I'm not sure many people are going to buy it.
Only one (physical) table needs to be generated at first, but the values of the transition weights do need to be changed as the scribe proceeds through the manuscript.
Could it be used for English? I doubt it. The morphology of English and other Indo-European languages is too complex to for a reasonably-sized state machine to realistically generate.
Maybe for something more regular such as Pinyin Mandarin or Yale Cantonese BUT the output wouldn't fool anyone who knows those languages, and a linguist who had never seen them before would very quickly become suspicious - no discernable word order, common words repeated several times in a row.
What I'm proposing is that the words we see are the output of a random process, but that their order appears to be random, in the statistical sense. However, everyone who's studied the text agrees that the glyph distribution within words conforms to a well-defined syntax, and there are significant differences between different parts of the text. My theory explains these.
Of course, a well-designed cipher will appear to be random, and the input to the text generation process appears to be just that, but the problem is that there's no way of subsequently decrypting the text, as information (in the mathematical sense, as well as the everyday sense) would be lost.
I'll start a new thread ... (https://news.ycombinator.com/item?id=14091216)