VWS Talk 2: AI Notions of Context  
     The VWS Project      How a VWS Works      How to Join a
VWS      How to Give a
VWS      Relevant
Information      The VWS
Programme      People
Virtual Worldwide Seminar Project

AI Notions of Context

John McCarthy1
December 9, 1998


There will be overlap with my 1993 IJCAI article, http://www-formal.stanford.edu/jmc/context.html, but I'll emphasize new stuff.

We treat contexts as objects in (mostly) first order theories. The most basic relation is _Ist(c,p)_ asserting that the proposition _p_ is true in the context _c_. Every assertion, including those about contexts is made in a context, and therefore when we assert _Ist(c,p)_, we do it in some context _c0_ and therefore write

     c0: ist(c,p).

Besides propositions, there are individual concepts, and _Value(c,exp)_ denotes the value that the individual concept _exp_ has in the context _c_.

We consider relations among contexts such as that one context specializes another or has a certain relation to that of another.

Here are some ideas and topics that will be discussed.

  1. "What is context?" is a misleading question to be replaced by "What kinds of context objects are useful in AI, philosophy and linguistics?" There are many.
  2. All assertions about particular contexts are made in contexts. There is no outermost context, and the outermost considered so far may be _transcended_.
  3. Two contexts with the same true sentences may differ in their relation to other contexts.
  4. contexts associated with the beliefs of a person
  5. contexts associated with the contents of a database
  6. contexts associated with particular times and places, e.g. this abstract or the context associated with the actual lecture
  7. relations between the values of expressions in two different contexts asserted in a third, outer, context, e.g.

         c[this abstract]: Value(c[Stanford-Stockholm seminar of 1998 January 16],Lecturer)
         Value(c[Stanford Computer Science Department],John McCarthy).

  8. contexts associated with assumptions including counterfactuals, e.g.

         c[if another car had come over the hill when you passed that Lexus]

  9. contexts associated with mathematical theories, e.g. group theory or common sense theories about the different kinds of automobiles
  10. fictional contexts, e.g. that of the Sherlock Holmes stories
The problem for AI is to get theories expressed in languages of mathematical logic within which computer programs can reason about the relations between objects associated with different contexts.


Are UP! title.html
(Wed Dec 9 03:59:03 1998)

Gzipped, tarred version (so can look at locally)


1 John McCarthy
Room 208,
Gates Building 2A [click on the building for directions]
Computer Science Department
Stanford University
Stanford, California 94305-9020

Top of this page

 Contact: Virtual Seminar Coordinators
Last modified: Wed Feb 3 14:29:20 PST 1999