onto

ONTOLOGIES and THE SEMANTIC WEB: 

ONTOLOGIES and THE SEMANTIC WEB Guido Boella Dipartimento di Informatica Università di Torino

Disclaimer…: 

Disclaimer… Barry Smith: “the Semantic web is going to die in 6 months” Peter Patel Schneider: “it is not fair, ontology is there since 2000 years”

PLAN OF THE LESSON: 

PLAN OF THE LESSON What are ontologies for Existing ontologies WordNet Cyc SUMO DOLCE

Ontology: 

Ontology A discipline of Philosophy Meta-physics dates back to Aristotle Ontology dates back to 17th century • The science of what is (“being qua being”) • The study of what is possible • The study of the nature and structure of possibilia

What are ontologies : 

What are ontologies Philosophy: field of metaphysics which studies how is the world really made behind what appears to be Computer Science: Field of Artificial Intelligence which studies methodologies for knowledge representation Why in CS? Communication: among software agents and among agents and humans Knowledge sharing: reuse of resources for developing computer software Semantic Web!

Slide6: 

Ontology in Philosophy Each special science aims at truth, seeking to portray accurately some part of reality … No special science can arrogate to itself the task of rendering mutually consistent the various partial portraits: that task can alone belong to an overarching science of being, that is, to ontology. But … the proper concern of ontology is not the portraits we construct of it, but reality itself (Lowe 2001). Can we know reality as it is? Entity vs its representation

Ontology in CS: 

Ontology in CS • A specific artifact designed with the purpose of expressing the intended meaning of a (shared) vocabulary •A shared vocabulary plus a specification ( characterization) of its intended meaning “An ontology is a specification of a conceptualization” [Gruber 95] ...i.e., an ontology accounts for the commitment of a language to a certain conceptualization

Slide8: 

What are the fundamental components (primitives: they cannot be reduced to other ones) of reality? The basic question Eg.: Particolars (instances: eg. a rose) Universals (classes: eg. roses) Tropes (a particular value: the red of this rose) Properties (values of parameters: the color red) Analogies with DB Particolars: rows of a table Universali: tables (ie their definition) Tropes: the content of a field (attribute) Properties: the type of a field (eg. ‘integer’)

Ontologies for the Semantic Web: 

Ontologies for the Semantic Web Ontologies are the basic infrastructure for the Semantic Web. Semantic Web assumes the possibility to use shared vocabularies for describing resource content and capabilities, Their semantics is described in a (reasonably) unambiguous and machine-processable form. Describing this semantics, the intended meaning of vocabulary terms, is exactly the job ontologies

Applications for the SemWeb: 

Applications for the SemWeb Semantic Interoperability – Generalized database integration – Virtual Enterprises e-commerce - Information Retrieval – Decoupling user vocabulary from data vocabulary – Query answering over document sets – Natural Language Processing

But what kinds of ontologies do we need?: 

But what kinds of ontologies do we need? Upper level ontologies? Lightweight ontologies? (the minimal terminological structure, a taxonomy) Answer: different uses within the SemWeb Semantic access to a specific resource; the ontology describes structural relationships among terms which are relevant for the query. Meaning negotiation: for cooperation between multiple artificial agents: it requires the explicit representation of ontological commitment in terms of a rich axiomatization. Knowledge sharing: reuse of resources

Different uses: 

Different uses Application ontologies ( run time) – offer terminological services, checking constraints between terms – limited expressivity (stringent computational reqs.) • Reference ontologies ( develop. time) – establish consensus about meaning of terms (in general) – higher expressivity (less stringent computational reqs) - Mutual understanding more important than mass interoperability – understanding disagreements – establish trustable mappings among application ontologies

Formal Ontology: 

Formal Ontology • Theory of formal distinctions and connections within: – entities of the world, as we perceive it ( particulars) – categories we use to talk about such entities ( universals) • Why formal? – Two meanings: rigorous and general – Formal logic: connections between truths - neutral wrt truth – Formal ontology: connections between things - neutral wrt reality • Goal: characterizing particulars and universals by means of formal properties and relations.

