At this week’s lab meeting (the last meeting of the semester), Amanda Doucette, will be presenting their paper Inherent Biases of Recurrent Neural Networks for Phonological Assimilation and Dissimilation.
- Wednesday, December 9, 13:30–14:30 (Montreal time, UTC-4).
- Meetings are via Zoom. If you would like to attend the meeting but have not yet registered for this semester’s MCQLL meetings, please do so at this link. If you have already registered, please join using the link you received in your confirmation email.
When learning the phonological rules of a language, human learners display a simplicity bias. For example, patterns involving only one phonological feature are learned faster than patterns involving two or more. Previous computational models of phonological learning have required additional representation of repeated features to capture this learning bias. I propose two simple recurrent neural network models that are capable of capturing the simplicity bias without employing any bias-specific feature representations.
Amanda Doucette is a first-year PhD student from the Department of Linguistics. They are interested in computational models of language learning, and how language is represented in the brain.