Knowledge-Based Systems 15 (2002) 147±167 www.elsevier.com/locate/knosys On the insuf®ciency of ontologies: problems in knowledge sharing and alternative solutions Flavio S. Correa da Silvaa,1, Wamberto W. Vasconcelosb,*, David S. Robertsonb, Virginia Brilhanteb, Ana C.V. de Meloa,1, Marcelo Fingera,1, Jaume AgustõÂc,2 aInstituto de MatemaÂtica e EstatõÂstica, Universidade de SaÄo Paulo, Rua do MataÄo, 1010, 05508-900 SaÄo Paulo, Brazil bDivision of Informatics, University of Edinburgh, 80 South Bridge, EH1 1HN Scotland, UK cInstitut d'Investigacio en Intel.ligencia Arti®cial, Consejo Superior de Investigaciones Cientõ®cas (IIIA/CSIC) UAB, E-08193 Bellaterra, Catalonia, Spain Received 10 October 2000; accepted 2 February 2001 Abstract One of the bene®ts of formally represented knowledge lies in its potential to be shared. Ontologies have been proposed as the ultimate solution to problems in knowledge sharing. However even when an agreed correspondence between ontologies is reached that is not the end of the problems in knowledge sharing. In this paper we explore a number of realistic knowledge-sharing situations and their related problems for which ontologies fall short in providing a solution. For each situation we propose and analyse alternative solutions. q 2002 Elsevier Science B.V. All rights reserved. Keywords: Ontologies; Knowledge sharing; Knowledge representation and inference 1. Introduction formalisms of different systems [52]. However, as pointed out in Ref. [16], even when an agreed correspondence One of the bene®ts of formally represented knowledge between ontologies is reached that is not the end of the lies in its potential to be shared. The opportunity presented problems. In this paper we explore a number of realistic by computer interconnection technologies such as the Inter- knowledge-sharing situations and their related problems net is for locally produced knowledge bases and inference for which ontologies fall short in providing a solution. For mechanisms to interact in solving problems which are more each of such problems we propose and analyse alternative complicated than each individual system could tackle on its solutions. own. However, in many such situations we have to face The structure of this paper is as follows. In the rest of this problems concerning how we can ensure, or at least assess, section we further discuss our aimed class of knowledge- the reliability of different forms of cooperation, given that based systems (KBSs, for short) and their properties; we each system may have been developed by different people also present a parallel between knowledge sharing and and are likely to be based on different languages and form- object-oriented technologies (Section 1.2) and a brief alisms. discussion on the existing proposals for the communication Ontologies [24,33,35,52] have been proposed as the ulti- among KBSs (Section 1.3). We present the organisation of mate solution to problems in knowledge sharing. Ontologies the rest of this paper in the ensuing paragraphs. provide explicit mappings between shared concepts from Even when a single language can be employed to repre- sent the messages exchanged among the systems, this by no means implies that a single theory can be constructed for the * Corresponding author. Tel.: 144-131-650-2718; fax: 144-131-650- 6513. shared ontologies among those systems. This simpli®ed E-mail addresses: [email protected] (F.S. Correa da Silva), situation is obviously desirable and should be pursued [email protected] (W.W. Vasconcelos), [email protected] whenever possible. In Ref. [24] we ®nd the proposal of an (D.S. Robertson), [email protected] (V. Brilhante), [email protected] `ontology server', to act as a tool to construct large collec- (A.C.V. de Melo), m®[email protected] (M. Finger), [email protected] tions of shared ontologies. In Section 2 we address the dif®- (J. AgustõÂ). 1 Fax: 155-11-3818-6134. culties inherent to this approach. 2 Tel.: 134-9-3-5809570; fax: 134-9-3-5809661. In Ref. [28] we ®nd the differentiation between KBSs and 0950-7051/02/$ - see front matter q 2002 Elsevier Science B.V. All rights reserved. PII: S0950-7051(01)00152-6 148 F.S. Correa da Silva et al. / Knowledge-Based Systems 15 (2002) 147±167 knowledge base repositories, being the latter used to store KBSs are best presented in terms of logical systems, declarative knowledge and information, akin and subject to although this does not necessarily mean that they have to the same care requested by a database to maintain integrity be implemented as such. If we concentrate on deductive and accessibility. Sharing declarative knowledge (e.g. logics we can ®nally achieve the conceptualisation of a contained in knowledge base repositories) can be quite KBS that we intend to adopt throughout this work. We diverse from sharing reasoning (i.e. deductive) knowledge, shall regard a KBS as a deductive theory, written in a parti- as shown in Section 3. cular logic language with well-de®ned formal semantics, Modelling the