In this paper, we investigate the relation between Guarino's seminal paper \Formal Ontology and Information Systems" and the DOGMA ontology-engineering framework. As DOGMA is geared towards the development of ontologies for semantic interoperation between autonomously developed and maintained information systems, it follows that the stakeholders in this project form a community and adds a social dimension to the ontology project. The goal of this exercise is to examine how the di erent terminologies and ideas relate to one and another, thus providing a reference for clarifying DOGMA's ideas and notation inside Guarino's framework.
An ontology is commonly de ned as: \a [formal,] explicit speci cation of a
[shared] conceptualization" [?]1. Ontologies constitute the key resources for
realizing a semantic Web. The main di erence between a conceptual schema and
an ontology is that the rst is intended for the development of one particular
information system in one organization and the latter for reuse and therefore
general for a particular domain [
Gruber's de nition was based on the de nition of Genesereth and Nilsson's
notion of a conceptualization [
Over the past years, quite a few (collaborative) ontology-engineering
methods have been developed, each with their own characteristics; e.g., the formalism
adopted, approach of agreement processes, application domain, etc. [
The goal of this paper is to investigate how the di erent terminologies relate to each other given the fact that Guarino's work aims to provide a de nition for ontology in computer science and a collaborative ontology-engineering framework for establishing semantic interoperability between autonomously developed information systems. The paper is organized as follows: Section 2 provides a summary and explanation of Guarino's ideas and de nitions, Section 3 describes the problem of semantic interoperability between autonomously developed and maintained information systems, Section 4 describes the notion of the fact-oriented collaborative ontology-engineering framework DOGMA, Section 5 discusses the relation between the two, followed by a conclusion in Section 6. 2
This section explains the terminology used in Guarino's most in uential work [
De nition 1 (Extensional notion of conceptualization [
However, Guarino argued in [
With the notions introduced by Guarino, the structure hD; Ri introduced
in [
Given a logical language L with a vocabulary V , Guarino provided an extensional and an intensional interpretation of the language.
De nition 5 (Extensional interpretation of a language [
De nition 6 (Intensional interpretation of a language [
Guarino argued that the set of intended models only provide a weak characterization of a conceptualization as an intended model does not necessarily re ect a particular world and can even describe a situation common to most worlds. Indeed, if one would take for instance a domain space with a domain of the following persons D = hP eter; Louis; J oe; Bonniei and where all words contain all possible marriage con gurations as well as all persons classi ed according to their gender (Peter and Joe being male, Louis and Bonnie female). If one takes as set of conceptual relations only the unary predicates for being male and female, the intended model for each world { via an I an appropriate I { will be the same.
Because an intended model can describe a situation common to many worlds, it only rules out absurd situations. One can merely create a set of axioms for a commitment K and a language L such that the models of the ontology approximates as close as possible IK (L). This set of axioms is what Guarino calls an ontology. Ontologies thus indirectly re ect an ontological commitment by approximating the intended models of that commitment. 3
Now that we have presented Guarino's ideas and de nitions, we will present the di erence between closed and open information systems. As we will see, for both systems an ontology will be needed that will replace the real word and hence one can state that the \same" exercise is done at an enterprise level for the former and at a domain level for the latter. Developing ontologies for open information systems, however, will prove to be more challenging due to the di erent perceptions of reality and use of language (e.g., jargon) used by the stakeholders representing the di erent autonomously developed and maintained information systems.
An information system is a system containing information in a database for a given application context of a given organization. The application context denes the functionality of such a system, which is prescribed by the organizations' requirements. The development of an information system thus involves the creation of a requirements speci cation and an agreement on the design. Costing is an important factor here and will also in uence the choice whether components will be selected for its implementation, outsourced, or even build from scratch. One should involve end-users during the requirements speci cation process for several reasons. Two of these reasons are the impedance mismatch between the jargon (used by end-users) and the business knowledge, and the end-users being experts on (their part of) the domain.
As shown in Fig. 1, domain experts and end-users observe the world. Domain experts try to abstract the world, whereas the end-users will interact with the world and are able to test the developed information systems by comparing the instances stored in the information systems with objects in the real world. Both domain experts and end-users will collaborate with a knowledge designer so that the latter can put the resulting agreements into a CASE tool to build a conceptual schema that described the business domain. The conceptual schema will { in turn { be used to generate parts of the processes, parts of the constraints and a data model of the information system.
The knowledge in such a conceptual schema is typically a mixture of domaingeneral knowledge and enterprise-speci c knowledge derived from the requirements. Domain knowledge can contain constraints that are shared or generic. Enterprise-speci c knowledge describes the applications and constraints local to the enterprise making use of the domain knowledge. Enterprise-speci c knowledge will often constrain the domain-general knowledge even further. For example, the knowledge that an ISBN identi es a book is part of the domain knowl"Real world" Business Domain
Observe/ Abstract
Designer AGREEMENT
(N.L.)
