The Architecture of a Multiple Representation System (Draft). (MRS),
Abstract
In a broad sense, much of computer programming is simulation. On encodes facts about some aspect of the world and writes programs to answer questions using those facts. This is true whether one is a writing a complex numerical code or an inventory control program or an expert system to do medical diagnosis. In non-numerical programs the selection of a good data structure to encode facts particularly important and is often more difficult than writing the code that processes it. Over the years, two distinct approaches to representation have arisen, which might be called theory -building and model-building . The theory-building approach is typified by a growing number of knowledge representation systems (e.g. KRL (ref), KLONE (ref), Prolog (ref). UNITS (ref)). Most of these systems provide language in which the programmer encodes facts about the world, including in many cases parital information (e.g. Arthur is Bertram's father or Allison is his mother. ) and quantification (e.g. All apples are red ). Most offer some general inference capabilities for reasoning about facts expressed within their language (e.g. inheritance in the frame languages). In the vocabulary of formal logic, a set of such sentences is called a 'partial theory'; and, if the sentence are true in the world being described, that world is called a model of the theory. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- May 30, 1981
- Accession Number
- ADA123296
Entities
People
- Michael R. Genesereth
Organizations
- Stanford University