Axiomatic ontologies: 

Axiomatic ontologies The axiomatization’s purpose is to exclude terminological and conceptual ambiguities Hard task (both conceptually and computationally), but it only needs to be undertaken once, before a cooperation process starts. The quality of a meaning negotiation process affects the trust in a service offered by the Semantic Web,

Degrees of ontologies: 

Degrees of ontologies

Knowledge representation vs ontology: 

Knowledge representation vs ontology Ontological truths vs. epistemic truths • Ontological knowledge holds necessarily! Not facts like in a DB • The semantics of generalization needs to be refined – All the telephones are artifacts – All the telephones are black [Woods 75, What’s in a link]

Ontologies vs. Conceptual Schemas: 

Ontologies vs. Conceptual Schemas • Conceptual schemas – Often not accessible at run time – Usually no formal semantics – attribute values taken out of the UoD – constraints relevant for database update • Ontologies – Usually accessible at run time – formal semantics – attribute values first-class citizens – constraints relevant for intended meaning

Slide18: 

Un esempio: Trope theory Esistono solo i Tropi (come categoria fondamentale) I particolari non sono altro che insiemi di tropi (le caratteristiche di un certo individuo) L’analogia con i DB Esistono solo i campi delle tabelle (solo essi occupano memoria, cioè spazio fisico nel DB) Le righe esistono solo come agglomerati di campi, e non sono quindi indipendenti da essi Le definizioni delle tabelle (e dei vincoli sugli attributi) stanno fuori dal DB vero e proprio. Sono cioè non il DB (la realtà), ma una descrizione ‘astratta’ delle sue caratteristiche N.B. Come i tropi, anche i campi non possono esistere senza la loro riga Sia gli universali che le proprietà non esistono autonomamente, ma solo come ‘astrazione mentale’ (dei particolari e dei tropi, rispettivamente) che noi facciamo per organizzare la realtà

Slide19: 

Un esempio di dibattito Opponent: Perchè, se i tropi sono ‘primitivi’ non possono staccarsi dal loro oggetto? Meglio pensare che le proprietà siano ‘modi di essere’ di oggetti (questi ultimi primitivi) Defender: Ma cosa c’è allora in un ‘particolare’ più dei tropi che lo caratterizzano? Opponent: C’è un ingrediente supplementare, un ‘substrato’ che tiene insieme l’oggetto Defender: Ma cos’è, ontologicamente, questo substrato? Un altro tropo? O un particolare? Opponent: ….

Slide20: 

Example problems What are facts (eg States (Cris is blonde) and Events (Cris went away))? What is existence? Do unicorns exist? does a cancelled or planned strike exist? Many alternatives For example: all 4 basic categories (particolars, universals, tropes and properties) are primitive

How to share knowledge : 

KB1 Diagnostic system for heart deases Structureof aortic valve KB2 System for scheduling surgical operations Structureof aortic valve How to share knowledge Ontology in Computer Science

Solution : 

KB1 Structureof aortic valve KB2 Solution How is it possible Diagnostic system for heart deases System for scheduling surgical operations

Communication: 

Communication ma: Starting town := Workplace; Arrival town := Meeting;

PLAN OF THE LESSON: 

PLAN OF THE LESSON What are ontologies for Existing ontologies WordNet Cyc SUMO DOLCE

WordNet : 

WordNet From lists of words: <tiger, dog, animal, mammal, beast, cat, persian, felinus> To structured dictionaries: NB: the same word can belong to more than one SynSet

Slide26: 

WordNet: Cognitive Science Laboratory of University of Princeton (english) end of ’80s. (http://www.cogsci.princeton.edu/~wn/index.shtml) EuroWordNet: EU projects. (multilingual - ILC-Pisa for italian) ’90s ItalWordNet: IRST-ICT (Trento). end of ‘90 Different versions and many others…

Slide27: 

WordNet relations (original)

Note that : 

Vagueness of some relations POS means Part Of Speech (noun, verb, etc.) different POS are linked in limited cases Note that Relations are unrelated to synsets

Slide29: 

Le relazioni di EuroWordNet

Slide30: 

Per molte relazioni sono definite anche le inverse, che per semplicità non ho riportato in tabella (ad es. iponimia  iperonimia; meronimia  olonimia; causa  causato_da; …) Alcune relazioni non sono definite tra Synset, ma tra singole parole. Questo vale ovviamente per la sinonimia, ma anche per la derivazione e per l’antonimia La tabella è molto più estesa della precedente; non riporto la tabella delle relazioni di ItalWordNet, che comprende 45 righe, al posto delle 18 che ci sono sopra Tutti i Synset coinvolti si riferiscono a classi (chitarra, andare, …) eccetto quelli che compaiono nell’ultima relazione, in cui uno dei due elementi collegati è un’istanza (Po, Roma) La seconda colonna è molto diversa da quella della tabella precedente. I numeri (1, 2, 3) si riferiscono ai cosiddetti “ordini semantici”, così definiti: 1: nomi concreti 2: nomi, verbi, aggettivi o avverbi indicanti proprietà, stati, processi o eventi 3: nomi astratti indicanti proposizioni indipendenti dal tempo e dallo spazio Osservazioni

Slide31: 

EuroWordNet Architecture

Slide32: 

