Ontologies: Hot topics and a future direction

1
Ontologies Hot topics and a future direction

• Alistair E. R. Campbell
• Hamilton College
• IISI Workshop on Strategic Research Directions in
  AI
• 26 June 2003

2
Outline

• Intro Grounded reasoning via ontologies,
  ontology defined.
• Hot topics
• Ontologies
• Languages
• Standards
• Future direction (a.k.a. what Alistairs up to)
• Automated taxonomization

3

• Hot Topics

4
Grounded reasoning

• Classic example
• Forall x (jdeax(x) gt krtac(x))
• jdeax(mveziip)
• krtac(mveziip)
• Correct by accepted semantics of FOL
• Equivalent
• Forall x (Man(x) gt Mortal(x)
• Human(socrates)
• Mortal(socrates)

5
Why Happier

• Mortal, Man, and Socrates are from our common
  ontological experience.
• Mortal property
• Human class
• Socrates a well-known instance
• Semantics/pragmatics justify the example
• Mortal applies to living things.
• Socrates really was human.
• Socrates really did die.

6
Ontology defined

• Gruber Explicit formalization of a
  conceptualization
• Taxonomic hierarchy plus
• Axioms constraining
• Gruber term interpretations
• Campbell term usage
• (two sides of the same coin)
• Logic supporting reasoning
• Helpful Taxonomic classifier
• In short Vehicle for achieving grounded reasoning

7
Some Hot Ontology Efforts

• CYC
• OpenCyc just released for Windows
• IEEE SUO
• SUMO (Suggested Upper Merged Ontology)
• Freely available
• Approx. 1000 terms
• Constraint axioms
• UMLS
• Their own upper model
• Medical domain ontology
• Merged from several sources, by hand

8
Hot Logic/languages for ontologies

• First order logic / (KIF)
• Semantic network logics / (SNePSUL)
• Description logics (DAML/OWL)
• Decidable subset of first order logic

9
A Particularly hot Language DAML/OWL

• W3C proposed standard
• Ontology specification
• For the semantic web
• Are the semantic web
• About the semantic web (meta-ontologies)
• Description Logic
• Decidable subset of First Order Logic
• Properties of objects, and properties of
  statements about objects
• Promoting ontology reuse via namespace inclusion
• Multiple Inference engines
• Manual ontology concept equivalence axiom
  (damlequivalentTo)

10
Equivalence is key

• We will always have
• Forall x (jdeax(x) gt krtac(x)
• krtac(mveziip)
• jdeax(mveziip)
• But if we also know
• jdeax SUMOhuman
• kratc WordNetmortal
• mveziip BritanicaSocrates
• That reasoning was grounded
• Can be trusted
• Might be useful to someone else

11

• Future Direction

12
Ontologies for flat data sets

• Flat data set
• Discrete elements describing a domain
• No taxonomic hierarchy
• Example DoD Metadata Registry and Clearinghouse
  (XML tag registry)
• Problem Produce a system that will take as input
  a flat data set and produce an output a taxonomic
  hierarchy of the records in that set.

13
Ultimate Goal

• Induce a taxonomic structure for a flat data set
  (semiautomated)
• Identify properties to taxonomize over
• Probably manually in the short term
• Search for a taxonomy that meets constraints
• Shape
• Balance
• Link flat data set elements to existing
  ontologies (fully automated)

14
Recognition of problem space size

• Given a set of discrete attributes ?, where Va is
  the set of values for attribute a, of
  taxonomies
• Examples
• lt2,2,2,2,2,2,2gt space T(?) 1.9x1027
• lt4,2,5,4,2,2,2gt space T(?) 5.85x10222

15
Mapping to lexical ontology

• WordNet dense lexical ontology
• gt100,000 synsets with short glosses
• XML tags short glosses and labels
• Task identify words in the label
• Lexical chaining using glosses
• Disambiguate polysemous senses
• 70 accuracy initially
• (initial finding, not a systematic study, yet)

16
Irony

• Big problem with keyword-based web searches
• Ontologies are supposed to mitigate by giving
  semantics to web content
• But (much) ontology mapping technology uses
  keywords
• Next step more robust NLP techniques.
• Read term glosses for ontological content
• (parse and understand to nontrivial extent)
• Provide more accurate matches
• Identify partial matches more precisely

17
Summary

• Ontologies moving from theory to practice
• Mature Ontology logics, construction tools
• Medium mature distributed ontology systems
• Maturing Ontology integration algorithms
• Immature Automated ontology construction
  techniques