17-05-2026, 07:06 PM
(17-05-2026, 06:47 PM)Mauro wrote: You are not allowed to view links. Register or Login to view.So, if I understood: you take a seed text, then you choose a word from it, then you modify it, then you check if it's a 'valid' word by testing if it can be generated by the 'ledger' table. If it cannot, you iterate until you find a 'valid' word'. Is it correct?
If it is, I have an additional question: which algorithm do you use to modify the chosen source word?
You have the basics yes!
It usually starts with an existing word from the local source sheet or from the current page, then tries to produce a small mutation of that word.
The mutation step is simple:
- choose an existing source word
- make a small edit
- test whether the result is legal under the ledger
- reject it if it fails
- try again
The edits are ordinary edit-distance operations:
- substitute one glyph
- insert one glyph
- delete one glyph
So for ED1, it performs one of those edits. For ED2, it performs two edits. The ledger then acts as a validator. It checks whether the resulting word has legal glyph transitions for its position: beginning, middle, and ending.
The generator also has some 'don't look stupid' guardrails that prevent things like runaway vowels or globs of consonants. Basically, what a scribe would look at and say... uh.. no. daiiiiiiin is not a good word. It doesn't specifically look for consonants or vowels when it does the mutation, it just uses the ledger letter weights. Then it applies the don't look stupid rules.

