Frame-shifting and Conceptual Blending in Large Language Models - Understanding Human Language Processing

Explore a thought-provoking lecture from UC San Diego's Seana Coulson examining the intersection of cognitive semantics and large language models in understanding human language comprehension. Delve into research findings that demonstrate how metrics from large language models correlate with human behavioral and neural responses to language processing. Discover the fascinating contrasts between human joke comprehension and the understanding capabilities of text-trained language models, while examining the role of hierarchically structured background knowledge and conceptual mapping in flexible language processing. Learn about the current limitations and potential implications of language models in replicating human-like understanding through this 59-minute presentation from the Simons Institute's "Unknown Futures of Generalization" series.