[ACT-R-users] Declarative Rule Interpreter

James Peters jpeters at rhsmith.umd.edu
Sat Aug 28 10:26:15 EDT 2004


I am working on a rather large project to implement a practical, generic 
declarative rule interpreter in ACT-R based on the prototype in Tutorial 
#7.  I believe that this is an important project for the following 
reasons:

1.  In theory, all knowledge starts as declarative and is compiled into 
productions.  Since step by step instructions and other forms of process 
knowledge are a key part of the knowledge an expert brings to any task, 
ACT-R needs to have some way of representing, executing, and 
proceduralizing this sort of knowledge.  ACT-R has a mechanism for 
proceduralization (admittedly an incomplete and evolving one, but there is 
a mechanism) but the Tutorial Appendix is the only mechanism I know of for 
executing declarative process knowledge (i.e., rules).

2.  In my domain (accounting), experts rarely have sufficient practice at 
any task to proceduralize their knowledge, at least the higher level 
knowledge.  Clearly, high-order knowledge contains primitive components 
that are highly practiced (like basic math skills), but accountants 
constantly face new issues that require working through declarative 
structures to resolve.  In addition, they always need to explain their 
actions and, therefore, are constantly using declarative representations 
of process knowledge (since procedural knowledge is unavailable for 
explanation except if the expert observes their own actions and then 
retrospectively infers the procedural knowledge used). 

Finally, administratively, managing large production systems is very 
difficult in ACT-R.  I need to manage large knowledge bases to properly 
model my subjects.  Declarative knowledge is much easier to manage because 
it fits into a relational database structure very nicely.  I am building 
an Access database to manage chunks.  The database will merely produce a 
text file with the "add-dm" and "chunk-type" statements needed to feed the 
information into ACT-R.  The system will also be able to track links 
between productions and chunks through the chunks embedded in the 
productions slots in the same way ACT-R instantiates productions, which 
should help manage the productions as well.  It would add the "P" 
statements needed to feed productions into ACT-R as well.  This will give 
me a very powerful tool to sort, select, edit, and display chunks and 
productions is just about any way I want to.  I haven't progressed too far 
with the database side because I need to build more extensive, generic, 
productions in ACT-R to work with declarative knowledge, particularly 
process oriented declarative knowledge.  That is, before I build a massive 
database, I need an ACT-R system that can handle it.

If you are working on a similar project(s), or know of someone who is, I 
would love to hear from you.  There are no other ACT-R researchers at the 
U. of Maryland and so I could use someone to talk to. 

Thanks.

Jim
James M. Peters, Ph.D.
Department of Accounting and Information Assurance
R. H. Smith School of Business
University of Maryland
College Park, MD  20742-7215
TEL - (301) 405-0570
FAX - (301) 314-9414
jmpeters at umd.edu
http://www.rhsmith.umd.edu:8001/jpeters/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.srv.cs.cmu.edu/pipermail/act-r-users/attachments/20040828/a9a1bd7a/attachment.html>


More information about the ACT-R-users mailing list