Inter Lingual Index (ILI) is a mapping among Synsets, without structure Top-ontology is a structured representation of more general concepts (see below) Domain-Ontologies are lists of Semantic fields, i.e., arguments (e.g. Sport, football, …) Three types of links I: links independent from languages: related ILI with top and domain ontologies II: links relating synsets of the national WordNet to ILI III: language dependent links relating synsets of a language (see above list) Note that

Slide33: 

Top-Ontology

Slide34: 

Concepts are defined by features First order entities: Origin (natural or artificial?; O index in the figure) Natural Living Plant Human Creature Animal Artifact Form (substance or object with a form? F index) Substance Solid Liquid Gas Object Composition (whole or group?) Part Group Function (function, U index) Vehicle Representation MoneyRepresentation LanguageRepresentation ImageRepresentation Software Place Occupation Instrument Garment Furniture Covering Container Comestible Building

Slide35: 

For second order entities: Situation Component (feature of the situation described, C index) Cause Communication Condition Existence Experience Location Manner Mental Modal Physical Possession Purpose Quantity Social Time Situation Type (type of situation; T index) Dynamic BoundedEvent UnboundedEvent Static Property Relation Third order comes with no features

Slide36: 

Note that More specific concepts defined by bundles of features (e.g., containerU+objectF+artifactO ) Top-level concepts are not SynSets! e.g., Container belongs both to the top ontology and to some SynSet Top-ontology links 1310 Base Concepts, common to all languages, selected on: number of associated relations place in the taxonomy frequency in corpora No inferential mechanism using the relation ItalWordNet è stored in relational DB Only graphic browsers

PLAN OF THE LESSON: 

PLAN OF THE LESSON What are ontologies for Existing ontologies WordNet Cyc SUMO DOLCE

Slide38: 

The Cyc project (from enCYClopedia) since 1984 (http://www.opencyc.org/). Now it includes 1million concepts, but the public version OpenCyc includes only 6.000 concepts and 60.000 relations “So, the mattress in the road to AI is lack of knowledge, and the anti-mattress is knowledge. But how much does a program need to know to begin with? The annoying, inelegant, but apparently true answer is: a non-trivial fraction of consensus reality - the millions of things that we all know and that we assume everyone else knows” (Guha & Lenat 90, p.4) Cyc

Slide39: 

2 components

Slide40: 

#$Texas #$capital: (#$Austin) #$residents: (#$Doug #$Guha #$Mary) #$stateOf: (#$UnitedStatesOfAmerica) CycL Units i.e., Frames with slots Example of Unit of an instance #$ prefix designates Units. Slots are Units too (SlotUnits)

Slide41: 

#$residents #$instanceOf: (#$Slot) #$inverse: (#$residentOf) #$makesSenseFor: (#$GeopoliticalRegion) #$entryIsA: (#$Person) #$specSlots: (#$lifelongResidents #$illegalAliens #$registeredVoters) Example of Slot Unit slots are just binary relations They have domain (#$makesSenseFor) and (#$entryIsa) Relations on relations (#$inverse e #$specSlots)

Slide42: 

The Constraint Language ‘restricted quantification’ for predicate logic Eg every person has a mother and the difference in age is 16

Slide43: 

#$Person #$genls: (#$Living) #$name: (#$PersonName) #$residentOf: (#$city) #$mother: (#$Person) #$inheritedSlotConstraints: (#$AgeOfMotherConstraint)   #$AgeOfMotherConstraint #$instanceOf: (#$SlotConstraint) #$constraintInheritedTo: (#$Person …) #$slotsConstrained: (#$mother) #$slotConstraints: (#$GreaterThan (#$Diff (v #$age) (u #$age)) 16))))) Efficiency issues: Separate slot definition from the slot constraint: ie Constraint Language

Slide44: 

#$mother #$instanceOf: (#$Slot) #$inverse: (#$motherOf) #$makesSenseFor: (#$Person) #$entryIsA: (#$Person) #$entryFormat: (SingleEntry) Definition of slot #$mother: Why more efficient Constraint expressed in CycL, specialized language for constraint

Slide45: 

Inference in Cyc (Put #$Giorgio #$mother #$Lucia) do operation: Store (Put) in slot #$mother of Unit #$Giorgio, the value #$Lucia ask: (Get #$Lucia #$mother-of) Ask (Get) who is mother of (#$mother-of) #$Lucia. Which is the result?

Slide46: 

Two possible answers: 1. ( ) Empty list: no children 2. (#$Giorgio) Lucia is the mother of Giorgio It depends on the type of Get and Put: with or without inference The one associated with the slot #$mother: #$mother has a #$inverse, ie #$mother-of Which inference ?

