Dept of Psychology, University of Pennsylvania
Word learning in honor of Lila Gleitman: Perception of structure from word and world
Abstract: It is tempting to conclude that children learn the meanings of words by observing their circumstances of use (e.g., observing that the word “dog” often co-occurs with dog-sightings). If this is the case though, how do children ever learn the vast majority of the words that they know? Consider most of the words in this abstract, many of which a 3-year-old produces and understands: like “what”, “not”, “language”, “do”, “think”, “learn.” Can these words be learned by observation of their circumstances of use? There are no what-sightings that go with “what”, and no not-sightings that go with “not”; thinking-sightings often look like sleeping-sightings and sitting-sightings. How do children go about learning these “hard words” despite no explicit instruction? I will present research, some of which was done with my longtime collaborator Lila Gleitman, that is designed to answer these questions. I’ll focus on the unexpected role that word-to-world pairings nevertheless play in the learning of hard words. I’ll propose a framework for word-to-world mapping in which perception of the referent world itself offers us significant structure, and the syntactic structure we gather from the language is connected to these representations. This connection, and the structural representations on both sides of the word-to-world coin, allow us to see what we shouldn’t be able to see, and hear what we shouldn’t be able to hear. I’ll offer experimental evidence that our perception of the world includes rapid extraction of event structure, and hypothesize that this allows access to abstract relational meaning even in young children. These representations play an important role in understanding how situational contexts permit children to learn even the most abstract of terms, such as symmetrical predicates (e.g., the meaning of “equal”) and truth-functional negation (e.g., the meaning of “not”).