Statistical testing of the functional load hypothesis and implications for phonological theory

The idea that phonological neutralization is constrained by the communicative function of lexical contrast has a long history (e.g., Trubetzkoy 1939, Martinet 1955, Hockett 1967). A few case-studies have asked whether occurring neutralizations have a lower impact on lexical contrast than non-occurring ones, either by creating fewer homophones (e.g., Silverman 2010, Kaplan in press) or by incurring a relatively small drop in lexical-level entropy (Surendran and Niyogi 2006). Results have been mixed: some case-studies are consistent with the functional load account (Silverman 2010, Kaplan in press) while others contradict it (e.g., King 1967) or are judged inconclusive (e.g. Surendran & Niyogi 2006). However, if functional load is just one of many influences on sound change we would expect to find many individual ‘counterexamples’ to the functional load hypothesis. Consequently, we need a set of case-studies that can be assessed jointly using statistical methods. As part of an ongoing investigation of relationships between functional load and contrast neutralization, we are building a database of diachronically recent mergers in languages for which there exist frequency-coded phonemic word lists. The database currently includes data related to forty distinct mergers occurring in German, French, Korean, and several varieties of English. This data reveals a statistically significant difference between the distributions of these mergers versus comparable non-occurring mergers, in three measures of the functional load for phonemic oppositions: changes in homophony, lexical entropy and segmental entropy.