Title: Naive physics, event perception, lexical semantics, and language acquisition
Abstract: How do children learn the language-specific components of their native language? How is language grounded in perception? Knowledge of the meanings of utterances containing unknown words presumably aids children in the process of determining their meanings. A complete account of such a process must ultimately explain how children extract utterance meanings from their non-linguistic context. Algorithms are presented which utilize a cross-situational learning strategy whereby the learner finds a language model which is consistent across several utterances paired with their non-linguistic context. This allows the learner to acquire partial knowledge from ambiguous situations and combine it across situations to infer a unique language model. M scAIMRA learns word-to-meaning and word-to-category mappings from a corpus pairing utterances with sets of expressions representing the potential meanings of those utterances hypothesized by the learner from the non-linguistic context. M scAIMRA's syntactic theory is embodied in a fixed context-free grammar. D scAVRA extends M scAIMRA by replacing the context-free grammar with a parameterized variant of X theory. K scENUNIA incorporates a more comprehensive model of universal grammar supporting movement, adjunction, and empty categories, as well as more extensive parameterization of its X theory component.
I advance three claims about event perception and the process of grounding language in visual perception. Notions of support, contact, and attachment play a central role in defining the meanings of simple spatial motion verbs in a way that delineates prototypical occurrences of events described by those verbs from non-occurrences. Support, contact, and attachment relations between objects are recovered from images by a process of counterfactual simulation. This imagination capacity, while superficially similar in intent to traditional kinematic simulation, is actually based on a drastically different foundation which takes naive physical constraints such as substantiality, continuity, and attachment relations between objects to be primary. This theory of event perception has been implemented in a program called A scBIGAIL which watches a computer-generated animated movie and produces a description of the objects and events which occur in that movie. A scBIGAIL's event perception processes rely on counter-factual simulation to recover changing support, contact, and attachment relations between objects. (Copies available exclusively from MIT Libraries, Rm. 14-0551, Cambridge, MA 02139-4307. Ph. 617-253-5668; Fax 617-253-1690.) (Abstract shortened with permission of school.)
Publication Year: 1992
Publication Date: 1992-01-01
Language: en
Type: book
Access and Citation
Cited By Count: 82
AI Researcher Chatbot
Get quick answers to your questions about the article from our AI researcher chatbot