Event Date: Tuesday, 21 May, 2019
Location: 10-12, Sala Riunioni, Pal. Venera
Speaker: Dr. Idan Blank (MIT & UCLA)
Title: When we “know the meaning” of a word, what kind of knowledge do we have?
Abstract: Understanding words appears to require both linguistic knowledge (stored form-meaning pairings and ways to combine them) and world knowledge (object properties, event structure, etc.). In this talk, I will pose two challenges for common distinctions between these knowledge sources. First, I will show that rich information about concrete objects could, in principle, be learned from word co-occurrence statistics in the absence of non-linguistic (e.g., perceptual) information. I will introduce a domain-general approach for leveraging these statistics (as captured by distributional semantic models, DSMs) to recover context-specific human judgments such that, e.g., “dolphin” and “tiger” are similar when considering size but different when considering habitat or aggressiveness. Second, I will demonstrate that “syntactic” knowledge of verb-argument structure (e.g., “eat”, but not “devour”, can appear without an object) can in part be predicted from distributional information (i.e., without explicit access to “syntax”). Nevertheless, only a small fraction of DSM dimensions is required for predicting argument structure – the rest captures semantic properties that are relatively divorced from verb syntax, and human judgments appear to be sensitive to both kinds of information. Together, these studies attempt to push against the upper bound on the potential complexity of distributional word meanings.
Idan A. Blank will join UCLA as an Assistant Professor of Psychology in July 2019. He received his PhD (2016) in Cognitive Science from MIT, working with Nancy Kanwisher and Ev Fedorenko, and continued working with Ev as a postdoctoral associate at the McGovern Institute for Brain Research. He studies the division of linguistic labor across distinct cognitive mechanisms during naturalistic comprehension (using fMRI), as well as the kinds of knowledge captured by distributional semantic models and artificial neural networks (using computational modeling). Previously, he studied mathematics, psychology, and theatre arts in the Lautman Interdisciplinary Program at Tel-Aviv University, where he received his MA (2011) studying visual face processing with Galit Yovel.