Computational neuroscientist Idan Blank to speak to joint LingCircle/抖阴传媒在线 NLP Seminar
Wednesday, February 4 | 11:15 a.m.鈥1:00 p.m. | Hellems N380 & Zoom
The CU 抖阴传媒在线 Linguistics community welcomes Idan A. Blank, Assistant Professor of Psychology and Linguistics at UCLA for a timely talk examining whether large language models (LLMs) truly understand the language they process.
Blank鈥檚 talk presents three studies that adapt experimental methods from human psycholinguistics and neurolinguistics to probe signatures of human-like comprehension in LLMs. First, he asks whether semantic information can influence syntactic processing in these models, as it does in humans, or whether certain stages of syntactic processing remain insulated from meaning. Second, he examines whether LLMs encode a foundational distinction in linguistic meaning鈥攖he difference between agents and patients. Third, he explores whether large vision鈥搇anguage models use visual context in pragmatically sensitive ways, distinguishing felicitous, under-informative, and over-informative referring expressions.
Together, these studies reveal both convergences and divergences between human and machine language processing. By decomposing 鈥渦nderstanding鈥 into theoretically grounded components, Blank offers a nuanced perspective on what LLMs do鈥攁nd do not鈥攕hare with human comprehension.
Dr. Blank leads the BlankLangLab at UCLA, where his research investigates how meaning is represented and integrated across biological and artificial systems. Combining functional neuroimaging, behavioral experiments, and computational modeling, his work explores how different information sources contribute to comprehension and how cognitive labor is distributed across systems. He earned his PhD in Cognitive Science from MIT and completed postdoctoral training at the McGovern Institute for Brain Research.
Zoom link:
All campus community members are welcome to attend, in person or remotely via Zoom.