Competitive latency problem

[Todd R Johnson]

problematic for certain kinds of tasks. 

In the model I'm working on, subjects briefly see thirty three-shape
exemplars of the form aba, such as (circle triangle circle) and (diamond
circle diamond). They are then presented a series of test triples and must
indicate whether the test triple was or was not seen during study.
Experimental results show that subjects sometimes base their decision only
on whether the form matches (e.g, aba vs. abb). The model uses a positional
encoding for each element of a triple and also stores the relationship (aba)
with each element:

	(s11 isa entity value star location pos1 relation aba exemplar e1) 

When given a test triple, one production attempts to retrieve an entity with
the same relation.
As a result of the representation, there are 90  entity chunks. With partial
matching on, the latency paramater (F) at 1, and goal activation set to 0,
Act-R takes nearly 90 seconds to retrieve the relationship. Even with F at a
more reasonable .05, it still takes around 4.5 seconds.
Turning on goal activation doesn't help much, because the shapes in the test
triple activate many of the entities.

It possible to produce more reasonable times by assuming that subjects
notice the aba pattern and store it separately. However, its easy to think
of tasks where this is not possible. For instance, the Sternberg model in
Unit 6 avoids this problem becasue Act-R is reset before each trial. But of
course subjects cannot simply reset their memory. I suspect that if the
model were written to retain all of the previously studied items, that each
successive trial would take slightly longer than the previous trial. I've
tried to do this, but it seems that Act-R 5 is no longer creating chunks on
the RHS, as in:

+goal>
      ISA         do-sternberg
      trial       =trial

The dummy trial chunk should allow the model to discriminate between old and
new memory sets, but in this case, I end up with a new goal where trial is
nil.

It seems odd to me that it would take so long to recognize an item when you
have studied many similar items. In fact, I would think that it would take
less time. Perhaps this has something to do with the difference between
recognition and recall. In Act-R to recognize, you must recall. Perhaps
recognition is a different process with a different time equation.

Todd