Similarity-based Phonological Generalization

Halle (1978) suggested that abstract, partial feature-based representations could explain the generalization seen in phonology. An alternative explanation for this is that the similarity of different sounds causes speakers to treat them similarly (Cristia et al. 2013). To test which explanation better predicts the experimental results of Cristia et al. (2013), I created a MaxEnt learner that uses similarity in its learning update to encourage generalization to similar segments. My learner\u27s predictions match the results found by Cristia et al. (2013) more accurately than a previously proposed MaxEnt learner (Moreton et al. 2017) that relies on generalization caused by abstract representations