Being a logic student who uses probabilistic machine learning, Farmer's use of the words refutation, proof etc. make me despair. This web page is an example:
" As clearly stated in the last sentence of the paper, our results provide evidence which, given the rich syntactic structure in the script (and other evidence as listed below), increases the probability that the script represents language."
It seems to me that at least 50% of this controversy is a culture clash between computational linguists and historical linguists.
It's a little odd to describe the Rao paper as using time-tested logic and math when the idea of conditional entropy as an indicator of whether a given symbol system is linguistic is in no way time-tested. I'm not a professional linguist, and it may well be a valid and interesting technique, but it is not time-tested.
How do you time-test models with a strictly limited and static set of test cases? Any scheme would suffer from over-training. Maybe we have to evaluate approaches using logic and math e.g. conditional entropy to attack problems spaces like this.
It's a little odd to describe the Rao paper as using time-tested logic and math when the idea of conditional entropy as an indicator of whether a given symbol system is linguistic is in no way time-tested. I'm not a professional linguist, and it may well be a valid and interesting technique, but it is not time-tested.