A Suggestion for a Linguistics with Connectionist Foundations - eScholarship

A Suggestion For A Linguistics With Connectionist Foundations George Lakoff Department of Linguistics University of California Berkeley, CA 94720 Abstract The theory of cognitive linguistics (as outlined in my book Women. Fire. and Dangerous Things and Langacker’s Foundations of Cognitive Grammar) converges with connectionist cognitive science in a variety of ways. This paper gives an overview of what those convergences are and what an overall theory of language with connectionist foundations might look like. 1. Some Larger Issues I would like to situate this lecture with respect to certain larger issues raised by connectionist research. I see connectionist research as being primarily concerned with the question: How is it possible for the brain to be the mind? That is. how is it possible for a physical brain to engage in reason? And how is language represented in the brain? Such questions are indirectly related to the question: How could human intelligence have evolved from the brains of other primates? Does human intelligence involve the use of capacities present in the animals front which we evolved? I will be presenting a number of linguistic results that bear indirectly on these issues, and will add a number of speculations that. if correct. might allow us to begin answering such questions. The linguistic results I will be presenting indicate that htunan reason uses some of the same mechanisms involved in perception and that human reason can be seen as growing out of perceptual and motor mechanisms. Before going on to this, however, it is important to make a distinction between connectionist modelling and the connectionist theory of mind. 2. Activation Patterns Over Portions of Topographic Maps Are Meaningful The various parts of the body are connected to the brain by neural networks. Indeed. they are literally ‘u-nannn:-I’ Anon ol-manta AF Ilnunounn :n Ilia L-min no-IL-ul ‘topographic maps‘. Such mappings preserve topological relations. but not relative size; e.g.. the topographic map of the thumb is next to that of the forefinger, but is much larger than it. The neural networks of the brain are an appropriate locus for a theory of how concepts are embodied for a very simple reason: Patterns of activation over topographic maps in the brain are meaningful, because of the way those portions of the brain are connected to the body. . » 9 Consider. for example, the pattern of activation that gr-ises in those neurons in the topographic map of the retina that characterize color. The patterns of activation over such neurons characterize color categories, and we experience those patterns as colors within those categories. Those same pattems .of activation, located in some other topographic map in the brain, would be experienced as something else: say, in the map of auditory space. they would be experienced as sounds. Activation patterns considered in isolation from their locations in the brain are meaningless in themselves; at particular locations in the brain. they are meaningful — not by virtue of the location per se but because of the way such locations are connected to the body. It is in this sense that the neural networks of the brain(asitishookeduptothebody)appeartobethe right locus for a theory of how concepts are embodied. They are thus a suitable basis for a theory of meaning, given the additional assumption that our perceptual systems respond in a lawlike way to external stimuli. 3. The Connectionist Theory of Mind versus Connectionist Modelling Connectionist cognitive science. as I understand it. is ultimately about the physical brain as located in the body. it is important to see how connectionist modelling fits into the overall program of connectionist cognitive science. Models of neural networks. simulated on a computer. are very different from the physical brain as it is located in the human body. Computer models of neural networks are disembodied. Their job is limited. Their job is to study the properties of particular neural architectures.