TERMINOLOGY – CONCEPT MODELLING – ONTOLOGY XXVI VAKKI SYMPOSIUM – 10-11-12 February 2006 - Vaasa, Finland IS ONTOLOGY OVERRATED? When I was asked to propose some titles for my talk… Is ontology overrated? : It was a joke Ontology Acquisition for Terminology Pr. Christophe Roche University of Savoie - Campus Scientifique 73 376 Le Bourget du Lac cedex – France christophe.roche@univ-savoie.fr http://www.ontology.univ-savoie.fr

? ONTOLOGY Knowledge Representation Language Oiled LOOM Protégé Ontonlingua OWL ONTOLOGY RDF Schema SHOE WebOde WordNet KIF TOVE Description Logic Representation OntoEdit DAML-OIL UML Language  Representation of Ontology ? As a matter  Natural Language ?  Logic ?  UML ?  Text and Ontology ?  Lexical Ontology ?  WordNet ?  Ontology Building from Text ?

Some principles…  What is it for ?  Ontology is an Object  Representation  Ontology is a “Science” Object : Product, Result of a Practice Today we focus too much on representation than on knowledge. Science: systematic study and knowledge of a domain Building : Ontology does not belong to Linguistics  Knowledge  Principles  Ontology Management  Building  Exploiting  Maintaining - Updating

« TERMINOLOGY – CONCEPT MODELLING – ONTOLOGY » TABLE OF CONTENTS « TERMINOLOGY – CONCEPT MODELLING – ONTOLOGY »  Terminology  Concept Modelling  Ontology  Representation Language  Knowledge  Text  Wordnet, UML, Naming the World  Conclusion

TERMINOLOGY ISO 1087-1:2000 special language : language for special purposes (LSP), language used in a subject field and characterized by the use of specific linguistic means of expression. subject field : Domain, field of special knowledge terminology 1 : set of designations belonging to one special language terminology 2 : science studying the structure, formation, development, usage and management of terminologies designation : representation of a concept by a sign which denotes it concept : unit of knowledge created by a unique combination of characteristics. Note: Concepts are not necessarily bound to particular languages. They are, however, influenced by the socialor cultural background which often leads to different categorizations. characteristic : abstraction of a property of an object or of a set of objects. generic relation (genus-species relation) : relation between two concepts where the intension of one of the concepts includes that of the other concept and at least one additional delimiting characteristic Intension : set of characteristics which makes up the concept essential characteristic : characteristic which is indispensable to understanding a concept delimiting characteristic : essential characteristic used for distinguishing a concept from related concepts

TERMINOLOGY Handbook of Terminology terminology 1 : the set of special words belonging to a science, an art, an author, or a social entity. terminology 2 : the language discipline dedicated to the scientific study of the concepts and terms used in specialized languages. concept : a unit of knowledge abstracted from a set of characteriscs attributed to a class of objects, relations or entities. term : a word (simple term), multiword expression (complex term), symbol or formula that designates a particular concept within a given subject. generic relationship : the hierarchical relationship between a general concept and a series of subordinate concepts that inherit its properties but which are distinguished from one another by at least one delimiting characteric.

“signified” [signifié] “signifier” [significant] Community of Practice Language for Special Purpose Lexicology { specialized words } usage words ? Terminology { terminology units } Terminology : objectives : to eliminate ambig concept “signified” [signifié] réel denomination “signifier” [significant] designation

? Semiotic Triangles Ogden – Richards Saussure Scholastic signified conceptus signifier referent vox res signified (meaning) : a value in system concept : a set of characteristics

