Statistics 101 and the Detection of Linguistic Relationships

Abstract

[Introduction] A recurrent theme in the writings of Alexis Manaster Ramer is the use and abuse of mathematics in linguistic inquiry. A number of his articles deal specifically with the use and abuse of statistics and probability in historical linguistics, and it is through his work in this area that I came to develop an interest in historical linguistics. This paper discusses some attempts to apply statistical methods and arguments to the problem of detecting linguistic relationships - determining whether two or more languages share a common ancestor. Of course, this particular problem does not exhaust the field of historical linguistics, but I will say nothing about the application of statistics and probability to such further problems as the reconstruction of extinct languages, the processes of language change, and glottochronology. I will examine three statistical tests for linguistic relationship- or more precisely, two tests and one argument- that have been proposed in the literature by Robert Oswalt (1970), Donald A. Ringe, Jr. (1992), and Joseph Greenberg and Merritt Ruhlen (1992). Many reactions to these sorts of proposals amount to little more than blanket dismissal of statistical methods or, at the opposite extreme, uncritical kowtowing to the authority of numbers. This is due, in part, to ignorance of probability and statistics on the part of many historical linguists. I will argue, however, that even a very basic understanding of some fundamental statistical concepts suffices for a critical examination of these proposals on their own terms. To this end, I will present an elementary example of statistical hypothesis testing, and discuss a number of issues that arise. I will then show how similar issues arise for the three tests of linguistic relationship.

Additional details