Quantifying Phonological Feature Co-Occurrence
Abstract
Similar Place Avoidance is a gradient phonological pattern with two components. First, two consonants in a CVC sequence are marked if they both have the same place of articulation (Pozdniakov & Segerer 2007; Mayer, Rohrdantz, Plank, et al. 2010). As a subset of this larger trend, consonants that do have the same place feature are less likely to co-occur if they also share identical subsidiary features, such as [sonorant] or [continuant] (Yip 1989, Padgett 1995). To take an example from English, words like tote and nine (identical coronal segments) are more marked than are ten and night (both coronals but with different values of [sonorant]), which in turn are more marked than pen and might (labial + coronal segments) (Berkley 2000). The traditional means of quantifying the strength of association for Similar Place Avoidance has been a statistic known as O/E, or the observed-over-expected ratio (Pierrehumbert 1993).Despite its pedigree, an in-depth analysis of whether or not O/E is an appropriate measure of association for Similar Place Avoidance data has never been conducted. This is the primary research question to be addressed, and the answer is that O/E is not adequate for comparative studies, particularly of the scope required in typological surveys. Wilson & Obdeyn (2009) do raise some objections to the way that O/E has been applied, noting that some uses of O/E fold in confounding variables. However, their focus is on how best to compare computational models that are dependent on different theoretical understandings of Similar Place Avoidance. This dissertation provides an in-depth analysis of O/E along with its companion statistic phi, suggested for Similar Place Avoidance by Mayer, Rohrdantz, Plank, et al. 2010). The most serious issue with both statistics is how they are dependent on the margins of the contingency table, a known property of phi (Zysno 1997) and demonstrated here for O/E. I propose replacing O/E and phi with another theory-neutral statistic found in the social sciences known as Yule’s Q (Yule 1900, 1912), as it is not margin dependent. Arguments for the superiority of Yule’s Qfor Similar Place Avoidance studiesare provided on theoretical grounds and backed by analyses of both simulated and empirical linguistic data.
Degree
Ph.D.
Advisors
Sundquist, Purdue University.
Subject Area
Linguistics|Statistics|Language
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.