CONCEPT MODELLING  Concept Definition  Concept Modelling meaning of a term = denoted concept  Concept Definition definiendum = definiens Concept = set of characteristics - extra linguistic - system of concepts How to represent the definiens ? the definition simply consists of a definiendum followed by a definiens, i.e. an expression that gives the meaning of the word, with some punctuation. The words definiendum and definiens are Latin participle forms of the verb definire 'to define', and they mean 'the one being defined, the one to be defined' and 'defining, the one that defines', respectively. By the way, the meaning of that verb was originally related to limiting, then to designating by limiting; the verb is derived from the noun finis 'boundary, limit, border, end'. Similarly, a formula used for defining something usually has the form definiendum = definiens, e.g. F = ma (force equals mass times acceleration). (Most formulas are not definitions, of course, and it depends on the system of physics or other realm of knowledge which formulas are taken as definitions. A formula which is a definition in one system might be a derived one in another.) Even a definition written as a normal sentence often consists of a definiendum = definiens pattern, just with a word like is or means instead of an equals sign. Example: A liger is a hybrid (cross) between a tiger and a lion. But a simple definition could be written in a different form, and for stylistic reasons this is quite common. It is normal to write a new word after its definiens: A hybrid (cross) between a tiger and a lion is called a liger. salve veritate ? two expressions are “synonymous” in a language L iff they may be interchanged salve veritate in each sentence of L in which they occur.  Concept Modelling - scientific approach “formal” langage - system - definition in natural language = comment

CONCEPT MODELLING  “Formal” Language - System Why ? : objectives of Terminology : - precise (without ambiguity) - consensual - re-usable to get off the problems risen by NL What is it ? - “theoretical concepts” - rules (syntax) - operations (reasoning) => to build a representation of the concepts of the domain.

A specification of a conceptualization CONCEPT MODELLING  “Theoretical Concepts” Semi formal : Class, Relationships (hierarchy) class-def white-wine subclass-of wine slot-constraint has-color has-filler white The trap of formal science : if you accept the hypotheses, you are obliged to accept its construction (definition) Formal : first order Logic (Predicates) hypothetical – deductive systems Person : (and Animal (all (restrict hasParent Person)) A specification of a conceptualization ONTOLOGY (Gruber)

ONTOLOGY Web search result for: Ontology

ONTOLOGY Methontology Sensus IDEF5 SHOE LOOM Mikrokosmos KIF Ontonlingua OWL Conceptual Graphs WordNet 2 Questions : - How can I find my in a such jungle ? - Why ontology is so popular ? DAML-OIL Semantic Networks Protégé Enterprise Ontology RDF Schema Schemas Description Logic OntoEdit WebOde BSDM TOVE WebOnto Oiled Cyc Kaon KR Ontology CommonKADS

Ontology : Why ? Common Language A Myth : A shared and common understanding of some domain that can be communicated across people and computers So, Ontology is a very popular subject. Why is it so popular ? Common Language - no communication - no knowledge sharing - no knowledge exchanging … without agreement on the meaning of terms

Collaborative Engineering Ontology : What is it for ? …to enable communication and knowledge sharing between people and computers ! “Ontologies are finding applicability in many other areas of information systems engineering, for example, in database design, in object systems, in knowledge based systems and within many application areas, such as datawarehousing, knowledge management, computer supported collaborative working and enterprise integration.” Ontology.org Collaborative Engineering Search Engine Data Base Knowledge Management Semantic Web Information System Multi-Agent Systems Natural Language Interoperability E-Commerce Communication Is a general view of ontology possible ?

 Ontology : What is it ? A Knowledge Engineering Point of View  There is today an agreement on the definition :  Set of Concept Definitions and Relationship Definitions What is an Ontology ? Short answer: An ontology is a specification of a conceptualization. In the context of knowledge sharing, I use the term ontology to mean a specification of a conceptualization. That is, an ontology is a description (like a formal specification of a program) of the concepts and relationships that can exist for an agent or a community of agents. This definition is consistent with the usage of ontology as set-of-concept-definitions, but more general. Tom Gruber  Vocabulary of Terms « An [explicit] ontology may take a variety of forms, but necessarily it will include a vocabulary of terms and some specification of their meaning (i.e. definitions). » “Ontologies: Principles, Methods and Applications” M.Ushold & M.Gruninger. Knowledge Engineering Review, Vol.11, n°2, June1996  There is today an agreement on the objective :  Communication and Knowledge Sharing between Human and/or Software Agents « The main purpose of an ontology is to enable communication between computer systems in a way that is independent of the individual system technologies, information architectures and application domain. » www.ontology.org  A Knowledge Engineering Point of View

