CAs as Computational Primitives
- Existing models and theories support CAs as the means of
recognising conceptual categories. This accounts for both
long-term and working memory.
- A slight theoretical extension along with a working model shows that
CAs can compete with each other.
- Hierarchies are also a means of classifying. A dog is a
member of the mammal category. It is straight forward to
add Hierarchies to a CA-based model. I don't know of a model
to do this but am working on one.
- Of course people do more than categorise.
- CAs for sequences. Most of the early CA modelling work was
done with sequences in mind.
- CAs for Structures. Structures are essentially inter-CA
connections. No modelling work has been done on this, but
at least theoretically it is supported by neural connections and
priming data.
- CAs for Rules. I've shown how CAs might be used for variable
binding and thus for rules in general. I plan on exploring this
problem via modelling over the next two years.
- CAs for processes. I've described a parsing specific processing
mechanism. General processing might be limited to a small number of
flexible mechanisms. A CA-based model would involve mostly learning
categories and relationships, not processes. A few new processes
may need to be developed but the number should be small.
