Floating gallows

[obelus]

Taking the hypothesis that gallows aren't an inherent part of words, but are only added to words, I tried comparing the statistics for all words containing [k] and [t] against the statistics for the same words without them (in paragraphic text only, leaving out labels, etc.).

Yes, gallows can move a lot, with strong preferences for the beginning, next to vowels (counting [c] and [h] as vowels, they function just like [e]). Insertion near or at the end is less frequent.

This way of thinking about gallows placement draws attention to glyph sequences that may be precipitated by the presence of gallows, but are improbable otherwise.  For example:  the rare bigram ye occurs 18 times in the source text, but 321 times "split" by a gallows glyph as y[gallows]e.  The odds are strongly against randomly floating gallows having split almost all of these instances by chance.  If we strip the text of all gallows glyphs, then salt them back in at random positions, only ~23 instances y[gallows]e occur. Random placement is of course not promoted by anyone here as an underlying rule;  as a basis for calculation, however, it can highlight possible laws of gallows-attraction.

If all you have is a hammer... then everything looks like a histogram of character-character correlation frequencies.  The occurrence of trigrams containing a central gallows glyph is summarized below.  Characters preceding a generic gallows glyph are in rows, characters following it are in columns, in descending order of frequency.  Each matrix element represents the count of corresponding trigram pattern matches in IT paragraph text, preceding[gallows]following, divided by the count of trigrams when gallows positions have been globally randomized.  The numerical scale thus gives the factor by which observed frequency deviates from the prediction of a random-placement model:

   

• The range of values is set by the improbably frequent pattern y[gallows]e.
  
• Another conspicuous pattern with elevated frequency is o[gallows]EVA-vowel.
  
• The frequency of l preceding a gallows glyph is elevated, as is the frequency of S (as Sh) following a gallows glyph.
  
• As noted by @nablator, word-initial gallows are over-represented (entries in the top row -- gallows following a space -- are mostly greater than unity).
  
• As noted by @nablator, word-final gallows are under-represented (entries in the left-most column -- gallows preceding a space -- are mostly less than unity)...
  
• ...but, since the color contrast between 0 and 1 is low, suppressed-frequency deviations are not visible in this graphic.  The matrix element with the lowest value is o[gallows]r, because the plentiful bigram or is split by gallows just once, in okraiin (the only attested example of a gallows glyph directly followed by r).  ~278 instances of o[gallows]r would be expected by chance, giving us a candidate law of gallows-repulsion.
  
• Shown here are fractional deviations from the random model, not absolute counts.  The latter are dominated by o[gallows]e, o[gallows]a, and c[gallows]h as bench.