Specification of a Conceptualisation Ontology : What is it ? - 2 components : - a vocabulary of terms , - a set of definitions. Common Language Properties : - consensual - coherent - precise - sharable Vocabulary Specification of a Conceptualisation < … > Reasoning : - queries - assertions - inferences Terms = concept’s names

? Ontology : What is it ? < … > Term’s Meanings ? < … > Thing The Knowledge Engineering point of view: “What exists is that which can be represented” “An ontology is a shared description of concepts and relationships of a domain expressed in a computer readable language” class-def white-wine subclass-of wine slot-constraint has-color has-filler white Person : (and Animal (all (restrict hasParent Person)) - agreement on definition (Gruber). (agreement on content ?)

Representation Languages “An ontology is a shared description of concepts and relationships of a domain expressed in a computer readable language” Logic-based languages - clear and formal syntax and semantics - sound inferences - A concept is a well formed formula - A concept is the intension definition of a set - operational languages Properties : - consensual - completeness - coherent - soundness - precise - can be shared (interchange format) A concept (category) is an unary predicate. ‘form(x)’ = independant(x)  abstract(x) Logic is necessary

Representation Languages Frame-based languages Artificial Intelligence Frame System, Conceptual Graph, Semantic Network. - A concept (class) is a set of slots - Facets are associated to slots - Concepts are organized according to the « sub-class » relationship (a simple or multiple inheritance relationship) - The concepts are structured into graphs or taxonomies - The meaning of a term is the concept denoted by the term

Representation Languages : An Example OWL The Web Ontology Language - DAML : DARPA Agent Markup Language - OIL : Ontology Inference Layer Formal semantics and reasoning support Human readable form Logic Frame Class-def defined adult-elephant subclass-of elephant slot-constraint age has-value (min 20) OWL Web Interchange format WEB Syntax (XML & RDF)

The Representation Language Problem Too nice to be true… ?  Always Coherent ?  Can be Re-Used ?  Really Compatible ? The bride is too beautiful  Really Shared ?  Really Consensual ?  Do I agree with the vocabulary of terms ?  Do I agree with the (formal) meaning of terms ?  Is the conceptualization really common and shared ?  How do I build such a domain conceptualization ?

The Representation Language Problem “An ontology is a shared description of concepts and relationships of a domain expressed in a computer readable language” Logic-based languages - Human readable ? - Re-usable ? - Merging ? Enterprise Ontology (Define-Class Activity-Or-Spec (?X) "The union of Activity and Activity-Spec" :Iff-Def (And (Eo-Entity ?X) (Or (Activity ?X) (Activity-Spec ?X))) :Axiom-Def (Partition Activity-Or-Spec (Setof Activity Activity-Spec))) TOVE (define-class plan_action (?a) :def (forall (?alpha ?f ?s) (=> (holds (agent_constraint ?alpha (fluent_goal ?f)) ?s) (forall (?ap ?s1 ?s2) (=> (and (subaction ?ap ?a) (leq ?s1 ?s2) (Do ?ap ?s1 ?s2 (intended ?s2)) (holds ?f ?s2))))) Axiom for the ‘dormant’ status of an activity (dormant, executing, suspended, reExecuting, terminated) (EQ 38) (∀ a,e, σ) holds(activity_status(a, dormant), do(e, σ)) ≡ ((&eksist; s) state(s,a) & e=commit(s,a) & holds(status(s,a,possible), σ)) | ¬((&eksist; s) substate(s,a) & e=enable(s,a)) & holds(activity_status(a, dormant),σ).

