architecture comparisons

[Bonnie E. John]

>Also of interest are papers that compare the features and/or predictions of
>two or more models created in different modeling architectures.
Kevin,

Yannick Lallement and I wrote a paper comparing an EPIC model with two Soar 
models (one in the "released" version of Soar circa 1998 and one in an 
"experimental" version of Soar that Ron Chong used).

Lallement, Y., & John, B. E. (1998)  Cognitive architecture and modeling 
idiom: An examination of three models of the Wicken's task. Proceedings of 
the Twentieth Annual Conference of the Cognitive Science Society, August 1998.


Conclusions in that paper:

+ All three architectures were adequate to make close quantitative 
predictions of a dual-task.

+ The "idiom" was as important as the architecture (where "idiom" is a 
"programming style" within an architecture).

+ Parallelism in cognition didn't matter because the time to perform the 
task was limited but the actions of the perceptual and motor peripherals.

+ I'm not sure you can actually find a task that differentiates between 
these cognitive architectures on performance time measures, because fast 
ones with little thinking will be dominated by the peripherals, and slower 
ones with a lot of thinking will be subject to individual differences in 
task strategies and prior knowledge.  (I didn't say it in that paper, but I 
think we'll have to go to learning to differentiate between these 
architectures, but of course that leaves out EPIC, and that's REALLY hard 
to do because learning has the problems above (being dependent on 
individual strategies and prior knowledge)).

Bonnie