Ontology (with a capital O) is a branch of philosophy allowing the study of the nature and identities of things. In this discipline, philosophers try to answer questions about what exists, what attributes characterize it, and how to group such things. Transferred to Artificial Intelligence,
ontologies (with a small o) are computer artifacts that symbolize a particular type of knowledge. According to Gruber, “for the AI system, what exists is what can be represented.” An ontology therefore explicitly specifies the objects, concepts and other entities supposed to exist in a domain of interest and the relationships they maintain between them. This can be summarized by the first definition put forward by Gruber “an ontology is the explicit specification of a conceptualization”. An explicit specification of a conceptualization can be made by extension or intentionally. The extension specification lists all possible interpretations of the vocabulary elements used by the conceptualization to name its elements. This is not always possible if the universe of discourse or the set of possible interpretations is infinite. However, intentional specification constrains the intended meaning of vocabulary items using a set of appropriate axioms. The set of these axioms captures the intended interpretations corresponding to the specified conceptualization and excludes unintended interpretations.
On the other hand, an ontology is an axiomatization of the meaning of a vocabulary used by a conceptualization of a domain of interest. The manner of axiomatization led Uschold and Gruninger to give a continuum of ontology types (Figure 4.1). The spectrum expresses the expressiveness of the semantics as well as the formality of the
ontologies which increase from left to right. We refer to the two poles of the spectrum as “weak semantics” and “strong semantics” respectively. On the weak side, we can express very simple semantics; on the strong side, we can express arbitrary and complex semantics. An ontology therefore goes from a simple set of terms with little or no explicitness to a simple notion of taxonomy (knowledge with hierarchy or minimal structure), to a thesaurus (words and synonyms), to a conceptual model (with a lot of knowledge), to a logical theory (very rich, complex, coherent and very important).
Knowledge in
ontologies is formalized using five types of components: concepts (classes), relationships, functions, axioms and instances. Ontology concepts are generally organized into taxonomies. Sometimes the notion of ontology is diluted, in the sense that taxonomies are considered complete
ontologies.
Concepts (classes) are used in a broad sense. A concept can be anything about which something is said and, therefore, could also be the description of a task, function, action, strategy, reasoning process, etc. For example, Means of Transport, Bus and Local Bus are concepts in a travel ontology.
Relations represent the types of connections between concepts in the domain. For example, the arrivalPlace relationship connects Travel with Location in a travel ontology. Related concepts are related types of arguments. Subclass and part-of are special, domain-independent relationships. The subclass relationship captures generalizations in a hierarchy allowing the main concepts of a domain to be understood in a parsimonious manner. For example, the concepts Bus, Local Bus and Means of Transport can be organized in a hierarchy as follows: Local Bus Subclass Bus and Bus Subclass Means of Transport.
Two features of subclass relationships deserve special attention, because they provide us with useful additional information about the meaning (or "semantics") of a subtype relationship. These two features are:
1. Disjunction. A subclass relationship is "disjoint" if each instance of the superclass belongs to at most one subclass. If multiple subclass participations by a single instance are possible, we call this subclass relationship “overlap”. For example, Local Bus and Shuttle are separate.
2. Completeness. A subtype relationship is "complete" if every instance of a superclass participates in at least one subclass. For example, Local Bus and Shuttle form a complete classification of the Bus class. If participation in the subclass is optional, the subclass relationship is called partial.
Functions are special relationships that show a particular connection between an argument and other arguments. For example, the function Trip-Price(trip distance, vehicle Brand, serives Set(amenities)) calculates the price of a trip based on the trip, vehicle type, and amenities served during the trip.
Axioms are used to model sentences that are always true. Examples of axioms, every means of transport has a starting point, Bus is a kind of Means of transport and Local Bus is a Bus whose starting place, destination place and stops are in the same Location.
Instances are used to represent elements. A1287 and Ahmed are instances of the concepts travel and person respectively.
Depending on the subject of conceptualization, ontologies can be categorized into the following classes: Representation ontologies, high-level ontologies, general ontologies, task ontologies, method ontologies, and domain ontologies.
Representation ontologies: they capture the representation primitives used to formalize knowledge in a given knowledge representation paradigm. The most representative examples are the frames ontology (http://www.aiai.ed.ac.uk/project/enterprise/enterprise/ontology-code/frame-ontology/), OKBC (http://ksl.stanford.edu/DAML/aqua/web/okbc-ontology/
) and semantic web languages such as RDFS (https://www.w3.org/TR/rdf-schema/) and OWL (https://www.w3.org/TR/owl-features/).
General ontologies : used to represent a common sense of reusable knowledge between domains (e.g. events, time, space, bahavior, mereology). The ontology of Mereology (https://plato.stanford.edu/entries/mereology/) is one of the most classic examples of a general ontology. It defines the Part-Of relationship and its properties.
High-level ontologies : describe very generic concepts which other ontologies specialize in. Location and spatial point are two high-level concepts. SUMO (http://www.adampease.org/OP/) is an example of a high-level ontology.
Domain ontologies : they are reusable in a specific field (medical, pharmaceutical, engineering, law, business, automobile, etc.). A domain ontology provides a vocabulary for naming the concepts of that domain and their relationships and expresses the elementary theories and principles governing that domain. There is a clear boundary between domain and high-level ontologies. Concepts in a domain ontology are usually specializations of concepts already defined in high-level ontologies, and the same can be true for relationships. The travel agency concept of the travel ontology specializes the agent concept of the high-level SUMO ontology.
Task ontologies: describes the vocabulary linked to a generic task or activity (diagnosis, planning, classification, etc.) by specializing the terms in high-level ontologies. A task ontology provides a systematic vocabulary of terms used to solve problems associated with tasks that may or may not belong to the same domain.
Method ontologies: they provide definitions of relevant concepts and relationships applied to specify a reasoning process to achieve a particular task. The domain diagrams in Chapter 3 are examples of method ontologies.
Application Ontologies: Often extends and specializes the vocabulary of domain and task ontologies for a given application.