Slide47: 

Other inferences in CycL Inverse relations (see above) specSlot-genlSlot: slots having specification or generalization: #$fatherOf #$specSlot #$parentOf If (put #$Luigi #$padreDi #$Marta) Cycl infers (#$Luigi #$genitoreDi #$Marta) TransfersThro: values of a slot are propagated #$book #$writtenIn #$language #$book #$partOf #$section #$writtenIn #$transfersThro #$partOf If (put #$theLightHouse #$writtenIn #$english) (put #$section1 #$partOf #$theLightHouse ) Cycl introduce automaticamente (#$section1 #$writtenIn #$english)

Slide48: 

Other inferences in CycL MutuallyDisjointWith: specify disjunct units #$malePerson #$mutuallyDisjointWith #$femalePerson #$individualObject #$mutuallyDisjointWith #$collection Each time an instance is added to malePerson Cyc checks if it is not already in femalePerson Analogously #$covering and $partitionedInto Coextensional sets: Two Units have necessary the same instances: #$mouse #$coExtensionalSets #$rat if (#$Milu #$instanceOf #$mouse) Cyc asserts (#$Milu #$instanceOf #$rat) (and viceversa)

Slide49: 

Inheritance is extended Other inferences in CycL Standard inheritance Si applica allo slot #$allInstances (tutte le istanze di una unit); Se #$persona #$nazionalità: (#$stato) #$studenteUnivTorino #$genL: (#$persona) #$nazionalità “default per #$studenteUnivTorino = #$Italia” Allora, quando si asserisce #$studenteUnivTorino #$allInstances (… #$Sandra …) Cyc ottiene (per default) #$Sandra #$nazionalità #$Italia

Slide50: 

Ma l’ereditarietà si può anche applicare ad altri slot: Nell’esempio che segue, alla coppia <#$possiedeAuto, #$categoria> #$persona #$possiedeAuto: (#$modelloDiAuto) #$modelloDiAuto #$categoria: (#$utilitaria #$media #$sport #$lusso) #$Carla #$instanceOf #$persona #$possiedeAuto ° #$categoria “default per #$Carla = #$lusso” Allora, quando si asserisce #$Carla #$possiedeAuto (… #$AutoXXX …) Cyc ottiene (per default) #$AutoXXX #$categoria #$lusso Other inferences in CycL N.B. Ho appositamente messo tra virgolette i due default di esempio: non c’è modo in Cyc per specificare queste ereditarietà in modo dichiarativo, ma bisogna usare una funzione apposita.

Slide51: 

Mantenimento di definizioni (toCompute): zio =def fratello ° genitore (formalmente composizione di funzioni). In Cyc: #$zio #$toCompute (#$computeByComposing #$fratello #$genitore) Classificazione: se triangolo =def poligono con esattamente tre lati equiTria =def poligono con esattamente tre lati uguali allora tutti gli equiTria sono triangoli Attivazione di “demoni”. Essi sono procedure, associate agli slot, che vengono eseguite se lo slot viene modificato. Ad esempio, se per gli studenti ho lo slot #$esamiSostenuti e lo slot #$crediti, al primo slot si può agganciare un demone che viene attivato ogni volta che si aggiunge un nuovo esame, incrementando automaticamente il numero di crediti. Other inferences in CycL

Slide52: 

#$Lesmo #$instanceOf: (#$ProfessoreUniversitario) #$insegnaIn: (#$UniversitaDiTorino) #$name: (#$Leonardo) Demons #$UniversitaDiTorino #$instanceOf: (#$Ateneo) #$sede: (#$Torino) #$rettore: (#$RinaldoBertolino) #$Torino #$instanceOf: (#$Città) #$inRegione: (#$Piemonte) #$insegnaIn #$instanceOf: (#$Slot) #$makesSenseFor: (#$ProfessoreUniversitario) #$entryIsa: (#$Ateneo) #$entryFormat: (#$SingleEntry) #$afterAdding: (#$MyComputeByComposing #$risiedeInRegione #$sede #$inRegione)

Slide53: 

Applicazione dei meccanismi inferenziali I metodi visti in precedenza non vengono sempre applicati. Alcuni, infatti (ad esempio la classificazione) sono piuttosto inefficienti. E’ compito dell’utente specificare quali meccanismi desidera siano applicati. All’atto della get (inferenza backward) La get non esiste; esistono invece 4 versioni di potenza differente: get0: nessuna inferenza (come accesso a database) get4: inferenze semplici e molto efficienti (#$inverse, #$toCompute, #$genlSlots, #$TransfersThro, #$coExtensionalSets, …) get8: inferenze plausibili (che non abbiamo visto), es. Ragionamento per analogia, uso di ‘strutture’, … get6: inferenze complesse: ereditarietà completa, classificazione, demoni

