The Grounding of Word Meaning:
Data and Models
Papers from the AAAI Workshop
Michael Gasser, Chair
Language begins with the acquisition of word meaning. A normal child masters tens of thousands of words over the course of the ensuing years. Understanding how this happens is fundamental to understanding how language works as well as to the development of viable computational approaches to language acquisition.
Early research on the development of word meaning focused on constraints that were internal to the linguistic system or to relationships between language and symbolic thought. More recently, however, there have been attempts to "ground" or "embody" meaning in general cognitive or perceptual processes or in the body of the organism interacting with its environment. The advent of connectionism has opened up the possibility of models that relate analog systems to discrete categories, thus of explicit models of perception and motor control. Recently some connectionists have begun to model the acquisition of word meaning, usually focusing on the way in which meaning arises out of a system perceiving the world. This work is consistent with the growing grounding or embodiment view among empirical researchers. However, psychologists, linguists, anthropologists, and biologists may be unaware of what neural networks, especially those with built-in modularity, are capable of, and AI researchers could benefit from greater familiarity with the data. The goal of this workshop was to bring together these different communities.