Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Modular Ontologies - A Formal Investigation of Semantics and Expressivity Jie Bao, Doina Caragea and Vasant G Honavar Artificial Intelligence Research Laboratory, Department of Computer Science, Iowa State University, Ames, IA 50011-1040, USA. {baojie,dcaragea, honavar}@cs.iastate.edu ASWC, Sept 7, 2006, Beijing, China 1 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Outline • • • • Desiderata of Modular Ontologies Abstract Modular Ontology (AMO) Semantics & Expressivity Comparison Summary & Conclusion ASWC, Sept 7, 2006, Beijing, China 2 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Modularity ASWC, Sept 7, 2006, Beijing, China 3 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory A Modular Semantic Web Visualising the Semantic Web by Juan C. Dürsteler ASWC, Sept 7, 2006, Beijing, China 4 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Modular Ontologies • What is modular ontology? – An ontology that is composed by a set of smaller (semantically) connected component ontologies • Why modular ontology ? – – – – – Collaborative Ontology Building Selective Ontology Reuse Selective Knowledge Hiding Distributed Data Management Large Ontology Storage and Reasoning ASWC, Sept 7, 2006, Beijing, China 5 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University OWL Limitations • owl:imports: syntactic modularization • No localized semantics owl:imports – Reasoning is possible only with the integrated ontology • No partial reuse – Reuse all or nothing – E.g. OpenCyc OWL file needs 9 hours to load into Protege ASWC, Sept 7, 2006, Beijing, China Syntactic import: “copy and paste” 6 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Modular Ontology Approaches 1998 2002 CTXML DFOL 2003 2004 2005 2006 C-OWL Role<->Concept Mapping DDL P-DL OWL P-OWL (Planning) E-Connections CЄ(SHIF(D)) ASWC, Sept 7, 2006, Beijing, China Є CIHN (SHOIN(D)) s + 7 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Requirements • Semantic soundness – Reasoning correctness – Module autonomy • Needed language features – Concept relations – Role relations –… ASWC, Sept 7, 2006, Beijing, China 8 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Localized Semantics Integrated ontology Modular ontology Materialized Global Model Local Models ASWC, Sept 7, 2006, Beijing, China 9 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Exact Reasoning Integrated ontology Modular ontology Cv D Cv D ASWC, Sept 7, 2006, Beijing, China 10 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Directional Semantic Relations Dv E Dv E ASWC, Sept 7, 2006, Beijing, China 11 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Transitive Reusability Cv D Dv E Ev F Cv F ASWC, Sept 7, 2006, Beijing, China 12 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Decidability C v D is answerable in finite steps ASWC, Sept 7, 2006, Beijing, China 13 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Language Features • Concept Subsumption • Concept Construction with Foreign Concepts • Concept Construction with Role Restrictions. • Role Inclusion • Role Inversion • Role Construction • Transitive Role • Nominal Correspondence ASWC, Sept 7, 2006, Beijing, China 1:C v 2 : D 1:C u2 : D 9(1 : R):(2 : D ) 1:P v 2 : R 1:P = 2:R¡ 1:P u2 : R Trans(1:P), 1:P used in 2 1:x= 2:y 14 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Outline • • • • Desiderata of Modular Ontologies Abstract Modular Ontology (AMO) Semantics & Expressivity Comparison Conclusion ASWC, Sept 7, 2006, Beijing, China 15 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Local Points of View agents agents The domain Multiple observers of a domain ASWC, Sept 7, 2006, Beijing, China 16 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Abstract Modular Ontology (AMO) Δ1 DL1 Δ2 DL2 2 r13 1 r23 1 r13 DL3 Semantics ASWC, Sept 7, 2006, Beijing, China Δ3 17 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University (General) Domain Relations Δ1 neighbourOf r13 Δ3 friendOf r13 ASWC, Sept 7, 2006, Beijing, China 18 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Image Domain Relation Δ1 r13 Δ3 ASWC, Sept 7, 2006, Beijing, China 19 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Concept Image Legm 1 r13 Agent3: "these objects in my mind state correspond to the concept Leg1! Leg from agent 1’s mind state" ASWC, Sept 7, 2006, Beijing, China 3 ! : r 13 (L egm 1 ) 20 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Role Image Δ1 Δ3 P1! P 3 r13 Agent3: "these object pairs in my mind state correspond to object pairs P from agent 1’s mind state" ASWC, Sept 7, 2006, Beijing, China 21 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Possible AMO Expressivity Features ASWC, Sept 7, 2006, Beijing, China 22 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Semantic Soundness Definitions • Localized Semantics: local domains {Δi} are not necessarily identical • Decidability (of concept C w.r.t AMO O): there is an algorithm to decide in finite steps whether there is a common model <{mi}, {rij}> of C and O. • Directional Semantic Relations: mj ² C m j µ D m j mi ² C i à j µ D i à ASWC, Sept 7, 2006, Beijing, China j 23 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Semantic Soundness Definitions (2) • Reusability C D Cm i µ D m i Ci ! j µ Di! j • Transitive Reusability Δ1 Δ2 Δ3 (an agent can infer local constraints based on observing constraints in other agents’ points of view) ASWC, Sept 7, 2006, Beijing, China 24 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Semantic Soundness Definitions (3) • Exact Reasoning – Compatible beliefs of agents may be combined. – Local models M can be merged into an integrated model M' s.t. Physical World Local Models Integrated