyway, this question is: given that daiin is the most frequent word in the overall text, is the reason that it appears most because:
- this is just the most frequent word so it appears most frequenty
- this is the word that most frequently results from a small (or zero) change from the recent words.
And then of course the question if or how we can possibly detect this.

Yes, in a strict sense, almost everything could be derived from "daiin" if you allow multiple small steps, because the whole system forms a single connected network. But that is not specific to "daiin". In fact, it is true for any word inside that network. From any node, you can reach any other through chains of small edits.

So the important question is not whether everything can be derived from “daiin”, but whether the system is actually centered around it. And the graph suggests it is not.

"daiin" is not especially central in the network. It does not directly connect to a large portion of the vocabulary (only 5%), and there are other words that are more central in terms of connectivity and average distance.

So while everything can be connected to "daiin", the same is true for any other word. The structure looks more like a dense web than a system expanding from a single root.

This makes it more likely that "daiin" is simply very frequent, rather than the main generative source.

The background behind my thoughts is related to a question (doubt) I have about the autocopy method.

I’m actually not very convinced by the autocopy idea either.

From what I’m seeing in the CUVA data, almost all the vocabulary (around 94% of tokens) is inside the Levenshtein ≤2 network. That suggests a very dense and coherent system of variation, and not in a form of tree (there is no source, at least easily seen).

To me, that looks less like words being generated locally during writing by copying and slightly modifying the previous one, and more like a system where the word forms already belong to a sort of dictionary, and the writer is selecting from there (codebook?). In other words, it feels more like a predefined generative system than an on-the-fly autocopy process.

Also, the small set of words outside the main network are almost all hapax, and many of them can be explained as combinations of existing tokens from inside the network. That could point to cases where the author is no longer following the usual formation patterns and instead combines known pieces. (Oh, I don't have this word in the dictionary now! Let's join some words and work further)

So my current intuition is that the structure we see is more consistent with a prior generated scheme and controlled variation, rather than simple local copying from one word to the next.