Slide54: 

All’atto della put (inferenza forward) In teoria, put come get. In pratica, un unico livello: put4 Applicazione dei meccanismi inferenziali (2) Perchè? Le inferenze sopra il livello 4 sono molto inefficienti sempre, per cui è il caso di farle solo se richiesto Le inferenze sotto il livello 4 sono molto efficienti all’atto della put, molto inefficienti all’atto della get. Es. (get #$Lucia #$motherOf) Applicare l’inverse a (put #$Giorgia #$mother #$Lucia) è immediato e, in tal caso, anche rispondere alla get sopra è immediato. In caso contrario, rispondere alla get richiede controllare tutte le #$Person per verificare se hanno (#$mother #$Lucia)

Slide55: 

Una conclusione sulle inferenze: TMS La conoscenza introdotta tramite meccanismi inferenziali è spesso incerta (per default). V. esempio categoria dell’auto di Carla (slide 40). Cosa succede se viene introdotto manualmente il dato che Carla ha comprato un’utilitaria? Prima alternativa: inconsistenza ® errore Seconda alternativa: alcuni dati sono più incerti di altri: un dato introdotto manualmente è più sicuro di uno inferito tramite inheritance. Per mettere in pratica la seconda alternativa, Cyc usa un TMS (Truth Maintenance System), che in realtà si applica anche a casi più generali di quello dell’esempio. Se Cyc sa che è vero Fatto1, che è vera la regola (non default) Fatto1 ® Fatto2, può dedurre Fatto2. Se viene asserito che Fatto2 è falso cosa si può fare? Si può negare che è vero Fatto1, o cancellare la regola.

Slide56: 

Alcune considerazioni ontologiche Topolino è una #$Thing? Person è una #$Thing? 327.451.666 è una #$Thing? ‘Mangiare al ristorante’ è una #$Thing? Ovviamente sì, anche se non è reale. Sì, della classe (come entità) si sanno alcune cose (come la cardinalità – circa 6 miliardi) Lo è diventata in questo momento! Di essa si può dire che è un numero che è stato usato come esempio nelle lezioni di Lesmo. Sì, ma può essere di due tipi: un’entità ‘intera’ (IndividualObject), o una collezione. ‘La difesa della Juve’ è una #$Thing? E’ certamente una classe (di eventi). Può diventare un’entità se di essa (come classe) si vuol dire qualcosa (v. sopra esempio Person) E’ una collezione, ma, come Person, può avere un’individualità come classe. Influenza è un #$IndividualObject o una #$Collection?

Slide57: 

Top level of Cyc List of concepts is on the website (http://www.opencyc.org/  downloads) Some descriptions of concepts #$Thing: Universal set. Every constant belongs to it

Slide58: 

#$Intangible: set of things not made of matter. E.g., a Collection #$Collection is #$Intangible even if the instances belonging to it are. ATTENTION: tangible is different from perceptible. Light is perceptible but not made of matter (at a certain leval of detail) #$Individual: Set of things which are not sets. #$Individual includes physical objects, numbers, relations, groups #$Individuals have parts but not members #$IntangibleIndividual: The set of intangible individuals.They have no matter, volume color…: times, ideas, algorithms, numbers, distances. No sets! #$TemporalThing: set of things with a temporal extension: it is possible to ask about them “when”? it includes actions, tangible objects. Abstract and matematical things are not in the set since they are atemporal

Slide59: 

Conclusions about Cyc A huge and complex system which includes both an ontology and a reasoning system Pros: - Critical mass - Inferential capacity - Optimization of specialized inferences Cons: - Too complex - Opacity of ontological choises - Some failures (relation with natural language)

Slide61: 

Version 1990 Version 1997 Perchè ?

PLAN OF THE LESSON: 

PLAN OF THE LESSON What are ontologies for Existing ontologies WordNet Cyc SUMO DOLCE

Slide63: 

SUMO (Suggested Upper Merged Ontology) is a project of IEEE (Institute of Electrical and Electronic Engineering), started at mid ’90s. SUMO can be found on the IEEE working group SUO (Standard Upper Ontology) http://suo.ieee.org “Standard specifying a ‘upper ontology’ which computers will use for applications like data interoperability, research of information, automated reasoning and natural language processing. An ontology is like a dictionary or thesarus but with more details and structure and that a computer can use. An ontology is a set of concepts, axioms and relations in a certain domain. An ‘upper ontology’ is limited to ‘meta’- concepts, generic, abstract, covering a wide number of domains. Sumo will provide no individual domain but a general standard and structure to be used in the different domains (eg. engineering, medical, finance, etc.).” SUMO