The Representation Language Problem Using a same formal language (logic) is not a guarantee of consensus ! - There is a consensus about the syntax and the semantics of the language. - It does not mean a consensus on the knowledge express with this language. Epistemological Problems :  definition of a concept  a concept is not a wff  a set is not a concept  an essential property is not a relation  ... Logic is necessary, but a posteriori, not a priori.

The Representation Language Problem “An ontology is a shared description of concepts and relationships of a domain expressed in a computer readable language” Frame-based languages “The mercury is a both a metal and a liquid” MIKROKOSMOS « In this ontology, you should not expect to find : any kind of guarantees, warrantees, or liability for correctness or precision, formally clean or theoretically "pure" concepts, complete consistency; guaranteed absence of contraditions; etc » - More epistemological than logic (class, slot, relationships) - Less sound Epistemological Problems :  a technique (representation) does not define a knowledge theory (conceptualization)  subsumption is more than an inheritance relationship  an essential property is not an attribute  ...

? WHAT IS THE PROBLEM ? Go back to the definition… The main objective of ontology from the computer science point of view is normalization based on a specification of a conceptualization It means that the main objective is : - not to define meaning of terms which would imply a linguistic theory, - not to understand the world which would imply an epistemological theory, - but to define concepts based on a computational language in order to manipulate entities. A more representation-oriented approach than knowledge-oriented. Go back to the definition…

Knowledge Epistemology Representation Linguistics

 What « Language - Theory » to choose ? Knowledge & Language The ontology needs to be specified in some language Intention Language « Object » Theory « Agent »  What « Language - Theory » to choose ?

Knowledge – Representation - Visualization Visualizations Representation Languages There is no pure fact. A fact is what it is only under the view of the concept which founds it. Theory 1 Theory 2 Theory n World Information Space

Knowledge – Language - Representation  World : « Every cat is on a mat »  Theory : Existantial Graph of Pierce  Representation Languages LF : [Cat:  ] ® (On) ® [Mat] [Cat: @every ] ® (On) ® [Mat] CGIF : [Cat: @every *x] [Mat: *y] (On ?x ?y) ( On [Cat every ] [Mat] ) KIF : (forall ((?x Cat)) (exists ((?y Mat)) (On ?x ?y))) Logic : (x:Cat) ($y:Mat) on(x,y)  Visualization

Visualization UML Protégé

Theory Visualization Language

 What « Language - Theory » to choose ? Knowledge & Language The ontology needs to be specified in some language Intention Language To choose a language which helps me to model and understand the world « Object » Theory « Agent »  What « Language - Theory » to choose ?

ONTOS ( ? ) + LOGOS (Language – Science – Reason) Etymology ONTOS ( ? ) + LOGOS (Language – Science – Reason) ONTOS : ? « I am a man » in Spanish : « Soy un hombre » « I am ill » in Spanish : « Estoy enfermo » Soy un hombre Estoy enfermo to be : ser , estar French Nouns : Être – Étant German Nouns: Sein – Dasein English : Being ? Beingness ?

Science of Being as Being “Science of Existence” Etymology Science of Being as Being Etymology & Philosophy Soy un hombre Estoy enfermo Essence “Science of Being” Metaphysics Properties “Science of Existence” Phenomenology

Epistemology  An Ontology must reflect the structure of the world !  There are different kinds of knowledge Terminological & Ontological Logic Logic of Judgments Logic of Reasoning Concept : knowledge applied to a plurality of things governed by a same law. Ontology : science of being as being independently of its particular states. => Different Languages Concept, Set, Class Essential property, Atrtibute Relationship => Different “Theoretical Concepts”

Logic  A neutral language no epistemological principal : rewriting system => “good” formal properties  Epistemological problem Unary predicate : Concept or a Property Apple (x) , Red (x) Binary Predicate : Property or a Relationship Color (x,y) , GreaterThan (x , y)

