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

[Jerry.Ball at williams.af.mil]

Yes, this would give you the ability to activate chunks that contain the
retrieved chunk as a slot value, however, it would not activate chunks that
are the slot values of the retrieved chunk (as the goal chunk does).

 From an architectural perspective, if the retrieval buffer is considered to
be part of the "focus of attention" along with the goal buffer, then an
argument can be made for spreading activation from the (slots of the) chunk
in the retrieval buffer as well as from the chunk in the goal buffer.  This
argument is made stronger if the chunk in the goal buffer is essentially
just another declarative memory chunk as has been suggested in defense of
removal of the goal stack.  That is, all declarative memory chunks in the
focus of attention should spread activation to declarative memory (not just
the goal chunk).  

Of course, this begs the question -- is the chunk in the retrieval buffer in
the focus of attention?

Jerry

-----Original Message-----
From: Frank E. Ritter [mailto:ritter at ist.psu.edu]
Sent: Friday, November 29, 2002 9:37 AM
To: Jerry.Ball at williams.af.mil; act-r-users+ at andrew.cmu.edu
Subject: Re: [ACT-R-users] Spreading Activation and Retrieval Chunks


At 15:24 -0700 27/11/02, Jerry.Ball at williams.af.mil wrote:
>In ACT-R 5.0, only the goal chunk spreads activation to declarative memory.
>This means that before a chunk can spread activation, it must become the
>goal chunk.  Thus, if a model retrieves a chunk from declarative memory and
>wants to retrieve chunks associated with the retrieved chunk, the retrieved
>chunk must first become the goal chunk.  This makes it difficult to create
a
>model that follows associative links without putting the goal of following
>the associative links into the declarative chunks themselves.
>
>Given that fact that activation in ACT-R 5.0 only spreads to directly
>related chunks (i.e. there is no propagation beyond directly related
>chunks), allowing activation to spread from the retrieval chunk as well as
>the goal chunk would provide a secondary mechanism for spreading activation
>and would eliminate the serial bottleneck of having to push retrieval
chunks
>into the goal buffer (which takes at least 1 production just to shuffle
>chunks around) to follow associative links.

couldn't you make the retrieval chunk part of the goal chunk, and get 
some, albeit, a lighter activation propogation?

Cheers,

Frank