Slide64: 

SUMO Components

Slide65: 

Structural Ontology SUMO describes the SUMO primitives (asserted StructuralOntology (instance instance BinaryPredicate)) (asserted StructuralOntology (instance instance AntisymmetricRelation)) (asserted StructuralOntology (domain instance 1 Entity)) (asserted StructuralOntology (domain instance 2 Class)) Instance relation (asserted StructuralOntology (instance subclass BinaryPredicate)) (asserted StructuralOntology (instance subclass PartialOrderingRelation)) (asserted StructuralOntology (domain subclass 1 Class)) (asserted StructuralOntology (domain subclass 2 Class)) Subclass relation (asserted StructuralOntology (=> (subclass ?C1 ?C2) (forall (?X) (=> (instance ?X ?C1) (instance ?X ?C2))))) example axiom

Slide66: 

Structural Ontology (2) Inverse relation (asserted StructuralOntology (instance inverse BinaryPredicate)) (asserted StructuralOntology (instance inverse SymmetricRelation)) (asserted StructuralOntology (domain inverse 1 BinaryRelation)) (asserted StructuralOntology (domain inverse 2 BinaryRelation)) (asserted StructuralOntology (=> (and (inverse ?R1 ?R2) (instance ?R1 BinaryRelation) (instance ?R2 BinaryRelation)) (forall (?X1 ?X2) (<=> (holds ?R1 ?X1 ?X2) (holds ?R2 ?X2 ?X1))))) An axiom for inverse A differenza di Cyc, non ‘procedure di inferenza’, ma formule logiche

Slide67: 

Base Ontology (the top-level) Entity: ("x) instance (x, Entity) everything is an Entity ($x) instance (x, Entity) Entity is not empty ("c) instance (c, Class) ® subclass (c, Entity) Every class is a subclass of Entity Axioms

Slide68: 

Physical: ("x) Physical (x) ® [($y,z) located (x, y) Ù existant(x,z)] Every physical entitye has a place (y) and temporal instante (z) Base Ontology (again) Base Ontology (below top-level) Intentional process

Slide69: 

Base Ontology (below top level) Intentional Processes

Slide70: 

Process: ("x) Process (x) ® ($y) subProcess (x,y) Every process has a subprocess. Base Ontology (axioms on processes) subProcess: ("x,y) subProcess (x,y) ® ($t) existant (y,t) Every subprocess exists in a temporal instant IntentionalProcess: ("x) IntentionalProcess (x) ® ($y) agent (x, y) Every intentional process has an agent ("x,y) subProcess (x,y) ® [(" z) located (y,z) ® located (x,z)] A subprocess is located as the process ("x,y) subProcess (x,y) ® WhenFn(x)=WhenFn(y) Ú during (WhenFn(x),WhenFn(y)) Temporal inclusion of subprocesses ("x,y) IntentionalProcess (x) Ù agent(x,y) ® CognitiveAgent(y) Ù ($z) hasPurposeForAgent (x, z, y)

Slide71: 

Conclusions about SUMO A real ontology: no reasoning capabilities but only descriptions of concepts and properties Pros: - Knowledge separated from reasoning - Wide ontology - Integration of different ontologies Cons: - More transparent than Cyc but not enough - Limited number of axioms - Not efficient reasoning - “Political” issues The language in which SUMO is expressed is KIF (Knowledge Interchange Format) and uses its inferential capabilities

PLAN OF THE LESSON: 

PLAN OF THE LESSON What are ontologies for Existing ontologies WordNet Cyc SUMO DOLCE

Slide73: 

Dolce (and clean ontologies)

Slide74: 

Sweetening ontologies with Dolce Dolce (Descriptive Ontology for Linguistic and Cognitive Engineering) from Istituto per le Scienze e le Tecnologie Cognitive del CNR (Trento-Roma), in the EU project WonderWeb (http://wonderweb.semanticweb.org/) Dolce doesn’t want to be a universal ontology, but as a starting point to compare existing ontologies and making hidden assumptions explicit Dolce has cognitive orientation, it expresses ontological commitments of natural language and common sense Ontoclean is the methodology underlying Dolce for building ontologies

Slide75: 

OntoClean 4 fundamental facets of concepts Identity: the possibility to distinguish two instances in base of characteristic property Eg. ‘Person’: ‘having same fingerprint’ Dependence: Property P depends on property Q, if, when Q is true, P is true too Eg. ‘have children’ depends on ‘being father’ Rigidity: if a property is essential for all instances Eg. ‘Person’ is rigid; ‘Student’ is not rigid Unity: all parts identified with a unifying relation Eg. ‘Firm’: ‘being employed in the firm’

Slide76: 

