1. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies1 Natural Ontologies : Understanding and Addressing Alignment Issues Kurt Conrad The 9th International Protégé Conference July 23 rd 2006, Stanford, California

2. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies2 Introduction What is the Ontology Management Team? What is the Ontology Management Team? Issue: Semantics are Naturally Dynamic. Issue: Semantics are Naturally Dynamic. Approach: Analyze Transformational Origins of Semantic Changes Approach: Analyze Transformational Origins of Semantic Changes Strategy: Formalize MetaKnowledge Strategy: Formalize MetaKnowledge Approach: Analyze Impact of Human Values on Semantic Association Approach: Analyze Impact of Human Values on Semantic Association Method: Semantic Consensus Acceleration Method: Semantic Consensus Acceleration

3. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies3 The Ontology Management Team OMT is a spin-off of the Ontolog Forum OMT is a spin-off of the Ontolog Forum Particpants Particpants Kurt Conrad Conrad@SagebrushGroup.com Kurt Conrad Conrad@SagebrushGroup.com Conrad@SagebrushGroup.com (408) 247-0454 (408) 247-0454 Bob Smith Bob@1TallTrees.com Bob Smith Bob@1TallTrees.com Bob@1TallTrees.com (714) 536-1084 (714) 536-1084 Bo Newman Bo.Newman@KM-Forum.org Bo Newman Bo.Newman@KM-Forum.org Bo.Newman@KM-Forum.org (540) 459-9592 (540) 459-9592 Joe Beck Joe.Beck@EKU.edu Joe Beck Joe.Beck@EKU.edu Joe.Beck@EKU.edu (859) 622-6359 (859) 622-6359

4. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies4 OMT Mission and Focus Clearly differentiate Ontology Management and Engineering activities. Clearly differentiate Ontology Management and Engineering activities. –Understand factors driving effective management of ontology engineering projects. Bridge theoretical, problem, and engineering domains Bridge theoretical, problem, and engineering domains –Produce improved methodologies for ontology development. –Develop reliable methods for driving (natural) ontological alignment within working groups. –Identify and develop methods to ensure quality and alignment of needed conceptual models. Leverage new insights and methodologies to deal with more general issues of policy development within large enterprises. Leverage new insights and methodologies to deal with more general issues of policy development within large enterprises.

5. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies5 OMT Policy Vector #1 Issue: Implicit policymaking by technologists. Issue: Implicit policymaking by technologists. –Managements abdication of policymaking responsibilities. –Occurs when the policy implications of design decisions are poorly understood. –Risk increases when dealing with emerging technologies. Strategy: Make policy decisions transparent. Strategy: Make policy decisions transparent. –Identify critical decisions that have significant downstream policy impacts. –Develop analytical and governmental processes to enable understanding and resolution of associated policy issues by the appropriate stakeholders.

6. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies6 OMT Policy Vector #2 Issue: Disruptive impact of incomplete consensus and Dynamic Semantics on ontology engineering efforts. Issue: Disruptive impact of incomplete consensus and Dynamic Semantics on ontology engineering efforts. Strategy: Expand ontology development methodologies to deal effectively with Dynamic Semantics. Strategy: Expand ontology development methodologies to deal effectively with Dynamic Semantics. –Understand the role of natural ontologies in sensemaking. –Develop methods to assess semantic distance, volatility, and drift. –Develop procedures for negotiating and narrowing-the-gap when semantic conflicts are encountered. –Develop a language for defining semantic policy that is usable by both policy makers and ontological engineers.

7. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies7 OMT Policy Vector #3 Issue: Other areas of organizational behavior are increasingly experiencing the alignment issues typically associated with ontological engineering initiatives. Issue: Other areas of organizational behavior are increasingly experiencing the alignment issues typically associated with ontological engineering initiatives. Strategy: Generalize ontology management practices to: Strategy: Generalize ontology management practices to: –Deal with broader issues of organizational meaning. –Resolve semantic issues in other policymaking domains. –Balance and prioritize semantic alignment efforts across initiatives. –Establish benchmarks for semantic accountability.

8. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies8 Issue: Semantics are Naturally Dynamic

9. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies9 Semantics are Naturally Dynamic Initial uncertainties and inevitable semantic changes undermine the alignment of formalization efforts. Initial uncertainties and inevitable semantic changes undermine the alignment of formalization efforts. Dynamic Semantics (DS) results from the interplay of three agent types and their associated ontologies Dynamic Semantics (DS) results from the interplay of three agent types and their associated ontologies –Individual Agents –Social Agents –Automated Agents

10. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies10 Formalization Doesnt Stabilize Semantics Meaning isnt an inherent property Meaning isnt an inherent property –Ultimately the product of human imagination and creativity Complex set of mechanisms both drive and limit changes to perceived meaning Complex set of mechanisms both drive and limit changes to perceived meaning DS ultimately impact automated systems DS ultimately impact automated systems –From individuals (changes to conceptualization) –From groups (dynamic / evolving consensus) Making semantics explicit doesnt necessarily limit or slow upstream change Making semantics explicit doesnt necessarily limit or slow upstream change

11. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies11 Approach: Analyze Transformational Origins of Semantic Changes

12. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies12 Analyzing Dynamic Semantics While analysis of agents and agent types can be used to identify DS, it doesnt provide enough detail to drive specific responses While analysis of agents and agent types can be used to identify DS, it doesnt provide enough detail to drive specific responses Need more sophisticated models that let us look beyond agent types to specific Semantic Classes and properties Need more sophisticated models that let us look beyond agent types to specific Semantic Classes and properties –Knowledge Vector Model –MetaKnowledge Continuum Model

13. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies13 Knowledge Vector Model D ƒ I ƒ K KU Describes a continuum of Knowledge Artifacts Describes a continuum of Knowledge Artifacts –Data (D) is transformed (ƒ) into Information –Information (I) is transformed into Knowledge –Knowledge (K) represents the point of actionable synthesis of all event-specific (K E ) and prior (K P ) knowledge –K enables a Knowledge Utilization Event (KU) –KU is either an action or a decision Boundaries are not necessarily discrete Boundaries are not necessarily discrete

14. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies14 MetaKnowledge Continuum Model Also describes a continuum without discrete boundaries Also describes a continuum without discrete boundaries MetaKnowledge (MK) comprises a range of KAs MetaKnowledge (MK) comprises a range of KAs –Like the generic term MetaData –The term MetaKnowledge is used deliberately mD, mI, and mK (specific to this model) document mD, mI, and mK (specific to this model) document –Critical semantic properties of each class of KA –Knowledge about the transformations that produced them D ƒ I ƒ K KU mDmImK

15. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies15 Interpretive Semantics Deal with the interpretation and meaning of symbols Deal with the interpretation and meaning of symbols Knowledge about how symbols map to concepts Knowledge about how symbols map to concepts –Potentially ambiguous (multiple meanings, double entendres) –Result of observational and symbol selection behaviors Answers the question: What is it? Answers the question: What is it? Maps to mD in the MK Continuum Model Maps to mD in the MK Continuum Model

16. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies16 Contextual Semantics Deal with context and pattern recognition Deal with context and pattern recognition Knowledge about how KAs (or parts) relate to Knowledge about how KAs (or parts) relate to –Transformation and representation behaviors –Other KAs –The real world Answers the questions Answers the questions –What kind? What is it about? –Who, where, when? –How, especially How does this fit? Maps to mI in the MK Continuum Model Maps to mI in the MK Continuum Model

17. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies17 Contextual Semantics Principles Appears to be the first place that values play a significant role in sense-making Appears to be the first place that values play a significant role in sense-making Meaning often expressed in historical terms that imply future semantics Meaning often expressed in historical terms that imply future semantics –You always.. –They always… –It always… Contextual Semantics often implied, incomplete, or missing Contextual Semantics often implied, incomplete, or missing –Supplied by the interpreting agent

18. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies18 Aspirational Semantics Deals with underlying motivations, drivers, rationality Deals with underlying motivations, drivers, rationality Knowledge about how KAs are synthesized and optimized to enable behavior Knowledge about how KAs are synthesized and optimized to enable behavior Answers the question Why? Answers the question Why? –Infrequently documented –Often tacit and implicit Individuals provide, when missing Individuals provide, when missing –Routine source of semantic breakdowns Maps to mK in the MK Continuum Model Maps to mK in the MK Continuum Model

19. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies19 Behavioral Semantics Meaning, as described in behavioral terms Meaning, as described in behavioral terms Often involves complex semantic chains (scenarios) which comprise Often involves complex semantic chains (scenarios) which comprise –Events –Conditions –Other behaviors Representations range from tacit to explicit Representations range from tacit to explicit –Culture –Law

20. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies20 Strategy: Formalize MetaKnowledge

21. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies21 Addressing Dynamic Semantics Ignore semantic volatility Ignore semantic volatility –Leave critical semantic properties largely implicit and tacit –Appropriate for non-critical issues Stomp Stomp –Make semantic properties more explicit –Formalize without addressing sources of volatility –Good when you can get away with it Embrace Embrace –Understand change vectors –Balance explicit, implicit, and tacit representations –Implement mechanisms to identify and/or leverage natural misalignments as they emerge

22. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies22 Semantic Nirvana / Artificial Utopia Formalization is optimized for computerized inference Formalization is optimized for computerized inference –First Order Logic typically considered the most expressive representation Two-step semantic resolution model Two-step semantic resolution model –Symbol interpretation Maps symbol to concept –Axiomatic component Used to document, communicate, and potentially infer behavioral implications Limitations Limitations –Limited Contextual and Aspirational Semantics –Practical issues limit expressiveness »Availability of subject matter experts »Truth is unbounded but resources are limited »Can you read KIF?

23. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies23 Dynamic Semantic Findings Formalizing meaning in the face of uncertainty is Formalizing meaning in the face of uncertainty is –Difficult, at best –Potentially chaotic Therefore, an improved ability to identify and understand Dynamic Semantics enables Therefore, an improved ability to identify and understand Dynamic Semantics enables –More resiliency to be built in, in the first place –Less remediation. Fewer false starts –Changes to be more easily anticipated and reacted to –Semantic change to be used as a resource for enhancing delivered value

24. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies24 MetaKnowledge Formalization Strategy Identify areas of potential semantic conflict Identify areas of potential semantic conflict –Resolve conflict, as appropriate –Integrate semantic models, where each has value Avoid sub-optimization around machine-processable semantics Avoid sub-optimization around machine-processable semantics –Evaluate each Semantic Class for potential value –Consciously balance or optimize the explicit representations –Document implicit and tacit MK triggers that are only usable by individuals and groups Where practical, expand the range of targeted KUs and associated behavioral semantics Where practical, expand the range of targeted KUs and associated behavioral semantics

25. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies25 MetaKnowledge Formalization Benefits Doesnt have to be difficult or expensive Doesnt have to be difficult or expensive Supports and enables rapid prototyping Supports and enables rapid prototyping –Scalable from small projects to large knowledge architectures –Reporting model starts right and improves through time Leverages implicit and tacit knowledge within the organization Leverages implicit and tacit knowledge within the organization –Enables re-contextualization and Knowledge Perpetuation –Less brittle than other approaches Doesnt preclude use of logic-based formalizations Doesnt preclude use of logic-based formalizations –Can speed and document emergent consensus –Helps ensure alignment of human behavior with axioms Expected to be the foundation for many emerging Ontology Management practices Expected to be the foundation for many emerging Ontology Management practices

26. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies26 Approach: Analyze Impact of Human Values on Semantic Association

27. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies27 Decisionmaking is Expensive Relies on a series of complex and sophisticated activities: Relies on a series of complex and sophisticated activities: –Data collection and interpretation. –Organization and pattern recognition. –Identification of motivations, causalities, and implications. –Synthesize actionable knowledge. Evolutionary advantages to be gained by reducing cost (time and effort) of decisionmaking. Evolutionary advantages to be gained by reducing cost (time and effort) of decisionmaking. –e.g., OODA Loops.

28. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies28 Values Cut Decisionmaking Costs… Usually abstracted (decontextualized) in order to be applied across a variety of behavioral contexts. Usually abstracted (decontextualized) in order to be applied across a variety of behavioral contexts. Often comprise significant implicit and tacit knowledge components. Often comprise significant implicit and tacit knowledge components. –Including truly tacit knowledge that defies articulation. Represents a form of bounded rationality. Represents a form of bounded rationality. –Doesnt just impact the amount of knowledge used in decisionmaking (sufficiency). –Impacts pattern recognition and other transformative behaviors. Enables much of the decisionmaking process to remain implicit and tacit, operating at a subconscious level. Enables much of the decisionmaking process to remain implicit and tacit, operating at a subconscious level. Values act as a technology. Values act as a technology.

29. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies29 …But at the Risk of Poor Decisions Values can drive sub-optimal decisions. Values can drive sub-optimal decisions. –Limit perception of contrary and mitigating evidence. –Reinforce understood and accepted interpretations. –Constrain associated semantics to those consistent with established behavioral patterns. Risk increases with changes across or within specific behavioral contexts. Risk increases with changes across or within specific behavioral contexts. Risk increases when the impact of values on decisionmaking goes unrecognized. Risk increases when the impact of values on decisionmaking goes unrecognized.

30. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies30 Potential Organizational Issues When unaccounted for, implicit and tacit values in systems can drive: When unaccounted for, implicit and tacit values in systems can drive: –Semantic breakdowns. –Polarization and conflict. –Group think. »Perpetuation of hidden biases. »Missed opportunities. »Inability to perceive risks.

31. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies31 Knowledge Vector Model D ƒ I ƒ K KU Provides basis for understanding the origin of values and the impact of values on semantic association. Provides basis for understanding the origin of values and the impact of values on semantic association. Agents use knowledge to execute behaviors. Agents use knowledge to execute behaviors. –Automated agents require all elements of the Knowledge Vector Model (KA, ƒ, & KU) to be explicit. –Individuals and organizations can leverage implicit and tacit knowledge elements to skip steps. People have a strong incentive to skip steps to make things easier. People have a strong incentive to skip steps to make things easier. The scientific method does not come naturally. The scientific method does not come naturally.

32. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies32 Impact of Knowledge on Values

33. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies33 Values Values (including Principles, Interests, and Expectations) are KAs that represent networks of prior knowledge. Values (including Principles, Interests, and Expectations) are KAs that represent networks of prior knowledge. Most importantly, values are routinely disassociated with major portions of the originating knowledge network. Most importantly, values are routinely disassociated with major portions of the originating knowledge network. Disassociation is a form of knowledge compression. Disassociation is a form of knowledge compression. –Reduces processing time, communications time, recall time. –Allows values to be used as abstract KAs that can be applied in a variety of behavioral contexts. Wisdom relates to judgment and the ability to work intelligently with multiple sets of values. Wisdom relates to judgment and the ability to work intelligently with multiple sets of values. –Conflict avoidance, conflict resolution. –Balance short-term and long-term perspectives. –Avoid unary values-based, sub-optimal decisionmaking.

34. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies34 Semantic Segmentation Model Sensemaking is a multi-step process. Sensemaking is a multi-step process. Knowledge Vector elements are associated with a distinct classes of semantic properties. Knowledge Vector elements are associated with a distinct classes of semantic properties. –Interpretive semantics. –Contextual semantics. –Aspirational semantics. –Behavioral semantics. Values directly impact the association of increasingly sophisticated meanings throughout the Knowledge Vector. Values directly impact the association of increasingly sophisticated meanings throughout the Knowledge Vector.

35. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies35 Values Impact Semantic Associations Perception. Perception. –What KA are considered irrelevant, noisy? –What KA are considered important, valuable, potentially useful? Interpretation. Interpretation. –What concepts are associated with inputs / symbols? –What referents are associated with the concepts? –How is ambiguity to be resolved?

36. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies36 Values Impact Semantic Associations Contextualization. Contextualization. –Which context / sensemaking structure is most appropriate? –How does this KA fit? How should it be positioned? –What rules should be used to organize and relate KAs? –What patterns emerge? Are they useful or distractions? –What can be inferred? What implicit K is relevant / important? –Are the sources credible? Has this K proved its value in the past? –Is the K applicable at this time, in this situation? –Is this consistent with history and trends?

37. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies37 Values Impact Semantic Associations Aspirations. Aspirations. –How do the KAs relate to goals and objectives? –What are the motivations, intentions, and strategies of the relevant agents? –Do the agents support my interests or are they in conflict? –How does potential conflict influence K acquired from those sources? –Is the K actionable, or do critical K gaps exist? Behaviors. Behaviors. –What casual models apply? –What behaviors / results can be expected? –What are the risks and probabilities? –What alternatives and contingencies exist?

38. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies38 Ambiguous Semantics of Values

39. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies39 Conclusions Values, not facts, drive decisions. Values, not facts, drive decisions. Values may represent the ultimate semantic technology. Values may represent the ultimate semantic technology. Values-based analysis methods: Values-based analysis methods: –Facilitate better ontological alignment across individuals and organizations. –Improve the quality of policymaking. –Stabilize organizational context, requirements, and specifications for engineering efforts. –Improve the quality of semantic formalization and articulation.

40. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies40 On the Horizon Values-based ontology development process. Values-based ontology development process. Introduction of values-based logic to complement existing axiomatic models of semantic formalization. Introduction of values-based logic to complement existing axiomatic models of semantic formalization. Extend values-based conceptual alignment methods to address a broad range of policy development and governance issues. Extend values-based conceptual alignment methods to address a broad range of policy development and governance issues.

41. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies41 Method: Semantic Consensus Acceleration (SCA)

42. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies42 SCA Origins Methodology developed to address a range of issues and initiatives. Methodology developed to address a range of issues and initiatives. –Metadata framework to improve semantic interoperability. –Web search, retrieval, and navigation. Cross-cultural teams Cross-cultural teams –Comprised marketing, engineering, technical documentation personnel. –Leveraged multiple value systems to illuminate semantic conflicts. –Given a Semantic Workpackage, comprising a set of related terms. –Identified and differentiated individual concepts and key relationships (including organizational scope). –Documented the level of consensus and any outstanding knowledge gaps. –Delivered conceptual requirements and specifications to engineering team.

43. © 2006 Kurt ConradNatural Ontologies: Alignment Issues & Strategies43 SCA Process Step 1: Research Terms Step 1: Research Terms –List any additional terms and sources Step 2: Inventory Semantics Step 2: Inventory Semantics –List concepts and draft / identify working definitions –Identify core concept and document supporting rationale –Map related concepts & relationships to core, document behavioral relevance Step 3: Normalize Semantics Step 3: Normalize Semantics –Refine & validate working definitions to produce candidate definitions –Document supporting rationales Step 4: Normalize Terminology Step 4: Normalize Terminology –Select normative term for each concept –Document terms origins and history –Document any alternative semantics –List proper and improper aliases