Proceedings

To agree or not to agree? Evaluating measures of similarity in consonant harmony

---

Abstract

In languages with consonant harmony, certain consonants are required to agree by the value of some feature (Gafos 1996; Hansson 2010; Mackenzie 2009, 2011, 2016; Rose and Walker 2004). It has been argued that the consonants which are required to agree are above a particular threshold of similarity (Rose and Walker 2004). Similarity metrics have previously been proposed to capture other phonological patterns, such as OCP-Place patterns (Frisch et al. 2004; Pierrehumbert 1993) and more general co-occurrence restrictions (Doucette et al. 2024). A number of studies have attempted to apply such similarity measures to consonant harmony patterns (Arsenault 2012; Mackenzie 2009, 2011, 2016). However, these metrics have yet to be applied predictively to determine where consonant harmony arises in the inventory of a given language. In this paper, we establish a baseline for this issue by evaluating a few possibilities developed in the literature. While the pairs that participate in harmony are not universally more similar than all non-participating pairs, participants in harmony have a higher probability of being similar than comparable pairs of non-participating consonants. No single similarity measure universally predicts participation in consonant harmony; the best model of similarity is dependent on the specific language. Assuming that similarity predicts which consonants are required to agree in all languages with consonant harmony, then we still do not have the right definition of similarity.

Keywords: consonant harmony, phonology, similarity, self-supervised speech models

How to Cite:

Quevedo, L. & Mooney, K., (2026) “To agree or not to agree? Evaluating measures of similarity in consonant harmony”, Proceedings of the Annual Meetings on Phonology 2(1). doi: https://doi.org/10.7275/amphonology.3694