Why facets? Constraints on the subsumption relation A non rigid class (-R) cannot subsume a rigid one (+R) Ex. ‘Legal Agent‘ cannot subsume ‘Person’ A class with identity (+I) cannot subsume one without (-I) A class with unity (+U) cannot subsume one without (-U) Ex. ‘Amount of Matter‘ cannot subsume ‘Physical Object’: if some amount is removed from a quantity the quantity is not itself anymore. A person remains the same even if hair is cut Dependent properties (+D) cannot subsume independent ones (-D) Ex. ‘Park‘ cannot subsume ‘Location’

Slide77: 

Il top level di Dolce P.S. top-level above dashed line

Slide78: 

Endurant e Perdurant Endurants are always completely presents with all their parts: objects Only a part of a Perdurant (events) is present at a certain moment of time Concepts of change: only Endurant can change (mantaining identity) but Perdurant cannot change since their parts are distributed over time Relation among Endurant e Perdurant: partecipation; Endurant participate (have a role) in Perdurants. When I go home, I participate in the event of going as the agent

Slide79: 

Qualities (color) are essential components of entities. Similar to properties but are individual not classes (eg Person). The color of a rose is one of its qualities. Another rose can have the same color, but it is another individual, even if they have the same values for the measures of the color. Abstract appears also in other ontologies. They exist outside space and time (numbers, sets)

Slide80: 

1.      This rose is red 2.      Red is a color 3.      This rose has a color 4.      The color of this rose turned to brown in one week 5.      The rose’s color is changing Red is opposite to green and close to brown

Slide81: 

Descriptions e Situations Newest extension of DOLCE. The base classes are description and situation. Description subsumes S-description (Situation Description) C-description (Concept Description): Each C-description describes how an entity appears in a situation il modo in cui un’entità appare in una situazione. C-descriptions are parts of a S-description The ”Descriptions” module contains the largest and most peculiar extension to DOLCE. The module implements the so-called theory of ”descriptions and situations” in the form of a ”design pattern” that can be applied to many domains without important modifications. Basic relations are satisfied-by (links a s-descrizion and a situation) and selects (link a c-descrizion with an entity)

Slide82: 

METHOD PLAN TECHNIQUE PROJECT GOAL S-DESCRIPTION REGULATION OBLIGATION COMMITMENT SCRIPT NORM CONTRACT PROMISE DESCRIPTION references ENTITY satisfied-by SITUATION encompasses = there is a c-description part of the s-description which describes the entity expects = there is a course part of the s-description which sequences the perdurant PERDURANT involves = there is a functional-role part of the s-description which is played by the endurant ENDURANT INFORMATION -DESCRIPTION INTERNAL -DESCRIPTION PHYSICALLY-DEPENDS-ON exactly one entity SOCIAL -DESCRIPTION PHYSICALLY-DEPENDS-ON more than one entity FORMAL -DESCRIPTION CLASS-OF -DESCRIPTIONS INFORMAL -DESCRIPTION THEORY TERMINOLOGY TOPIC refines S-DESCRIPTIONS CLASSIFICATION regulates envisages constrains

Slide83: 

A clinical condition (situation) has an associated diagnosis (s-description) made by some agent. Examples A case in point (situation) is constrained by a certain norm (s-description) A murder (situation) has been reported by a witness (functional role) in a testimony (s-description) Information science as a topic (s-description) references the manipulation of data structures (situation), both as a pure or applied science (parent s-descriptions)

Slide84: 

Some axioms

Slide85: 

Critic of the top-level of WordNet In Dolce there is the difference between concepts (person) and material roles (student), based on meta-property of ‘rigidity’. No ontology respects this!!

Slide86: 

Conclusions about Dolce It is not an ontology nor an inferential system, but a metodology (OntoClean) Based on the metodology they propose a “top-top level” (Dolce) Pros: - Philosophical foundation - Base for judging ontological choices Cons: - Manual hand work - Not directly usable

Slide87: 

Formal Ontological Analysis • Theory of Parts • Theory of Wholes • Theory of Essence and Identity • Theory of Dependence • Theory of Qualities • Theory of Composition and Constitution

Slide88: 

Mereology • Primitive: proper part-of relation (PP) – asymmetric – transitive – Pxy =def PPxy \/ x=y • Axioms: Excluded models: supplementation: PPxy → Exist z ( PPzy /\ ¬ z=x) principle of sum: Exist z ( PPxz /\ PPyz /\ ¬ exist w(PPwz /\ ¬ (Pwx \/ Pwy))) extensionality: x = y ↔ (Pwx ↔ Pwy)

Slide89: 

