Check out this article in ScienceNews summarizing commentaries on two recent language studies in Science (Atkinson, 2011: ) and Nature (Dunn et al., 2011). Each of the studies has received a lot of attention and they are the subject of two special issues in press for Linguistic Typology, to which HLP Lab contributed on three articles. I will add a link to the special issue(s) once it comes out.
Atkinson (2011) proposed the serial founder hypothesis, according to which languages further away from the point of origin of language have simpler phonology. He presents evidence based on a data set constructed from the World Atlas of Languages. The study has been criticized for, among other things, the choice of data, the way phonological complexity was calculated, and the statistical methods employed in the approach. The paper that I co-authored with Peter Graff, Bill Croft, and Dan Pontillo assesses the mixed effect regression approach taken by Atkinson to account for genetic relations between languages. Bill Croft, of course, is at the University of New Mexico, Peter Graff is a graduate student at MIT, and Dan just joined Rochester’s graduate program. We provide an introduction to mixed effect regression, discuss when one can conclude that random slopes aren’t warranted, and extend the approach to account for language contact. Beyond the specific evaluation of Atkinson’s approach, we also hope that this paper will be of interest to anyone conducting data analysis of typological data. You might also be interested in the comment by Cysouw et al that just got accepted by Science. They discuss to what extent Atkinson’s findings replicate on another data set. As soon as the article is officially in press, I will post a link here. Another comment from our lab (work with Dan Pontillo and Peter Graff) under review for Science presents large scale statistical simulations that assess the Type I error rate of Atkinson’s approach.
Dunn et al (2011) present a novel statistical approach to assess whether there is any typological evidence for word order universals (well, novel for typological research; their approach has been employed to evolutionary biology). They argue that there is only evidence for lineage-specific trends, rather than cross-lineage universals. In one comment to appear in Linguistic Typology, Hal Tily and I have discussed alternative evidence from behavioral paradigms (artificial language learning and iterated artificial language learning) that does seem to provide evidence for cross-lineage universals (although I think the main message of our comment is that there are other methods that should be pursued in addition to statistical analyses of typological data, which suffer a lot from data sparseness). Hal is currently a post-doctoral fellow at MIT. In another comment, Bill Croft, Tanmoy Bhattacharya, Dave Kleinschmidt, Eric Smith and I discuss the trade-offs of the statistical methods employed by Dunn et al (a trait evolution model for language change, implemented in the software BayesTraits). Tanmoy and Eric are at the Santa Fe Institute, Bill is at the University of New Mexico.