Proceedings

Phone2Vec

---

Abstract

In this paper, we investigate whether a generic word embedding model can be applied to phonetic segments to discover phonological classes. The data come from English and we consider the question both from the direction of the classes that emerge from the model and from the perspective of the phonology: what classes do we expect to come from the model? Our general prediction is that classes we know are supported by the phonotactics should emerge, but classes that are only evidenced by alternations will not.

Keywords: word2vec, phonotactics, English, embeddings, features, natural classes

How to Cite:

Hammond, M., (2026) “Phone2Vec”, Proceedings of the Annual Meetings on Phonology 2(1). doi: https://doi.org/10.7275/amphonology.3546