KRYPTON: A Functional Approach to Knowledge Representation 1983

1
KRYPTON A Functional Approach to Knowledge
Representation (1983)
RJ Brachman RE Fikes HJ Levesque

• CIS 6930 (Knowledge Representation)
• Fall 2002, University of Florida
• Eric Spellman

2
Outline

• KRYPTON marries frames and logic.
• Current frames have problems which KRYPTON tries
  to alleviate.
• Architecture of KRYPTON
• Reasoning in KRYPTON

3
Problems with Frames

• Strict interpretation too limiting.
• Either Elsie or Bessie is the cow
• Need for expressive power forces a loose
  interpretation as non-factual descriptions.
• Lack of precision leads to misinterpretation.
• Counting the kinds of rocks
• Number of links between concepts

4
Motivation

• Problems with Frames
• Facilities misused for extra expressive power.
• Meaning depends on direct interpretation of data
  structures.

• KRYPTON Philosophy
• Separate structural from assertional
  knowledge.
• Specify behavior explicitly, independent of
  representation.

5
KRYPTON Architecture

• Two components
• TBox handles structural knowledge
• ABox handles assertional knowledge

dog
beagle
Symbols
TBox
Dog(X) -gt Friendly(X)
Question Is a beagle friendly?
ABox
6
TBox

• Defines taxonomies of concepts and roles.
• Strictly defines relationships with language of
  operators.
• Define beagle as
• (VRGeneric dog
• breed-name
• beagle)

7
ABox

• Defines logical sentences about the concepts and
  roles.
• First-order logic plus predicates from the TBox.
• Adds assertional power on top of frames.
• Dog(X) -gt Friendly(X)

8
Functional Specification

• ABox
• TELL(s) -- asserts s is true.
• ASK(s) -- checks if s is true.
• TBox
• TELL(sym, t) -- defines sym as t.
• ASK1(t1, t2) -- checks if subsumes.
• ASK2(t1, t2) -- checks if disjoint.

9
Relating the Boxes

• Frames become one-placed predicates P(x)
• Slots become two-placed predicates Q(x,y)
• Two strategies for implementation
• Axiomitze with meaning postulates
• Extend standard logical inference

10
Conclusion and Discussion Points

• KRYPTON divides knowledge into definitions and
  assertions.
• This allows its definitions to be precise while
  maintaining expressiveness.

• How do we translate TBoxs definitions to logic
  so that we can reason with them?
• Are the current TBox operators enough?