Model (consensus) ASWC, Sept 7, 2006, Beijing, China 25 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Outline • • • • Desiderata of Modular Ontologies Abstract Modular Ontology (AMO) Semantics & Expressivity Comparison Conclusion ASWC, Sept 7, 2006, Beijing, China 26 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University DDL Semantics [Borgida and L. Serafini, 2002] (R.D)I DI CI - r(CI)= (R .C)I 1:Dog into 2:Animal DDL : C into D r(CI) DI - r(CI)= (R .C)I CI DI C onto D 1:Dog onto 2:Hound r(CI) DI implicit domain disjointness ASWC, Sept 7, 2006, Beijing, China 27 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Subsumption Propagation Problem Cm1 Dm2 C into D C into E ? D into E Em3 DDL domain relations are not transitively reusable ASWC, Sept 7, 2006, Beijing, China 28 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Inter-module Unsatisfiability Problem Flym1 Bird onto Penguin Penguinm2 Birdm1 ~Fly onto Penguin DDL allows arbitrary domain relations: loss of reasoning exactness ASWC, Sept 7, 2006, Beijing, China 29 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University DDL: Pros & Cons • Pros – Localized Semantics – Directional Relation – Decidability transfer • Cons – No support for role relations – No general module transitive reusability – No general reasoning exactness ASWC, Sept 7, 2006, Beijing, China 30 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University E-Connections • Local domains are disjoint • It allows multiple “link” relations between two local domains • Links can be used to construct local concepts [Grau, 2005] owns Pet PetOwner ASWC, Sept 7, 2006, Beijing, China 31 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University E-Connections Semantics - (R.D)I = r (DI) (R.D)I=Δ1\r-(Δ2\DI) R DI R DI ASWC, Sept 7, 2006, Beijing, China 32 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory E-Connections: Pros & Cons • Pros – Localized Semantics – Decidability Transfer – Exact Reasoning (without generalized link) • Cons – Very limited transitive reusability – No support for inter-module concept subsumption ASWC, Sept 7, 2006, Beijing, China 33 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University P-DL [Bao et al., 2006] • Semantic Importing O1 (General Animal) ASWC, Sept 7, 2006, Beijing, China O2 (Pet) 34 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University P-DL Semantics ΔI • Domain relation: individual correspondence between local domains • Importing establishes oneto-one domain relations – “Copied” individuals are shared • Domain relations are compositionally consistent: r13=r12 r23 O ΔI 1 r12 2 x’ x C I2 C I1 r13 r23 x’’ C I3 – Therefore domain relations are transitively reusable. ΔI 3 35 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University P-DL Semantics (2) ΔI ΔI 1 r12 2 x’ x C I2 C I1 x r13 r23 x’’ ΔI C I3 CI Global model obtained from local models by merging shared individuals 3 36 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Partially Overlapping Domains • Ensure unambiguous communication between local models – satisfiability transfer – transitive reusability • Overlapped domains represent the consensus of agents • Non-sharing domains are still kept local. x CI 37 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University P-DL: Pros & Cons • Pros – Localized Semantics – Exact Reasoning – Stronger Expressivity – Transitive Reusability • Cons – Directional Semantic Relation does not always hold – Decidable only if all modules use the same decidable DL (e.g. OWL). ASWC, Sept 7, 2006, Beijing, China 38 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Summary: Semantic Soundness ASWC, Sept 7, 2006, Beijing, China 39 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Summary: Expressivity ASWC, Sept 7, 2006, Beijing, China 40 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Outline • • • • Desiderata of Modular Ontologies Abstract Modular Ontology (AMO) Semantics & Expressvity Comparison Conclusion ASWC, Sept 7, 2006, Beijing, China 41 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Summary • We discussed – Semantic soundness and expressvity requirements for modular ontologies – Comparsion of DDL, E-Connections and P-DL under the AMO framework – Analysis of several semantic difficulties and expressivity limitations of existing approaches ASWC, Sept 7, 2006, Beijing, China 42 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Conclusions • There is still no language or reasoner support for both general inter-module concept and intermodule role correspondence • Local domain disjointedness assumption of DDL and E-Connections may be partially relaxed. – to improve expressivity – to ensure reasoning exactness and transitivity reusability. ASWC, Sept 7, 2006, Beijing, China 43 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Open Problems • A consensus on expressive modular ontology language • A OWL-compatible syntax for modular ontologies • A reasoner that supports the expressive modular ontology language – Pellet and DRAGO are complementary to each other To be discussed at the Modular Ontology Workshop (WoMo) at ISWC 2006 , Athens, Georgia, USA, Nov 2006. ASWC, Sept 7, 2006, Beijing, China 44 Iowa State University Department of Computer Science Artificial Intelligence Research Laboratory Thanks! ASWC, Sept 7, 2006, Beijing, China 45 Department of Computer Science Artificial Intelligence Research Laboratory Iowa State University Semantic Soundness What are the logical consequences in an AMO? a b r13 y x friendOf friendOf ? What are the possible cause of semantic inconsistencies between two agents? a b r13 friendOf y x enemyOf What is the "objective" way to integrate knowledge of agents? a/x b/y friendOf ASWC, Sept 7, 2006, Beijing, China 46