This lecture first picks up on some topics that got squeezed out of earlier lectures on structured connectionist modeling. The main issue addressed is the controversy over distributed vs. localist representations. There is virtually no current technical work on this, but a loose notion that the brain must use highly distributed representations pervades much general discussion. The truth, which has been known for almost two decades, is that neither extreme position could possibly be right on basic computational grounds. Feldman's chapter in the Waltz and Feldman book on reserve is one source.

The other major topic covered is a detailed account of how X-schemas at the computational level can be expressed in terms of structured connectionist models. The most difficult problem is variable binding. Fortunately, and surprisingly, the temporal synchrony binding mechanism of Shruti, described in Lecture 11, extend nicely to the X-schema case. A postscript version of the latest paper on this is available as:

http://www.icsi.berkeley.edu/~shastri/psfiles/schemas.ps

you may also get to it via the references listed in the SHRUTI homepage