Perception and Attention

Memory and Thought

Language and Conceptual Systems

Education in Math, Science, and Technology

Foundations of Cognitive Science

The Neural Theory of Language and Thought

The World Color Survey

Learning Complex Motor Tasks

Perceptual Organization in Vision

Metaphors in Language and Thought

Cognitive Neuroscience of Memory and Cognition

Control of Automated Vehicles

Crosslinguistic Studies of Early Language Development

Understanding Explanatory Coherence

Children's Theories of Mind

Spatial Cognition

Neuropsychological Studies of Mind and Brain

Biologically Motivated Computer Vision

Soft Computing

Cognition and Action

The Neural Theory of Language and Thought

For the past 11 years Profs. Feldman and Lakoff have been collaborating on a joint research to determine how the human physical brain supports the representation of , composed of neurons that function via chemistry, can give rise to human concepts and human language. The project seeks to combine results from neuroscience, computational neural modeling, cognitive linguistics, and cognitive and developmental psychology. Much of their work has had the form of neurally plausible computational models of language learning and use. Topics they have addressed include: optimal Bayesian construction matching, learning spatial relations, learning concepts of hand motion, aspectual concepts and aspectual reasoning, metaphorical reasoning and language, and children's acquisition of grammatical constructions tied to basic sensory-motor experiences. Much of this work is aimed at constructing a general theory of neural grammar and first-language acquisition, as well as improved human-computer interaction. It has also led to the development of a new undergraduate course, CogSci 110, "The Neural Basis of Language and Thought." This project is being conducted jointly with the International Computer Science Institute (ICSI). (Professors Feldman and Lakoff)

The Neural Theory of Language (NTL) Group, headed by Profs. Feldman and Lakoff, has evolved to the point where it makes sense for it to become a research center within ICBS, working cooperatively with the Brain Imaging Center as well as existing ICBS activities. It will be called the Center for the Neural Study of Language (CNL) and be headed by Profs. Feldman and Lakoff.

The purpose of CNL is to extend a decade of interdisciplinary research in ICBS to answer the following question in as much detail as possible: how can the physical brain, which is composed of neurons that function chemically, gives rise to human concepts and human language? We call this The Neural Language Problem. The focus of CNL is on neural computational mechanisms capable of characterizing fine details of language and thought in neural terms. The NTL Group has attacked the problem of how the brain computes the mind by combining results from neuroscience and cognitive and developmental psychology with our specialties, which lie in computational neural modeling and cognitive linguistics. We believe that enough solid results have been achieved so that The Neural Language Problem can now be studied as a part of normal science.

For example, the learning of spatial relations concepts in a number of languages has been successfully modeled (by former student, Terry Regier in our lab) for a significant range of cases. Regier's work was based on Talmy's analysis began with earlier ICS research by cognitive linguist Leonard Talmy (now Chair of Cognitive Science at SUNY Buffalo) showing how systems of spatial relations concepts around the world can be reduced to a universal set of primitives. Regier's breakthrough was to show how those primitives could be computed by well-studied neural structures in the visual system. The technique used was to use structured connectionist methods of modeling neural computation to demonstrate that actual neural structures could compute spatial relations concepts used throughout the languages of the world. By this method, neuroscience is linked to cognitive linguistics via methods of neural computation.

This methodology is being carried forward in other studies. The learning of concepts of hand motion has been successfully modeled by former student, David Bailey (now at on the basis of computational models of motor synergy and complex motor schema. Aspectual concepts and aspectual reasoning (which concerns the structure of events) have been successfully modeled by another student, Srini Narayanan (now at SRI), using computational models of neural control mechanisms for motor schema. Narayanan has also successfully modeled metaphorical reasoning and language on the basis of sensory-motor inferencing mechanisms and metaphorical maps, fitting the cognitive theory of metaphor of Lakoff and Johnson.

These results, taken together, permit further development of a broad-based theory of neural semantics. Our next steps are to integrate grammar and phonology. This is extending our scope to include the work of Charles Fillmore and Eve Sweetser and the speech processing effort of Nelson Morgan of EECS and ICSI.

Nancy Chang is extending the NTL paradigm to the learning of grammar. She is building on prior ICS research in psycholinguistics and linguistics by Dan Slobin and Adele Goldberg which shows how the acquisition by children of grammatical constructions is tied to basic sensory-motor experiences, that can be modeled by techniques described above. The theory of grammar used is another ICS product - construction grammar developed by Charles Fillmore, Adele Goldberg, George Lakoff and others at ICS. Our group, in collaborative work, is demonstrating how the neural mechanisms we are already capable of modeling can characterize grammar. This exploits early ICS-based work on optimal Bayesian construction matching by Dan Jurafsky (now at Colorado) and grammar learning by Andreas Stolcke (now at SRI). This work may result in a general theory of neural grammar and of first-language acquisition, as well as improved human-computer interaction.

A neural theory integrating phonetics and phonology has begun to be developed by Ben Bergen. As part of our general program of neural explanation via models, a computational modeling of neural binding has been developed by Lokendra Shastri of ICSI and his students.

This effort has yielded a number of NSF grants and played an important role in the ICS Spatial Cognition program. We have also developed a new course, CogSci 110, to teach the current state of this and related research and are at work on a trade book to disseminate the results to the general public.

ICBS group on the neural theory of language has established a general research program for the computational modeling of the general neural mechanisms of thought and language, and that research has advanced sufficiently to establish a Center for the Neural Study of Language within ICBS.