In an initial case study in the area of purchasing, we find that much of currently non-automated guidelines information can be represented successfully using standard logicist methods. We show how to create very-high-level specifications with well-understood semantics. These specifications are useful as descriptive information in non-executable form; they help identify ambiguities, inconsistencies, and omissions in less formal guidelines information. We also show that these very-high-level specifications are partially executable using standard logicist inference methods such as rule-based programming and/or logic programming. This is useful for development of prototype software and simulations.
Our logicist approach provides a rich language and set of methods for
describing and propagating constraints, especially when compared to
other less formal approaches such as the Entity-Relationship model.
However, we discover that most guidelines information relies heavily
on common sense, and raises difficult knowledge representation
challenges in temporal, default, decision-theoretic, and
multiple-level reasoning that expose the limits of state-of-the-art
logicist methods.
[
IBM home page |
Order |
Search |
Contact IBM |
Help |
(C) |
(TM)
]
Last update: 1-8-98
Up to Benjamin Grosof's Papers page
Up to Benjamin Grosof home page
[ IBM Research home page ]