Rather than two statistically distinct authors, the data shows a continuous shift across the manuscript.
The two distinct languages idea goes back to observations by Prescott Currier (1976), who labeled two statistical profiles usually named “Currier A” and “Currier B.” His distinction was relatively clear-cut for the Herbal section but considerably more nuanced elsewhere. He was describing mixed sections and hedged his assignments (see https://www.voynich.nu/extra/curr_main.html). Subsequent researchers simplified this into a sharper division between Curri... (read more)
Nice analysis! I'm Torsten Timm the author of the discussed self-citation hypothesis.
My generator is indeed a simulation based on observations for the Voynich text. The question the generator tries to answer is are the micro observations for the Voynich text enough to explain the high level statistics we can observer for the text. We deliberately did not fine-tune the algorithm to pick an “optimal” sample for this presentation. Such a strategy is by itself questionable. Nevertheless, an exhaustive scan of the parameter space (involving thousands of automa... (read more)
The group around Bowern and Lindemann actually did such a test. They recruited 42 volunteers to write short language like gibberish documents. They come to the conclusion: "Gibberish tends to exhibit lower total information content than meaningful text; higher repetition of words and characters, including triple repeats; greater biases in character placement within lines and word placement within sections; positive autocorrelation of word lengths (i.e., a tendency for words to cluster short-short-longlong rather than short-long-short-long); and a weaker av... (read more)