Logic  Extend Logic Postulates written in logic itself “Formal Ontology” N. Guarino “ontological rigidity” : x Apple (x)  Apple (x)  (x Red (x)  Red (x)) - Useful to constrain “judgments” about world’s state of affairs Ontological rigidity apply to unary predicates in order do distinguish those which represent concetps from those which represent attributes  But to express a posteriori the nature of knowledge no new “theoretical concepts”  Concept not defined in terms of proprieties Property-oriented approach (DL) : a property is defined in terms of classes to which it applies

Protégé

Aristotelian Approach  Concept A concept is defined according to its essence (an attribute is an accident) Concepts are structured according to their difference  A concept is defined by “specific differentiation” « … the difference has two aspects, one with respect to the genus it divides and separates, the other the species it constitutes and forms, making up the principal part of the comprehension of the idea of the species. » Logic or the Art of Thinking, Arnauld & Nicole  Porphyry Tree (attributes flesh the skeleton)  No multiple inheritance delimiting characteristic : essential characteristic used for distinguishing a concept from related concepts

Essential property : defines a concept according its nature Epistemological properties : - focus on essence and not on state - concept and set are different notions - essential property and attribute are different notions Attribute : describes a state of an object Concept : a set of common attributes

OCW

 Designation versus Denomination (onomasiology) NAMING THE WORLD Hermogenes: Cratylus says, Socrates, that there is a correctness of name for each thing, one that belongs to it by nature. A thing’s name isn’t whatever people agree to call it - … - but there is a natural correctness of names, which is the same for everyone, Greek or foreigner. Cratylus, Plato  Designation versus Denomination (onomasiology) How to name the concepts in such a way it expresses the structure of the ontology? « … every species can be expressed by a single noun, such as ‘mind’ or ‘body’; or by two words, namely one for the genus and one for the difference; this called a definition, such as ‘thinking substance’, ‘extended substance’. » Logic or the Art of Thinking. Arnauld & Nicole Terminology Porphyry’s Tree

WORDNET  a SEMANTIC NETWORK of the English lexicon : - about 150,000 words organized in over 115,000 synsets (set of synonyms) - synsets are connected via linguistic relationships: synonymy hypernymy… - Development began in 1985 - Created and maintained at the Cognitive Science Laboratory of Princeton University  Is it an Ontology? - stricto sensu: No. - There is no definition of concept. - Linguistic relationships belong to linguistics not to conceptualization.  Lexical Ontology - word = lexicalized concept - hypernymy = lexicalized subsumption

WORDNET Wordnet in RDFS and OWL A related but distinct activity would be to describe the use of Wordnet as a basis for RDF/OWL class and/or property hierarchy. Wordnet's noun term (hypernym) hierarchy captures "an X is a kind of Y" relationships between English category terms based on conventional usage.  Can we “align” Wordnet (lexical resource) with Ontology (conceptualization)? - probably not.

UML  Unified Modeling Language : - Can we use it for Ontology Development ?  Ontology : Knowledge Engineering Community - Class/Subclass hierarchies - Relationships between classes - Class Attributes - Constraints UML : its scope includes more modeling tasks  UML : Software Engineering Community - Class Diagrams : Class Class/Subclass hierachies - Class Attributes - Object Constraints Language (OCL)

UML - Graphical notation - Standard (Object Management Group) - Widely Used - Requires less expertise than Protégé - RDF Schemas from class diagram - Less formal - Limitation for describing diagram

Le sens de l’information Condillac Condillac Research Group « Knowledge Engineering » TEXT & ONTOLOGY Pr. Christophe Roche christophe.roche@univ-savoie.fr http://www.ontology.univ-savoie.fr

? ONTOLOGY BUILDING FROM TEXT “Textual” Ontology Informal representation Formal representation Semi-formal representation

