The evolution of P/F vs T/K raises important questions

[Mauro]

I did not follow all the discussion on the "puffs" that went on for the last 20 years.  But the conclusion I recall from my own statistics was

• in their prev-char distributions, p/f resemble both k/t and ke/te;
  
• in their next-char distributions, on the other hand, they resemble ke/te much more than k/t
  

I don't know. I get very different results, I may be wrong of course but I'm rather confident my software works as intended. p and f are quite distant from both ke and te. k and t are close together, and so are ke and te. Much of the closeness of k and t is actually a consequence of the closeness of ot and ok. I suspect (by I did not test it) that the closeness of ke and te is  a consequence of the closeness of ot and ok, rather than a property of ke, te by themselves. Apparently the results of npcompl33t point in similar directions from a different (and mathematically much more sophisticated) point of view.

It would be interesting to see why you got different results. Maybe some assumption? The data in my post You are not allowed to view links. Register or Login to view. is specifically for Herbal A (rf1a-n transcription, all incert spaces treated as spaces, all words with a ? removed). I also considered the glyphs sequences 'cph', 'cfh', 'ckh', 'cth' to be separate glyphs, so the instances of p,f,k,t inside those sequences do not appear in my calculations. If you wish I can rerun my tests on the section/with the assumptions you prefer, to compare with yours.

One thing that people should keep in mind when researching this question is that the use of p/f is probably not algorithmic and not invertible.  They are obviously fancy versions of some glyphs or digraphs, that are used on the head line of a parag to flag it as a head line, and occasionally on other lines too.  They generally occur inside words at positions where we expect k/t to occur; but also too often as the first glyph of a parag.  So they may have been used as replacements of other glyphs too (like d).  The decision to replace each glyph or digraph by p/f in head lines probably was made by the Scribe and/or the Author at random, with unequal probabilities for different glyphs; which would explain the puzzling statistics.

I mostly agree with you, even if I don't find it as obvious that p/f are just fancy versions of something else. Surely they might be, even slightly probably (75%? 3:1 odds?) as the prior stands. But I would expect some statistical confirmation that p and f are indeed used in a similar way to XX and YY, just only in some particular, topological circumstance, ie. near the beginning of a page. From my tests I think I can say confidently that XX and YY are not 'ke' and 'te'.