Object representation issues
----------------------------

 
One of the reasons for paying careful attention to representation is that the
representation of data in working memory directly affects the generality of
learning.  A prime reason is that the conditions of chunks are based on the
working memory elements matched by productions and not the conditions in the
productions themselves. A second reason is that different representations
allow more or less direct access to different units of knowledge.  If a piece
of data can be accessed only by also accessing other data, chunking will be
unable to separate out only the relevant access.  To date we do not have an
complete theory of how a given situation should be represented in Soar;
however, we do have a set of general rules that should be obeyed when
deciding on a representation for objects in working memory.  Each of these is
discussed in detail below.

  1. Constants should appear only in places where specific values are
     explicitly necessary; they should not be used when equality between
     values is all that is needed;
     
  2. All relevant relations between objects and predicates on objects should
     be represented explicitly in working memory (as opposed to being
     implicit within production conditions);
     
  3. All functionally independent uses of a concept should be represented
     independently;
     
  4. Disjunctive tests should be handled by explicitly representing the
     disjunction in working memory (rather than by using <<>> in production
     conditions).

Menu

Constants

Predicates

Disjunctive tests