Event Details

March 12, 2026

Join Zoom Webinar

Passcode: 315888

Webinar ID:  969 8480 6154

Abstract

Self-supervised speech models (S3Ms) can be trained to efficiently recognize spoken words in naturalistic, noisy environments. However, we do not understand the types of linguistic representations these models use to accomplish this task. To address this question, we study how S3M variants optimized for word recognition represent phonological and morphological phenomena, comparing frequent English noun and verb inflections in -s to frequent adjectival and verbal inflections in -er. We find that their representations exhibit a global linear geometry which can be used to link English bases and regular inflected forms.

This geometric structure does not directly track phonological or morphological units. Instead, it tracks the regular distributional relationships linking many word pairs in the English lexicon—often, but not always, due to morphological inflection. These findings point to candidate representational strategies that may support human spoken word recognition, challenging the presumed necessity of distinct linguistic representations of phonology and morphology, and highlighting the importance of functional pressures in shaping representations.

Host: Eric Boxer

POC: Justina Gilleland

Speaker Bio

Canaan Breiss is an Assistant Professor at the University of Southern California. His research is in theoretical, computational, and experimental phonology, with particular interest in learning/acquisition, representation of overlapping and interacting phonological processes, and phonology’s interfaces with (morpho)syntax and the lexicon. He uses a diverse methodological toolkit, including computational and statistical modeling, corpus methods, and behavioral experiments.