how can ACT-R models age?

[Bonnie E. John]

>An aside: Now I think it is an interesting challenge to models like ACT-R 
>or SOAR about the interference you could get if you left knowledge from 
>previous experiments in them.  It has always seemed unfair that these 
>models are evaluated starting from zeroed out memory (unlike 
>humans).  Supposed you had to use the same ACT-R model for the next 
>experiment that you used in the last experiment and you weren't allowed to 
>remove any chunks (of course, any natural decay process in them could 
>continue to work).

I agree with this Stu, and have made it part of my research program.

It's exactly what Yannick Lallement and I did with the ATC-Soar model -- it 
had the hand-written and learned productions from the Wicken's task loaded 
before it started the ATC task.  much of that knowledge didn't get in the 
way because Wickens-task-specific knowledge had a test for being in the 
Wickens task, so it never fired in the ATC task. Some may say that is 
cheating, but boy those screens sure looked soooooo different that it 
wasn't a stretch for me to beleive that enough environmental cues would 
have been associated with any Wickens' rule that differentiated it from the 
ATC environment that it would never fire in the ATC environment -- and 
isn't what underlies the difficulty that people have with transfering 
knowledge from context to context?  We have a new paper submitted on this work:

Experiences building a zero-parameter model that learns to perform a 
complex, dynamic, computer-based task, Bonnie E. John & Yannick Lallement

We also did that years ago with NL-Soar and the NASA Test Director model 
(NTD-Soar). Boy was that model bloated -- having all that NL code in with 
the NTD stuff. There was no NTD test, because NL was needed in that task 
and we always thought of it as a general capability that shouldn't be 
task-related. But the words were so different in the NTD task that all that 
knowledge about other words didn't get in the way.
You can read about that in

Nelson, G. H., Lehman, J. F., & John, B. E.  (1994) Integrating cognitive 
capabilities in a real-time task.  Proceedings of the Sixteenth Annual 
Conference of the Cognitive Science Society, August 1994. pp. 353-358.

Nelson, G., Lehman, J. F., John, B. E. (1994) Experiences in interruptible 
language processing, In Proceedings of the 1994 AAAI Spring Symposium on 
Active Natural Language Processing, 1994.