Business
Domain Expert Observe/Interact => Test by instances
End users
ENTERPRISE CONTEXT - DEFINED BY REQUIREMENTS Design Tool
Abstraction from instances Communicate at instance level
Conceptual
Schema
Apps Information System
DB Schema DBMS DB edge. In an enterprise providing a movie rental service, however, the knowledge that a customer is only able to lend at most ve movies at a time is part of the enterprise knowledge.
The conceptual schema, often in the shape of a formal (mathematical) construct, actually replaces the business domain, as the business domain is not accessible inside a computer. This is necessary in order to store and reason about semantics the business domain. The formal semantics of an information system is then the correspondence between this system and the conceptual schema, which represents the business domain as perceived by the domain expert and the endusers. Once the system is adequately designed and implemented, a statement output by the information system can be correctly interpreted by end-users in terms of objects in the business domain if and only if such statement is derived from stored instances of fact types described in the conceptual schema. Those stored instances are mapped by the intensional semantics to observed relationships among those same objects.
In Fig. 1, the conceptual schema corresponds with the ontology in Guarino's terminology (i.e., the intensional description of the concepts and relations) and the database provides the extensional account of a particular situation.
But what happens when two or more autonomously developed and
maintained information systems need to interoperate and yet need to remain
autonomous2? The business domains of each information systems relates to the
shared domain knowledge of all humans involved. In order for these systems to
interoperate meaningfully, agreements on the domain knowledge by the
represen2 In general, semantic interoperability is de ned as the ability of two or more
information systems or their (computerized) components to exchange data, knowledge or
resources and to interpret the information in those systems [
ONTOLOGY
Shared World
Community
Agreement
leads to
Interaction Enterprise IS 1
Enterprise IS 2 tatives of these systems { which we will call a community { are needed. Again, as the world is not accessible in each one of those information systems, the shared domain needs to be replaced by another formal (mathematical) construct, called an ontology. As shown in Fig. 2, each autonomous information system has its own database which each provides an extensional interpretation of the conceptualization approximated by the ontology. In other words, each database of an information system can be regarded as a possible world. Fig. 2 also depicts that a community of stakeholders interacts to achieve agreements on what perceptions they share. The approximation of the shared reality that is shared among those stakeholders is then stored as an ontology, which will in turn be used the replace the shared world as the real world is not immediately available from the information systems. 4
In [
De nition 8 (DOGMA Ontology Descriptions). A DOGMA Ontology Description is an ordered triple h ; ci; Ki where is a lexon base, i.e. a nite set of lexons. A lexon is an ordered 5-tuple h ; t1; r1; r2; t2i where 2 is a context identi er, t1; t2 2 T are terms, and r1; r2 2 R are role labels. A lexon is a binary fact type that can be read in two directions: t1 playing the role of r1 on t2 and t2 playing the role of r2 on t1. Here, the usual alphabets for constructing the elements of T [ R are omitted for simplicity. ci : T ! C is a function mapping pairs of context identi ers and terms to unique elements of C, a nite given set of concepts. K is a nite set of ontological3 commitments. Each commitment is an ordered triple h ; ; ci where is a selection of lexons from the DOGMA ontology description, : ! T is a mapping called an annotation from the set of application (information, system, database) symbols to terms occurring in that selection, and c is a predicate over T [ R of that same selection expressed in a suitable rst-order language.
Context-identi ers are pointers to the origin of a lexon, and helps with the disambiguation of term- and role-labels. Within a context 2 and t 2 T , ci( ; t) is the de nition itself of the concept agreed by all users.
We furthermore make a distinction between community commitments and
application commitments [
As noted by [
Lexons are plausible fact types, which means that if one can think of (an) application(s) for such a lexon, they may be entered in the lexon base. The sets T and R for term- and role-labels in lexons correspond to the predicate symbols in V . The context-identi er provides an interpretation from terms to concepts. Since lexons are entered when they are plausible for one or more applications, the context-identi er actually corresponds to Guarino's interpretation function I. In other words, if one selects in the lexon base all lexons holding in a particular context with context-identi er , one is able to reconstruct Guarino's interpretation function I: all concepts x referred to by ci( ; t) (for each term t in those lexons) will refer to the interpretation of a unary predicate.