 Example : Relay  Extracting candidate terms - extracting candidate terms from corpus by automatic text analysis. linguistic expressions : “relay”, “voltage relay”, “threshold relay’, “electromagnetic relay”, etc. - words of usage - LSP lexicon - Terminology (designations?) Texts are first analyzed from a lexical and syntactic point of view in order to associate to each word its grammatical class and its canonical form. Then linguistic expressions can be automatically extracted using, for example, lexical-syntactic patterns (the “adjective – substantive” and “substantive – substantive” patterns allow to extract expressions like “electromagnetic relay”, “threshold relay”, “voltage relay” from a technical corpus about of relays). This lexicon is structured according to linguistic relationships like hypernymoy, hyponymy, synonymy meronymy and so on. Here too, these linguistic relationships can be “automatically” extracted from the corpus (syntactic analysis based on verbal patterns: “a voltage relay is a kind of relay”) or from the extracted expressions. For example, linguistic expressions made up of several words with the same “queue” (i.e. ending with the same words, for example with the same name) give interesting information about the structure of the domain conceptualisation.  Structured lexicon - hyponymy

 Example : Relay  Conceptual Structure names  lexicalized concepts hyponymy  subclass x VoltageRelay (x)  Relay (x) x On-OffRelay (x)  Relay (x) x ThresholdRelay (x)  Relay (x) Let us apply this conceptualization to a CMS (content management system) application. Documents can be semantically annotated based on this conceptualization [Kiryakov and al. 05]. Searching information on “threshold relay” does not return any information about “voltage relay”. This is not correct. As a matter of fact, for experts, all information about “voltage relay” concerns “threshold relay”: the conceptualization is not right. Where is the problem?  Hypothesis : Lexical and conceptual structures are isomorphic - Content Management System : . semantic annotation . Information retrieval

 Example : Relay  The “trap” of natural language - rhetorical figures : metonymy a <voltage relay> is a kind of <threshold relay> whose threshold value is voltage - incompleteness of language  The “correct” conceptual structure Texts fall within the language in action. Using rhetorical figures, such as metonymy and ellipse, is a very ordinary practice. One of the principles in textual linguistics, even in technical documents, is the incompleteness of texts. So we can not always directly deduce from texts the conceptual relationships.  Words of usage and designations

 Example : Turbine - The three main types of water turbines are Pelton wheels, Francis turbines, and Kaplan or propeller type turbines. - A Kaplan turbine is a type of propeller turbine in which the pitch of the blades can be changed to improve performance. - A propeller turbine is a Kaplan turbine with fixed blades… - A Kaplan turbine looks like a propeller turbine…

TO CONCLUDE  Textual Ontology : - consensus ? - re-use ?  Incompleteness of Text  The Lexical and Conceptual Structures do not fit  The Signified is not the Concept  Ontology is extra linguistic

TO CONCLUDE

« IS ONTOLOGY OVERRATED ? » CONCLUSION « IS ONTOLOGY OVERRATED ? »  The word “Ontology” is overused !  The scope is too large : - Knowledge engineering - Logic - Linguistics - Information System - Philosphy Concept : - informal : word - semi formal : frame, class - formal : predicate Knowledge Engineering : KIF, semi formal ontology Logic : formal ontology,description logic : Protégé Lingusitics : Lexical ontology, textual ontology Philosphy : phenomenology, analytic philosphy Information system :UML  More Representation-oriented than Epistemology-oriented

« IS ONTOLOGY OVERRATED ? » CONCLUSION « IS ONTOLOGY OVERRATED ? »  Go back to the first definitions generic relation (genus-species relation) : relation between two concepts where the intension of one of the concepts includes that of the other concept and at least one additional delimiting characteristic Architecte de données Pas de normalisation a priori Les principes épistémologiques guident la construction de l’ontologie, La modélisation formelle lève les ambiguïtés Si l’on doit tenir compte de la LSP : avec prudence. C’est une langue, signifié : signification contextualisée. Ces corpus n’ont pas objectif de définir des modèles conceptuels qui par définition de la langue. Structure du lexique ne se superpose pas à la structure conceptuel du domaine L’objectif ici n’est pas de décrire le sens mais les concepts qui décrivent les objets du monde delimiting characteristic : essential characteristic used for distinguishing a concept from related concepts essential characteristic : characteristic which is indispensable to understanding a concept

OCW

Protégé