Cognitive models have explained a great deal of behavioral and neurophysiological data. On the road to understanding the mind, cognitive architectures have specified a core set of representations and mechanisms common to a variety of models. Such architectures separate general functional components and their abilities from domain-specific instantiations, such as knowledge and strategies.
On the road to describing cognition in real-world situations, models have been becoming increasingly complex. As task complexity increases, a careful analysis of the components of the model is necessary. For anything but the simplest cognitive models in ACT-R, many of the procedures and data structures they define are often not evaluated: the specifics of many of the components of the model may be irrelevant to the story a model has to tell. The solution to this problem is under-specification. In what we call the Accountable Modeling paradigm, we suggest applying Occam’s razor and specifying only what is meant to be directly or indirectly evaluated. As a consequence, we arrive at models that can be more complex yet faster and easier to prototype, while still using the same core representations and mechanisms of the architecture.
This motivated ACT-UP, a reimplementation of the ACT-R theory (Anderson 2007). ACT-UP implements a substantial subset of the ACT-R theory. ACT-UP is a toolbox providing ACT-R functionality in a piecemeal fashion as well as commands at a higher abstraction level. The ACT-UP library is a stand-alone system, and independent of ACT-R 6. It provides a set of Lisp functions and macros; modelers interact with it on the basis of source code that follows Common Lisp syntax (see this tutorial for a step-by-step introduction). ACT-UP models predict the two major behavioral outcome types: choice and timing.
Unlike in ACT-R 6, modelers do not write out series of production rules to program the model. Instead, they use a programming language with its loops, conditionals and variable definitions. The evaluation is sequential rather than parallel (unless parallelism is needed).
In the following, we give a high-level overview of the cognitive functionalities of ACT-UP in relation to ACT-R 6.
Most of the tutorial models from the ACT-R 6 distribution have been re-implemented in ACT-UP; they come with ACT-UP.
Please visit the following links to get a head start on ACT-UP. (This is also included with the ACT-UP distribution.)
Slides from our 2010 MODSIM World presentation (Oct 2010)
Right-click the link and choose "Save Link As..." to save the document to your computer.
ACT-UP.ZIP (Beta version, updated regularly) Current Version: 27bc8ed Sun Jul 17 07:10:59 EDT 2011
Common Lisp: SBCL, OpenMCL/CCL, LispWorks work. Allegro: limited testing.
Subscribe to the low-volume ACT-UP mailing list for user support, and for development announcements.
Your feedback is important to us. Please click on the link below to send us your comments and suggestions.
David Reitter, Christian Lebiere, Jasmeet Ajmani, Carnegie Mellon University
Thanks: Dan Bothell for code contributions and advice.