The expressivity of segmental phonology and the definition of weak determinism

Jardine (2016) asserts a computational distinction between segmental and tonal phonology, arguing that certain tonal patterns require the strictly more expressive computational power of non-deterministic regular functions while segmental patterns require at most the power of weakly deterministic regular functions. We advance two claims bearing on this assertion. Empirically, we show that non-deterministic segmental patterns are in fact attested, focusing on the vowel harmony pattern found in Tutrugbu (McCollum and Essegbey 2018) and citing several others. We submit that these patterns are non-deterministic in exactly the same essential way as the tonal patterns discussed by Jardine (2016). Formally, we show that the definition of weakly deterministic regular functions offered by Heinz and Lai (2013) is incapable of distinguishing between non-deterministic patterns and the less complex weakly deterministic patterns it is intended to delimit. We offer a revised definition of weakly deterministic functions that makes the correct distinctions, subsuming the conditions imposed by Heinz and Lai (2013) under a proposed definition of ‘interaction’ between composed functions. We also conjecture that the inverse relationship between complexity and observed frequency is explainable by domain-general principles of learnability rather than a categorical cap on the complexity of a phonology-specific learning mechanism, contra Heinz and Idsardi (2011, 2013).

This talk is based on the manuscript with the same title available here: