MTS is the department’s Mind, Technology, and Society speaker series. It is hosted by a different faculty member each semester. Founded by a generous gift from Professors Robert Glushko and Pamela Samuelson, MTS brings researchers and industry professionals from across the globe to present a variety of interdisciplinary work in cognitive science. See our UCMerced CogSci youtube channel for videos of past MTS talks!
CIS graduate students, faculty, and staff, and all who are interested are invited! Members of other departments at UC Merced as well as the general public are encouraged to attend. (Note: current CIS Ph.D. students are required to attend MTS each semester in residence, to fulfill their COGS 250 course requirement).
Dr. Gašper Beguš talk "New ways of modeling language: informative imitation" will be from 2-3:30pm in SSM 104.
Abstract: Large language models have recently captured the attention of the linguistic community, but their architecture features several cognitively and linguistically implausible aspects. In this talk, I argue for building more realistic models that learn language from raw speech, incorporating articulators, the production-perception loop, and communicative intent. I propose the ciwaGAN model, which models language as informative imitation and features several desired properties such as communicative intent, learning from raw unsupervised audio, and embodied representations. I also present an interpretability technique, which allows us to perform linguistic experiments on the models. This same technique also helped us discover analogs of human vowels in whale communication. Modeling language in terms of informative imitation with realistic deep learning models can not only provide insights into human language and cognition but also help us create a framework of language studies that incorporates humans, animals, and machines.
Bio: Gašper Beguš is an Associate Professor at the University of California, Berkeley, where he bridges the fields of linguistics, artificial intelligence, and cognitive science. His research centers on interpretable AI and the study of language in humans, animals, and machines. Dr. Beguš leads efforts to develop techniques that help us better understand the inner workings of AI. In his Berkeley Biological and Artificial Language lab, he also builds the most realistic models of human language learning—creating an artificial baby language learner. Serving as the Linguistics Lead at Project CETI, a non-profit dedicated to listening to sperm whales, he researches one of the most intriguing communication systems in the animal kingdom. His recent research discovered that sperm whales have equivalents to human vowels. His research has been featured in numerous press outlets, including the Financial Times, The Atlantic, National Geographic, Quanta Magazine, Harvard Magazine, Noema Magazine, and others.
For more information or to sign up for email announcements, please contact the talk series organizer: cis-mts-lead@lists.ucmerced.edu.
