Methods for minimizing the confounding effects of word length in the analysis of phonotactic probability and neighborhood density

Abstract

Recent research suggests that phonotactic probability (the likelihood of occurrence of a sound sequence) and neighborhood density (the number of words phonologically similar to a given word) influence spoken language processing and acquisition across the lifespan in both normal and clinical populations. The majority of research in this area has tended to focus on controlled laboratory studies rather than naturalistic data such as spontaneous speech samples or elicited probes. One difficulty in applying current measures of phonotactic probability and neighborhood density to more naturalistic samples is the significant correlation between these variables and word length. This study examines several alternative transformations of phonotactic probability and neighborhood density as a means of reducing or eliminating this correlation with word length. Computational analyses of the words in a large database and reanalysis of archival data supported the use of z scores for the analysis of phonotactic probability as a continuous variable and the use of median transformation scores for the analysis of phonotactic probability as a dichotomous variable. Neighborhood density results were less clear with the conclusion that analysis of neighborhood density as a continuous variable warrants further investigation to differentiate the utility of z scores in comparison to median transformation scores. Furthermore, balanced dichotomous coding of neighborhood density was difficult to achieve, suggesting that analysis of neighborhood density as a dichotomous variable should be approached with caution. Recommendations for future application and analyses are discussed.