Page 1
Florian A. TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Naive Semantic Interoperability
Florian A. Twaroch
Institute for Geoinformation & CartographyVienna University of Technology
Annual Scientific Meeting – GeoGERAS 2006Pernegg, Austria
03rd July 2006
Page 2
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
MotivationMotivation BackgroundBackground HypothesisHypothesis MethodMethod ConclusionConclusion
Overview
● Problem Definition● Humans Conception of Space● Hypothesis● Review of the Literature● Sandbox Geography
– How to organize concepts?– How to change concepts?
● Conclusions● Outlook
Page 3
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation BackgroundBackgroundBackgroundBackground Hypothesis Method Conclusion
How to Build a Concept ?
• Experience
• Humans hold several concepts
• Concepts underlie change
Page 4
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation BackgroundBackgroundBackgroundBackground Hypothesis Method Conclusion
Semantic Interoperability
• … the transition of two mental models …
• Shared reality
• Objectivity (Frank & Mark 1996)
Page 5
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation BackgroundBackgroundBackgroundBackground Hypothesis Method Conclusion
Related Work
● Similarity Measures (Tversky 1977, Rodriguez & Egenhofer 2004)
● Formal Specifications of Interoperability with Image Schemata– Linguistic (Frank 1998, Frank & Raubal 1998)– „Toy Spaces“(Rodriguez & Egenhofer 1997)– „Real Space“(Rodriguez & Egenhofer 2000,
Rüetschi & Timpf 2005)(Gassner 1997)
Page 6
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation BackgroundBackgroundBackgroundBackground Hypothesis Method Conclusion
Questions
● How do we acquire image schemata about space?
● How can we overcome linguistic and meta-cognitive constraints ?
Page 7
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background HypothesisHypothesisHypothesisHypothesis Method Conclusion
Hypothesis
Spatial concepts (as needed for semantic interoperability) are pre-linguistic
concepts that can be described by a set of axioms.
Are pre-linguistic concepts universal ?
Page 8
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Developmental Psychology
• Theory theory: “Infants learn about the world by forming and revising theories. Their conceptual development is theory formation and change, their semantic development is theory dependent.”
• Development is driven by three forces
– Innate knowledge– Powerful learning abilities– Scaffolding from others
(Meltzoff & Gopnik 2002)
Page 9
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
How to Build a Concept - Revisited
Page 10
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Multi-Tiered Model (Frank 2000)
perceptions
cognitive objects
agents
Page 11
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Multiple Processors – Multiple StrategiesUse Case: Location Coding
● Newcombe and Huttenlocher (2000)
Self-referencedExternally referenced
Simple, limited
Sensorimotor learning (egocentric learning, response learning)
Cue learning
Complex,
powerful
Dead reckoning (inertial navigation)
Place learning
Page 12
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Further Modelling Aspects
● Bottom up vs. Top Down = Symbolic vs. grounded
● Deterministic vs. Stochastic
● Embodied vs. Disembodied
● Single agent vs. multi agent system
● Hierarchic vs. heterarchic
Page 13
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Sandbox Geography
● YES– Classification of Empirical Models– Abstract Models of Empirical Studies– Algebraic Specifications providing set of axioms for spatial relations
between objects– A symbolic mechanism for conceptual change
● NO– Cognitive architecture like SOAR, ACT-R, etc.– Model of the infant, tool for psychology– Artificial intelligence
Page 14
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Type Amount
Support
Page 15
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Theories for the Support of Objects
Page 16
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Sandbox Geography II
1. Expansion2. Contraction3. Revision / Combination4. Analogy
3.
1.
4.
2.
xx
(Sowa 2003)
Page 17
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Reasoning Towards New Theories
TRUERule explains
event
TRUE AND FALSERule explains and
does not explain event
FALSERule does not explain event
TRUE OR FALSERule is not tested
(enough) with eventT/F = 50 %
T > 50 %
T < 50 %
Page 18
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Observations and Expectations
1. T Rule gains weight (store observations to rule made before)
2. F Rule looses weight (store observations to rule made before)
3. FF Rule becomes invalid pool4. TT Rule is valid a gains weight5. TF Verified hypothesis is falsified
Neutral/Action?6. FT Falsified hypothesis is verified
Neutral/Action?7. TTF A chain were verification outweighs
falsification Rule is true8. FFT A chain were falsification outweighs
verification Rule is false
Page 19
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Properties of a Agent Based Framework
● Infant does not give up immediately – certain resistance against theory change (cf. Kuhn 1976)
● Pool of falsified theories (cf. Siegler and Chen 2002)
● An axiom that proofs a theory to be wrong is a case of specialization / generalization Detection of the Non Euclidian Geometry
Page 20
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis MethodMethodMethodMethod Conclusion
Summary
● Formal Specifications for Naive Interoperability have to consider– Experience (History)– Multiple Strategies, i.e. multiple overlapping concepts– Incremental Knowledge Representation– Mechanism to convert Knowledge Representations
● A framework with algebraic specifications suits these needs when embedded in a heterarchy– Contraction– Expansion– Revision– Analogy
Page 21
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis Method ConclusionConclusionConclusionConclusion
Conclusion
● The big question remains how to connect the simple parts of object ontologies.
● We point out some crucial aspects of the modeling process.
● Agent based simulations help to get sound theories in developmental psychology, we need human subject testing on various aspects.
● Psychology serves as an input for new computational models in other disciplines.
Page 22
Florian TwarochInstitute for Geoinformation and Cartography, TU Vienna
[email protected]
Motivation Background Hypothesis Method ConclusionConclusionConclusionConclusion
To be continued …To be continued …