Quantifying Phonological Feature Co-occurrence

This study argues that the observed-over-expected ratio, or O/E, is an inadequate metric for measuring the strength of consonant co-occurrence in Similar Place Avoidance. I advocate for the use of Yule's Q, an odds ratio based statistic that is not influenced by the relative proportions of labials, coronals, and dorsals in the dataset. This position is advanced on general statistical and linguistic considerations as well as through the analysis of empirical data from 32 languages. 

Parallel typological analyses are conducted using O/E and Yule's Q. Cross-linguistic comparisons using O/E suggest that CVC sequences with two coronals are the least marked of the homorganic pairs. The same analysis using Yule's Q suggests any place of articulation may be the least/most marked in a given language; there are no cross-linguistic preferences. The disagreement between the two statistics can be accounted for by the fact that O/E is sensitive to the margin totals: coronals only appear to pattern separately from the labials and dorsals in the O/E analysis because they are considerably more frequent than are labial and dorsal segments.

To advance the use of Yule's Q in the study of Similar Place Avoidance, the paper provides guidance on constructing confidence intervals, measuring/interpreting effect size, and appropriate use of significance testing. Two case studies on aspects of Similar Place Avoidance in Latin and Medieval Castilian illustrate the proposed methodology.