[ACT-R-users] Spreading Activation and Retrieval Chunks

[Dan Bothell]

--On Monday, December 02, 2002 12:07 PM -0700 Kevin.Gluck at williams.af.mil 
wrote:

> Dan recently mentioned that a "unified" state mechanism is under
> consideration for ACT-R 6.  Perhaps a unified spreading activation mechanism
> for the buffers would be worth considering, as well.

Design documentation of a potential framework for ACT-R 6 is something that I 
hope to have available in the not too distant future for discussion, but the 
details of specific functionalities are definitely something that I don't plan 
on putting in right away because I think they will require some serious 
discussions.

As for the issue at hand, this is one of those things that I think needs to be 
analyzed and discussed thoroughly.  I remember at some point in discussing 
buffers for ACT-R 5 that someone had suggested that all buffers (at least all 
real buffers i.e. not the state buffers) should be sources, but that isn't 
what we ended up implementing.  It does seem like a system that would be easy 
to describe and build (which is a plus from my perspective), but it does have 
some ramifications from a modeling perspective that do need to be considered. 
Such a system would break backward compatibility and probably "invalidate" 
many existing models' predictions (a simple porting of an old model to such a 
system could produce quite different predictions). It isn't unheard of however 
to have things other than the goal be sources of activation.  Looking back to 
ACT-R 4 with PM it was the case that the currently attended chunks of the 
perceptual systems (what would now be the chunks in the perceptual buffers) 
were sources of activation along with the goal, so perhaps it wouldn't be too 
big of a change.

Dan