I’ve been at a workshop on historical linguistics and reconstruction this weekend, at the University of Michigan. The slides for the paper I gave are available here. This conference is always fun and this year’s was no exception. This paper (when I write it up) will supersede the other things I have put up on these methods for Karnic languages. This one demonstrates different NNet networks for stable and borrowable vocabulary and parts of speech. It doesn’t talk about morphological data.
(Before anyone comments, I know that NNet is not a cladistic method; I changed my mind about how much I wanted to talk about and what I wanted to demonstrate between submitting the title and compiling the slides.)
Clarification: I should have added something about this before (at the risk of digging myself further into an embarrassing hole). In the first or second slide I mention my view that some of the linguistic computational work is (in my opinion) ‘bizarre’. I want to make it clear (as I did in the talk, but not on the slides) that that rather ill-chosen epithet only applies to a subset of what’s there. Particularly, it applies to Forster and Toth’s method, to the Australian data in McMahon and McMahon, and to some of the character choices and coding in some of the Ringe, Warnow and Taylor work. Forster and Toth is not at all accepted in linguistics, I’ve written about the Australian data problems in McMahon, and the potential circularity of the character choices in some of the IE data for RWT has come up at workshops. I should have made that clear in the text and I apologise about the implications of that slide.