content of the messages to be exchanged which includes the expected soundness and completeness between pairs of KBSs is a semantic problem. Since the results for the deductive theory with respect to the corre- construction of the corresponding logical theory of the inter- sponding semantic theory. A query is a logical conjecture lingua is carried through at the syntactical level, it can posed to the deductive theory, whose truth evaluation is happen that some important semantic issues become invisi- justi®ed by proof-theoretical means. ble in the implementation, nevertheless they may be still KBSs are built for `practical' problem-solving, hence the requirements to maintain consistency of the interpretation semantics of their corresponding deductive theories must be of messages. This issue is further analysed in Section 4. convincing enough as models for the problems being solved. The knowledge sharing strategy for the construction and Furthermore, the proof generation procedures must be application of KBSs may lead to more and more distributed appropriately implemented as ef®cient pieces of software, and fragmented networks of systems. Individual KBSs may and they must be appropriately wrapped as usable pieces of thus become smaller, and it may happen that a request not be software, with accordingly well-engineered interfaces. solvable by a single system, but still be solvable by an Building a full ¯edged KBS can be a highly costly endea- appropriate combination of operations from many systems. vour, as it requires a deep analysis of the system's problem In this case, a knowledge broker is demanded to split a domain and problem solving procedures, the engineering of single request into a collection of requests, gather the corre- an appropriate logical language for the system, the model- sponding operation execution results and assemble them to ling and formal reconstruction of the problem and its satisfy the demanding client. This situation is explored in solving procedures in terms of the semantics of the corre- Section 5. sponding logical language. It is also required that the actual Finally, in Section 6 we draw some conclusions and construction of the deductive theory which models the discuss the results presented. speci®c problem and procedures are taken into account, and, of course, the whole software engineering business 1.1. Knowledge-based systems: issues and features related to implementing the system, including testing, validation, code optimisation and interface development. Knowledge sharing is closely connected to knowledge- A historical account3 of the development of KBSs shows based systems, which are, on their turn, founded on knowl- that the ®rst systems constructed bene®ted from the exis- edge. The most commonly used de®nition of knowledge is tence of each other only with respect to the developed skills justi®ed true belief, as found in, for instance, the work of and generic methods in use, that could be passed away and Delgrande and Mylopoulos [18]. Hence, KBSs are computer exploited in the construction of the following systems. Little software systems that work with justi®ed true beliefs. We of the accumulated domain knowledge, deductive theories can expand this a bit, and propose that a KBS has as its and underlying languages could be inherited from one content collections of information structures Ð representa- system to another. tions of beliefs Ð for which we de®ne formal interpretations that grant them the status of true beliefs. Moreover, this 1.2. Knowledge sharing and object-oriented systems status can be checked by means of (most often deductive) proofs, that provide them a formal justi®cation. Recent developments in software construction technol- A large number of formal languages exist for KBSs, each ogy have suggested that a more ef®cient approach to the being chosen according to its adequacy for the representa- development of KBSs could be taken, that would lead to tion of the knowledge related to speci®c domains, ease-of- cost-effective construction of higher-quality systems. use, popularity, and existing supporting tools. For instance, Namely, techniques related to distributed object oriented bayesian and probabilistic languages [43] are convenient for systems development and implementation have in¯uenced uncertain knowledge; linear logics [30] are convenient for the development of KBSs towards knowledge sharing representing knowledge whose inferences may be resource- [24,27±29,41]. bounded, and so on. Attempts to provide heterogeneous A well accepted standardised architecture for implement- sources of knowledge with the ability to cooperate have ing distributed object oriented systems is the Common arisen in different contexts. Some projects related to this Object Request Broker Architecture (CORBA) proposed idea can be found [32,42,50]. A more formal counterpart to these proposals, oriented to formal speci®cation of soft- 3 This can be inferred from the reported experiences of development of ware, can be found in Refs.