[Article] Harvard physicist's take on the VMS

[DG97EEB]

... Schwartz's main point: The VMs text has too many statistical properties that have not been documented in any extant specimen of human writing known to be meaningful, even on their own, let alone together.

And that is the big problem with the LLMs: Their "deductions" are not really deductions, but the blending and distillation of the "deductions" that are out there.  If enough of their sources make a false claim (like "Elon Musk is a genius"), they will state that claim as a fact.

No, sorry, it s not true that "the VMS has many statistical properties that have not been documented in any extant specimen of meaningful human writing".  

For one thing, obviously no one has tested every specimen of meaningful human writing.  

But let's be generous and assume that what the lalamo meant was that "the VMS has many statistical properties that are distinct from those of most specimens of meaningful human writing that have been tested by Voynichologists". 

That statement is true enough, but hides the fact that practically all such tests have used samples of European or Semitic languages only.  

And most of those tests even used the wrong type of text, like novels or phlosophical treatises instead of herbals and materia medicas.

And many tests were character-based statistics, like digraph frequencies and per-character entropy, which are primarily properties of the alphabet and spelling, not of the language.

A "better truth" is that, in fact, the VMS text has many statistical properties that are a surprisingly good match to those of certain natural languages.  Just not the European and Semitic ones...

All the best, --stolfi

Would this work?

You are not allowed to view links. Register or Login to view.