Understanding the Neural Basis of Episodic Memory
                  (What's language got to do with it?)

                         Lokendra Shastri
             International Computer Science Institute
                           Berkeley, CA
                 http://www.icsi.berkeley.edu/~shastri
                        
We readily remember events and situations in our daily lives and acquire
memories of specific facts by watching a telecast, reading a newspaper,
or participating in a dialogue. This remarkable "one-shot" mnemonic ability
poses a challenge for cognitive and computational neuroscience.

It is reasonable to assume that the construal of an experience (or utterance)
in terms of an event is initially expressed as a pattern of activity over
distributed neural circuits in the brain. This expression, however, is per
force transient, since it must change continually as we interact with the
environment. Hence, the transient, activity-based expression of a
memorable event must be transformed rapidly into a persistent structural
encoding, or else it would be lost. How does the brain perform this rapid
transformation?

Events and situations are relational instances and not mere feature vectors.
What is the transient, activity-based representation of a relational
instance? What is the persistent, structure-based encoding of a relational
instance?

How do memories of specific events respond to partial cues, yet exhibit
strong pattern separation? How do such memories interact with representations
of general semantic knowledge during memory retrieval and inference?

I will review ongoing work on biologically grounded computational modeling
of memory and reasoning that attempts to address these questions, present
some results, and discuss some predictions about the representation and
processing of relational information in the brain. Some of the predictions
are supported by recent empirical findings, some others point the way to
interesting experimental studies.