DOGMA's is based on ORM [
The combination of the predicates with quanti cations and connectives already reduces the set of all possible models with those satisfying above constraints. The fact-orientented formalism already provides some constraints that restrict the intended models of the language; i.e. it leaves out some absurd situations. In DOGMA, role is identi ed by its context, role label, term and co-term. The roles in each of the following lexons are thus di erent: { hPerson Context, Person, with, of, Namei { hPerson Context, Dog, with, of, Namei { hPerson Context, Person, with, of, Agei { hProject Context, Person, with, of, Namei
Indeed, a group can argue that the rst and second lexon in the example above are referring to the same notions of \having a name", but such an agreement would actually be a constraint on these lexons that will be captured by community commitments that we will describe later on. Also, the rst and last lexon are deemed to be di erent since the term- and role-labels are residing in di erent contexts. Unless explicitly speci ed, the same labels in di erent contexts do not necessarily mean that those labels are referring to the same concepts.
A commitment k 2 K of the DOGMA Ontology Description corresponds with an ontology from Guarino's framework. It is a selection of lexon from the lexon base that is constrained such that it approximates as good as possible the domain it aims to describe. Those constraints correspond with the notion of axioms and typically include notions such as: type- and role hierarchies, totality constraints, uniqueness constraints, value constraints, etc.
Value constraints are interesting to note that they limit domain elements for the interpretation of concept referred to by a term. There are two types of value constraints: exhaustive sets and descriptions of sets. We start with an example of the rst type of value constraints: giving the term Size referring to the size of a drink at a fast food restaurant in the Fast Food context, one can constraint the instances of that concept to the following values f"Small", "Medium", "Large"g. These instances of values then would refer to the concepts of these di erent sizes in that context. The latter type { e.g. a regular expression { checks whether an instance complies with the condition and are therefore considered restraining the use of a unary predicate.
A community commitment further restrains all possible models of the lexons committed to. An application commitment will even further restrain those by providing additional lexons, constraints, and narrowing down all possible models by providing additional constants via the mappings.
Since the real world cannot be stored in a computer-based system, one needs to replace that real world with a database (and corresponding database schema) in order to reason about things in the real world. We assume that a database of an organization corresponds with one real world. The mappings provided by a commitment k 2 K are used how the constant symbols in a database are related in aforementioned predicates. For instance: assuming that one has the lexon hFast Food, Soda, with, of, Namei and a particular fast food organization has a table SSS containing records about di erent sodas with information such as the name in the eld NNN, one can provide an extensional account for the predicates "Food", "Name", "with" and "of" by adding the following mapping: MAP "SSS"."NNN" ON Name of Soda. and all symbols corresponding to the SQL query generated with this mapping will be used to populate the predicates with constants (e.g., "Pepsi", "Sprite," ...).
It is interesting to note that when relating DOGMA with Guarino's seminal work, the ontological commitment in DOGMA ontology descriptions does not allow constant symbols, but that the ontologies in the DOGMA ontology description do so for constraints and via the mappings.
In summary: { Guarino's vocabulary V would correspond with the union of DOGMA's T , R, values in value constraints in each K and symbols from records of each database described in the mappings of each K via queries. This vocabulary is then used for the language L. { Guarino's ontology corresponds with the constraints of one commitment k 2 K. In the case of a community commitment, the intended models are narrowed down with those constraints. In the case of an application commitment, the intended models are even further narrowed down. This is the case as the constant symbols of a database are then also used. Note that a database does not necessarily describe one world, but could even describe situations plausible for many worlds.
We can even argue that the mappings provided in an application
commitment are { albeit indirectly { and even further restriction of the intended
models, i.e., the mappings for a constraint on their own.
{ Guarino's seminal work is not particularly clear on the ontological language
L. On one hand it seems to be richer than the mere use of the predicate and
constant symbols of V , but on the other the constraints on those predicates
are left to the ontological commitment such as demonstrated in
[
From above, it follows that one needs to decompose the commitments and combine pieces with the lexon base to reconstruct Guarino's ontological commitment. In other words, there is a high cohesion between ontological commitments and ontologies in the DOGMA ontology-engineering framework. 6
This paper presented a thought exercise on relating a framework for collaborative ontology engineering with Nicola Guarino's seminal work "Formal Ontology and Information Systems". This exercise allows us to provide documentation to the reader on the DOGMA framework and might even clarify some notions presented by Guarino. As the DOGMA framework is mainly used for establishing semantic interoperability between autonomously developed information systems, the relation of the symbols in databases to concepts denoted by labels in the ontology was "new" compared to Guarino's framework. The work presented in this paper will furthermore allow us to revise or clarify some of the de nitions we presented in the past.
Acknowledgements This work was partially funded by the Brussels Institute for Research and Innovation through the Open Semantic Cloud for Brussels Project.