[ACT-R-users] Visual Perception in ACT-R

[Dan Bothell]

--On Friday, February 9, 2007 9:45 AM +0100 Marc Halbruegge 
<marc.halbruegge at unibw.de> wrote:

> Hi!
>
> I've a question concerning the feature scale of ACT-Rs vision.
> In Anderson et al. (1997, full citation below) on page 446, there's an H
> presented that is made up of Xs, like this:
>
> X   X
> X   X
> XXXXX
> X   X
> X   X
>
> The caption of the figure says "ACT-R can see either the H or the Xs
> comprising the H, depending on how ACT-R sets its feature scale"
>
> Now my questions:
> - Is there an ACT-R model available that demonstrates this ability?
> - How do I set the feature scale (not 'word' vs. 'letter' but
>   'H' vs. 'X')?
> - Or is this not yet implemented? (in the text, the claim is less
>   strong: "... we would want [ACT-R] to recognize either the H or the Xs
>   comprising the H.")
>
> Thanks for your answers!
>


That paper is describing a visual attention system which
is the predecessor of the current implementation, and I don't
really know the details as to how one would have done such a
thing (if possible) with that vision system.  There is no
mechanism in the ACT-R 6 vision module which would allow
the model to "see" an H automatically from those individual
X's.  However, it is possible to create your own features for
the model's visicon and thus you could add an H to the set
of features available to the model. If you gave it a value for
its kind slot which was different from the normal letters (which
are of kind text) then you could just specify that when making
the visual-location request.  So, to find an X you would use
something like this (along with any other constraints as needed):

+visual-location>
   isa visual-location
   kind text
   ...

and to find the H you would need to specify the new kind:

+visual-location>
   isa visual-location
   kind large-text
   ...


Hope that helps, and if you have questions about how to
augment the visicon let me know.

Dan