Welcome to Cognitive Science 110 / Linguistics 109 / Computer Science 182!

This class will be taught in Spr. 1999 on Mondays and Wednesdays, 11 - 12:30 PM, in 306 Soda(Note the classroom change!!!).
It is listed as Cog Sci 110 (cc# 16063), Ling 109 (cc# 52250), and Comp Sci 182(cc# 25007).
This is a 4-unit course, with 3 hours of lecture and 1 hour of discussion section (Fri 11-12 or 12-1) per week (in 320 Soda).

Jerome Feldman
     jfeldman@icsi.berkeley.edu, 731 Soda Hall . Office Hours: M 1-2PM.

George Lakoff
     lakoff@cogsci.berkeley.edu, 620 Barrows Hall, 1213 Dwinelle Hall. Office Hrs: TBA.
 

      TA: Benjamin Bergen
                      bbergen@icsi.berkeley.edu, 1310 Dwinelle. Office Hrs: Tu 11-12, Friday 9:30-10:30.

The Undergraduate Research Apprentice Program announcement .

Attention women undergraduates in EECS: you might want to read this announcement, as well as this one.

The final for this class is scheduled for Thurs, May 20, 5-8pm (Exam Group 17), location TBA.

          This is a course on the current status of interdisciplinary studies that seek to answer the following questions:

1.      How is it possible for the human brain, which is a highly structured network of neurons, to think and to learn, use, and understand language?

2.      How are language and thought related to perception, motor control, and our other neural systems, including social cognition?

3.      How do the computational properties of neural systems and the specific neural structures of the human brain shape the nature of thought and language?

        Much of the course will focus on The Neural Theory of Language (NTL), which seeks to answer these questions in terms of architecture and mechanism, using models and simulations of language and learning phenomena. The focus is less on where such functions are located in the brain, but rather on how  neural systems can carry out the computations necessary to characterize specific concepts, such as spatial relations concepts, aspectual concepts (used in structuring events), abstract metaphorical concepts, and so on.

        Though the answers to these questions are interesting in themselves, they are important for many other reasons as well. Most philosophical theories of mind assume that the mind is disembodied, and that human reason and language can be fully described and explained at a "functional" level, with no reference to actual neural mechanisms. Functionalist approaches to mind assume that mind is a matter of computational "software" that happens to be able to run on the brain's "hardware." This entails that brain structure plays no essential role in characterizing human reason. NTL research indicates the opposite, namely, that the human body and brain play a central in characterizing human reason, and that the peculiarities of brain structure and human computation enter crucially in shaping  the details of our capacity for thought and language. From the applications perspective, deeper computational understanding of meaning is required for information retrieval, search engines and intelligent agents.

        Because there is such a large gap between the cognitive and linguistic descriptions of thought and language and known
neurobiological details, NTL makes use of a neural bridging theory that includes five levels of description and several styles of modeling. The levels are:

• Level 1. The level of Conceptual and Linguistic Description as given by Cognitive Linguistics and other sub-disciplines of Cognitive Science.
• Level 2. The level of Computational Mechanisms (standard, well-understood ones, such as Feature Structures, Petri Nets, Model-Merging, and Mapping Structures). Central to the NTL approach is that mechanisms and particular models be reducible to the connectionist level.
• Level 3. The level of Structured Connectionist Networks. These are the simplest connectionist models of neural structures that are computationally adequate for thought and language.  These models are quite different from the back-propagation networks used in much of Artificial Neural Network research. (Examples: Regier's thesis, SHRUTI and temporal binding, recruitment learning).
• Level 4.  The level of Computational Neurobiology. Models of neural structures. (Example: Shastri's work on hippocampal models of episodic memory).
• Level 5. The Biological Level: Real neural structures with brain imaging (such as fMRI) closing the loop back to the Cognitive and  Linguistic level.

        This course should be useful for students in many disciplines.

• Cognitive Science students will find it necessary for comprehending the dispute between functionalist and biological approaches to mind.
• Computer Science students interested in artificial intelligence will be introduced to structured neural computational approaches to the study of intelligence, as well as findings about the brain and real  human conceptual systems from neuroscience and linguistics.
• Linguistics students will acquire a neural perspective on  language that differs markedly from from formalist viewpoints.
• Philosophy  students will learn about embodied theories of mind, which differ markedly from traditional disembodied theories of mind.
• Neuroscience students, who have mainly learned about localization issues or low-level neural computation, will have a chance to find out about the nature of human concepts and conceptual-level issues in neural computation.
• Psychology  students interested in either child language  learning or cognitive processing can get an acquaintance with neural learning models, both PDP and structured, and with neural processing models for perceptual, motor control, and inferential  behavior.
• Since the subject matter includes comparative language and conceptual structures, students from all disciplines will be exposed to the commonalities and differences in conceptual systems across cultures.