Extensionality and mereological invariance Extensionality: whenever the parts exist, x exists (the whole is always the sum of its parts) Mereological invariance: x always keeps its parts Examples of extensional entities: – Amounts of matter – Regions – Pluralities (pseudo-extensionality) Mereologically invariant (but non-extensional) entities: – A physical body (a lump of matter)

Slide90: 

Kinds of Whole • Depending on the nature of ω, we can distinguish: – Topological wholes(a piece of coal, a lump of coal) – Morphological wholes(a constellation) – Functional wholes(a hammer, a bikini) – Social wholes(a population) * a whole can have parts that are themselves wholes(with a different ω)

Slide91: 

Parts vs. components • A proper part is a componentiff it is a whole • We can have topological components, morphological components, functional components.…

Slide92: 

Essence and Rigidity Certain entities have essentialproperties. – John must have a brain. – John must be a person. Certain properties are essential to alltheir instances (compare being a personwith having a brain). These properties are rigid - if an entity is ever an instance of a rigid property, it must always be.

Slide93: 

Permanent vs. Essential Properties • Being always a student • Being necessarilya student

Slide94: 

Why bother with this? • Formal ontological analysisrequires analyzing all properties according to their meta-properties – This is a lotof work! • Why perform this analysis? – Makes modeling assumptionsclear, which: • Helps resolving known conflicts • Helps recognizing unkown conflicts – Imposes constraints on standard modeling primitives ( generalization, aggregation, association) – Elicits natural distinctions – …results in more reusable ontologies

Slide95: 

Resolving Ontological Conflicts • Two well-known ontologies define: – Physical Object is-a Amount of Matter(WordNet) – Amount of Matter is-a Physical Object(Pangloss) Amount of Matter – unstructured /scattered “stuff” – Identity: mereologically extensional – Unity: intrinsically none (anti-unity) • Physical Object – Isolated material body – Identity - three options: • None • Non-extensional • Extensional – Unity: Topological Conclusion: the two concepts are disjoint. Physical objects are constitutedby amounts of matter

Slide96: 

Overloading Subsumption Common modeling pitfalls • Instantiation • Constitution • Composition • Disjunction • Polysemy

Slide107: 

Particulars and Universals • Universals – Have multiple exemplifications – All abstract • Particulars: – Have no exemplifications – Can be either concrete or abstract Concrete entities are all particulars

Slide108: 

Instance-of vs. membership (1) • The problems of logical predication – x is an apple → Apple(x) – x is red → Red(x) • Instance-of vs. class membership – John is a member of “Person” → Person(John) – Tree1 is a member of “BlackForest” → BlackForest(Tree1) ?? (violates usual intended interpretation of unary predicates: property shared by all instances of the corresponding class. Doesn’t pass the “is-a” test )

Slide109: 

Instance-of vs membership (2) • Instance-of: – Particular-universal – Universal-universal • Membership: – Particular-particular

Slide110: 

Abstract vs. Concrete Entities • Concrete: located in space-time (regions of space-time are located in themselves) • Abstract - two meanings: - Result of an abstraction process (something common to multiple exemplifications) 􀀍 Not located in space-time • Mereological sums (of concrete entities) are concrete, the corresponding sets are abstract...

Slide111: 

Endurants: – All proper parts are present whenever they are present ( wholly presence, no temporal parts ) – Exist in time – Can genuinely change in time – Need a time-indexed parthood relation • Perdurants: – Only some proper parts are present whenever they are present (partial presence,temporal parts ) – Happen in time – Do not change in time – Do not need a time-indexed parthood relation

Slide112: 

Qualities and qualia • Linguistic evidence – This rose is red – Red is a color – This rose has a color – The color of this rose turned to brown in one week – The room’s temperature is increasing – Red is opposite to green and close to brown • Every entity comes with certain qualities that permanently inhereto it and are uniqueof it • Qualities are perceptually mapped into qualia, which are regions of quality spaces. • Properties hold because qualities have certain locations in their quality spaces. • Each quality type has its own quality space

Slide113: 

Features are “parasitic” entities, that exist insofar their host exists. • Features may be relevant parts of their host, or places (which are not parts of their hosts). • All features are essential wholes, but no common unity criterion may exist for all of them

Slide114: 

Participation relations • Hold between a perdurant and its involved endurants • Extremely relevant for domain modelling • Current axiomatization covers: – constant vs. temporary – complete vs. partial • Further distinctions are currently primitive (thematic roles) – Agent, Theme, Substrate, Instrument, Product – More is needed on event structure, intentionality, and artifacts to produce analytic definitions

Slide118: 

Conclusions on Ontologies Ontologies are essential for interoperability and reasoning A lot of efforts, projects and money Many proposals but yet no standart Menace for cultural diversity?