Sentando Las Bases De La Adquisición De La Gramatica En Representaciones Sensorio Motrices

Grounding the Acquisition of Grammar in Sensorimotor Representations
Tiago V. Maia and Nancy C. Chang
{maia, nchang}@icsi.berkeley.edu International Computer Science Institute 1947 Center St., Suite 600, Berkeley, CA 94704*

Abstract
Drawing on data from linguistics, developmental psychology and the neurosciences, we present a computational theory of the acquisition of early grammar byinfants. Based on the view that language is a mapping between form and meaning, we propose that a theory of language acquisition must be tightly integrated with a theory of the infant’s prelinguistic representations. Namely, the infant’s task is to learn how to map the linguistic form in the input to her representations of the corresponding scenes. We have developed a theory of prelinguistic cognitionbased on i) what is currently known about the architecture of the brain, and ii) the representational requirements for successful (sensorimotor) behavior in the world. We show how such prelinguistic sensorimotor representations can provide the basis for the acquisition of early grammatical forms, and thereby ground language in the world. Importantly, this is true not only at the lexical level, butalso at the grammatical level.

Introduction
Approaches to language in the Cognitive Linguistics tradition take linguistic units to be pairings between form and meaning (e.g. Langacker 1991). Typically, elements of form encompass properties of the speech or text stream, such as the identity or ordering of particular segments, while the semantic pole is equated with (non-linguistic) conceptualstructure, which is taken to be embodied (Lakoff 1987). That is, the meaning of language is grounded in human sensorimotor systems and, loosely speaking, corresponds to the activation of certain neural structures that represent, in a non-linguistic way, the scene or content associated with the linguistic form (e.g. Regier 1996, Bailey 1997). In this view, the acquisition of language amounts tolearning how elements of form map to elements of meaning. This implies that, among other things, one needs a solid theoretical understanding of the prelinguistic structures to which linguistic form will map. We assume that these prelinguistic structures exist because they are useful for behaving in the world, not merely because they pave the way for language. We base this assumption on anevolutionary line of reasoning: language entered the scene very late in evolutionary time, and as such was built on top of older cognitive skills that
The first author is also with the Dept. of Computer Science and Engineering, SUNY Buffalo, and the second author with the Electrical Engineering and Computer Sciences Dept., UC Berkeley.
*

we share with many of our relatives in the animal kingdom, such asthe ability to move, perceive scenes, act on objects, interact with conspecifics, etc. To some extent, ontogeny seems to proceed in the same manner – before acquiring language, the infant goes through an extended period of physical and social interaction with the environment. Piaget (1952) termed this period sensorimotor, arguing that it was characterized exclusively by motor and perceptualinteractions with the world, without any kind of representations or conceptual thought. This view, however, has been the subject of intense debate in contemporary developmental psychology (e.g. Mandler 1992). In AI, the grounding of single words (or, more generally, symbols) in perceptual and, to a lesser extent, motor systems has been a very active area of research over the last decade (e.g. Harnad1990, Regier 1996, Bailey 1997, Bailey et al. 1998, Steels and Kaplan 1999). It seems generally accepted that understanding the relation between single words and their underlying sensorimotor representations is fundamental for any model of language acquisition. Importantly, a growing body of evidence from the neurosciences shows that it might also be extremely relevant for language understanding....