        Our approach to teaching is hands-on to the extent possible. Students will be able to use to interactive software to learn in detail how neural models work, and homeworks will be in that format. We will use the PDP connectionist exercises from Kim Plunkett and Jeffrey Elman's  Exercises in Rethinking Innateness and on-line psychological experiments developed by Barbara Tversky at Stanford. In addition, we will adapt some existing programs for instructional use, as funding permits.

        Sections will meet once a week. There will be weekly problem sets, one major paper assignment and in-class midterm and final exams.

The Basic Models:

        The course is centered on teaching how four fundamental neural models work and what they reveal about language and thought.
 

• (1) Classical PDP (parallel distributed processing) models and Elman nets. These are the most discussed type of neural network models in the cognitive science literature.
• (2) Regier's hybrid PDP-structured connectionist models, in which structured models are used for characterizing spatial relations structures and a PDP-net is used for learning words for them in different languages.
• (3) Bailey's structured connectionist model for learning verbs of hand motion, given an existing computational model of the bodily actions.
• (4) Narayanan's model for general motor control and its extensions to reasoning about aspect (the structure of events and actions) and to the interpretation of metaphor in discourse.

Written Assignments:

Ten Homeworks, a Midterm, and a Commentary on a Behavioral and Brain Sciences paper.

Final exam: A traditional exam assessing what students have learned.

There will also be regular reading assignments aligned with the lectures.

Grading policy:

Final Exam: 15 %
Last Homework (BBS Commentary): 15%
Other Homeworks and Midterm (equal parts): 65%
Section Participation: 5%
The lowest grade out of the set of the first 10 homeworks will be dropped. Late homeworks will generally not be accepted.

Texts:

Regier, Terry. The Human Semantic Potential. MIT Press. 1995.

A course reader, including the following readings:

1. Stevens, Charles (1979). The Neuron. Scientific American, Sept 1979. W.H. Freeman and Co.

2. Crick, F. H. C. (1979). Thinking about the Brain. Scientific American, Sept 1979. W.H. Freeman and Co.

3. Cowan, Maxwell (1979). The Development of the Brain. Scientific American, Sept 1979. W.H. Freeman and Co.

4. Reichert, Heinrich (1992). Introduction to Neurobiology. Ch 5: Motor Systems. Thieme Verlag / Oxford University Press

5. Lakoff, George (1987). Women, Fire, and Dangerous Things: What Categories reveal about the Mind. Ch. 2. University of Chicago Press

6. Talmy, Leonard (1983). How Language Structures Space. In H. Pick and  L. Acredolo, eds., Spatial Orientation: Theory, Research, and Application. New York: Plenum Press.

7. Kay, P & McDaniel, C. (1978). The Linguistic Significance of the Meanings of Basic Color Terms. Language vol 54.3, 610 -646.

8. Plunkett, K. & Elman, J. (1997) Exercises in Rethinking Innateness: A Handbook for Connectionist Simulations. Ch 1 and Appendix B. , MIT Press

9.  Feldman, Jerome (1988) Computational Constraints on Higher Neural Representations, in Proceedings, System Development Foundation Symposium on Computational Neuroscience, E. Schwartz (ed.), Bradford Books/MIT Press, April 1988.

10. McCleland, J & Rumelhart, D. (1989) Explorations in Parallel Distributed processing - A Handbook of Models, Programs and Exercises. Ch. 5, MIT Press

11. Bailey, D., Feldman, J., Narayanan, S., & Lakoff, G. (1997). Modeling embodied lexical development.  in Proceedings of the Nineteenth Annual Meeting of the Cognitive Science Conference, Stanford , Stanford University Press

12. Lakoff, George (1993). The Contemporary Theory of Metaphor. In Ortony, Andrew, ed., Metaphor and Thought, 2nd edition. Cambridge, Engl.: Cambridge University Press

13. Lakoff, George and Mark Johnson. Excerpts on Image-schemas from Philosophy In The Flesh. in press. Ch 3, 4, 5

14. Fillmore, Charles (1985) Frames and the Semantics of Understanding. Quaderni di Semantica. vol.VI no. 2, Dec. 1985

15. Narayanan, Srinivas (1997) "Talking the Talk is like Walking the Walk" Proceedings of the Nineteenth Annual Meeting of the Cognitive Science Conference, Stanford , Stanford University Press

16. Narayanan, Srinivas (1996) Embodiment in Language Understanding: Modeling the semantics of causal narratives , AAAI Symposium on Embodied Cognition and Action, Cambridge Mass. AAAI Technical Report, FS-96-02, AAAI Press

17. Treisman, Anne (1996) The Binding Problem. Current Opinions in Neurobiology v.6, 171-178

18. Narayanan, Srinivas and Danial Jurafsky (1998) Bayesian Models of Human Sentence Processing.

19. Feldman, Jerome et al. (1998) Extending Embodied Lexical Development.

Useful / Fun / Educational  Links:
 

• UC Berkeley CS Division of the EECS Department
• UC Berkeley Linguistics Department
• The Institute for Cognitive Studies
• The Neural Theory of Language (NTL) group
• The Digital Anatomist Interactive Brain Atlas
• The Visual Illusions Page

 
• more . .