Analysis of patterns at the very start and end of lines, per line type

[quimqu]

One could test that theory by separating all lines into subsets based on their first glyph, computing the word frequency distribution of each subset (excluding the first word of each line) and comparing those distributions.

I did a fast check. The cosine similarity plot suggests that the first character of a line doesn’t create a separate "language" within the manuscript. It may nudge the line in a certain direction, but overall the vocabulary quickly falls back into the same general pattern as the rest of the text.