18-02-2026, 06:45 PM
(This post was last modified: 18-02-2026, 06:46 PM by Jorge_Stolfi.)
(18-02-2026, 02:58 PM)nablator wrote: You are not allowed to view links. Register or Login to view.With enough replacements you can transform the distribution of anything into anything and claim that Herbal-A and Herbal-B are not really different languages, but it doesn't prove anything and it doesn't explain why so many transformation are needed
That is not true. My description is way more complicated than it should be, but the transformation is merely mapping all gallows to f, all "dealers" d l r s to r, and deleting all other glyphs.
There is no reason to expect that this transformation would make the word frequencies of two different languages coincide.
In particular, the second plot suggests that the "translation" of language A to language B may replace some gallows by other gallows, but cannot create or delete them. And it cannot delete or create any dealers, nor double or collapse them, nor move dealers from one side of a gallows to the other side.
Okay, all these things could happen, but they would have to be balanced. If some dealers move from before the gallows to after the gallows, an equal number (counting tokens nor words) would have to move in the other direction.
Said another way, if you divide the lexicon into subsets that have the same "root", the plot suggests that the translation from A to B shuffles the words within each subset, but not between subsets.
All the best, --stolfi