Ontologies6.871: Lecture 22What is an Ontology?• A formal, explicit specification of a shared conceptualization.• A shared vocabulary that can be used to model a domain, i.e., the objects and/or concepts that exist, their properties and relations• Imposition of specific set of conceptualizations on a domain of interest– Tell me about analog electronics– Tell me about digital electronics• Definitions of terminology– and constraints between terms• Domain mini-theoriesOntology vs KB?• Can think of an ontology as a kind of KB, but:• Ontology serves different purpose:– Only needs to describe vocabulary, axioms– E.g. database schema is an ontology•KBincludes:– Specific knowledge needed for problem-solving• Scientific motivation: –Understand fundamental issues about human conceptualizationsMotivations• Engineering motivation:– Every knowledge-based system is based on an ontology of its domain– Explication of the ontology is a time-consuming component of the development process– Why not amortize the effort and share ontologies?• E.g. “core ontologies” such as space, time, quantities• Engineering motivation:• Scientific motivation:Pragmatic Motivations• Responding to the unexpected• Distributed Databases• Distributed Applications• Communicating Agents• Semantic Webdb1db2db3schema1schema2schema3Common OntologyMediatorAnalysisCommon OntologyDesignCommon OntologyApplication 1Application 2Key Question: What does he mean when he says <…>?PersonAspects of an Ontology• Content•Form •Purpose• DevelopmentAspects of an Ontology• Content– types of objects, relationships– e.g. the blocks world conceptualization includes:• Object Classes: Blocks, Robot Hands • Properties: shapes of blocks, color of blocks• Relationships: On, Above, Below, Grasp• Processes: stacking plan for a tower•Form •Purpose• DevelopmentAspects of an Ontology• Content•Form– Is the taxonomic relationship (instance-of, subclass) primary?– Are definitions of, or constraints on, terms provided?– Is the definitional language as rich as a full logic?– Is it process-centric or object-centric?•Purpose• DevelopmentAspects of an Ontology• Content•Form •Purpose– Knowledge sharing• E.g. Between people, software systems, agents– Knowledge reuse • E.g. When models or systems change– General (common sense) or domain specific• DevelopmentAspects of an Ontology• Content•Form •Purpose• Development– Is it acquired or engineered?– If acquired, what about:• Quality of knowledge• Diversity of content• Trust in knowledge• Unpredictable useBuilding an Ontology• Planning• Specification - consider scope and purpose• Knowledge Acquisition• Conceptualization - glossary of terms, top-down, bottom-up, middle-out• Integration - of existing relevant ontologies• Implementation • Evaluation - Clarity, Coherence, Extensibility, Minimal Encoding Bias, Minimal Ontological CommitmentExample Ontologiessee http://www.cs.utexas.edu/users/mfkb/related.html * ARPI Pla n n in g and Sch ed uli n g o n to lo gi e s * Aviat io n O n to lo g y * BPMO - Th e Busin es s P roc es s Ma n ag eme nt On to lo g y * CYC (a n d th e d e riva tive PDKB) * DOLCE - a D e s cri pt ive On to lo g y fo r L in g u is tic a n d C o g ni t ive En gi n e e r ing . * Dublin Core (bib liograph ic organ ization) * Th e Ent e rp ris e Onto log y (for b u s in e s s e n te rpr ise s) * On to lo g ies f or e tho lo g y (a nim a l be h av ior) , e .g . Log g e rh ead T u rt le * FrameNet (lexical reference) * Ge n e ra liz e d Up p er M o de l (for NLP) * Mik roko sm os (fo r NLP) * ON9 (t h e CNR- ITBM On to lo g y Libra ry) * OWL- S - Th e OWL (fo rm erly DAML) Se rvice s on to log y. * On to lin g u a O n to lo g y Libra r y * Op e n Min d datab a s e a nd OM CSNe t S em ant ic N et wor k * PharmGKB - Pha rmacogenetics and Pha rmacogeno mic s Kno wled ge Base * PSL (p ro ces s s pe cifi ca tion) * QoS (computers a nd networks) * SENSUS (fo r NLP) * STEP (for pr o du ct data ex ch an ge ) * SUMO (th e Su gg e s te d Upp e r M e rg ed On to log y) * the T wente Onto logy Collection * UMLS (b io me d icin e) * Wilkin s ' on t o log y (17t h ce nt u ry !) * Word Ne t (lex ica l re fe re n ce )Example Tools for Ontologiessee http://www.cs.utexas.edu/users/mfkb/related.html,http://www.xml.com/pub/a/2002/11/06/ontologies.html * Ch im a e ra * CODE4 * Ge n e ric Kn o wled g e -B a s e Ed itor * Ika ru s * JOE (Java O n to lo g y Edit or) * KAON * KACTUS * OilEd * On toE d it * On tos a u ru s * Prote ge * Sn oba s e * Stan ford On to log y Editor * Sym Ont o s * WordMapProtégéhttp://protege.stanford.edu/Some Large OntologiesPublished OnlineSemantic Network135 Semantic Types, 54 semantic relations, 975,354 conceptsUMLS biomedicinePublished OnlineSemantic Network70,000 termsextension of WordNetSensustext understandingPublished OnlineSemantic Network152,059 word forms in 115,424 synsetsWordNetlexical memoryPublished OnlineKIFAlso LOOM, OWL,Protege1000 terms,4200 assertionsSUMOupper ontologyPartially Online: 6000 Top ConceptsCYCL105concept types, 106 axiomsCYCcommon senseUMLSHeart DiseasesCarcinoid Heart DiseaseEndocarditisMyocardial IschemiaArrhythmiaCoronary DeseaseCoronary AneurismCoronary ArteriosclerosisAngina PectorisAngina Pectoris variantAngina unstableCoronary ThombosisCoronary VasopasmShock CardiogenicMyocardial StunningMyocardial InfarctionFigure by MIT OCW.CYC• Goal: Encode all of human common sense knowledge• Mechanization: human-entered axioms• Periodic review, reorganization, compaction, separation into distinct mini-theories, not mutually consistent• Driven by application domains• Often seems ad-hocCYC Top Level CategoriesThingIndividual ObjectIntangibleRepresented ThingEventIntangibleObjectCollectionStuffIntangibleStuffProcessInternalMachineThingRelationshipSlotAttributeAttributeValueSomethingExistingIntelligenceCompositeTangible&IntangibleObjectTangibleObjectTangibleStuffCYC OntologyCYC ExamplesCYC Examples (cont’d)Perspectives• Philosophy• Library and Information Science• Natural Language Processing• Artificial Intelligence• Semantic WebFredrik Arvidsson, Annika Flycht-ErikssonPerspectives• Philosophy– Objectives: Classify and categorize the world– E.g.: Aristotle … • Library