Ontologies for Agents: Theory and Experiences

Reconciling Implicit and Evolving Ontologies for Semantic Interoperability (p. 121-122)

Kendall Lister, Maia Hristozova and Leon Sterling

Abstract. This paper addresses current approaches to the goal of semantic interoperability on the web and presents new research directions. We critically discuss the existing approaches, including RDF, SHOE, PROMPT and Chimaera, and identify the most e.ective elements of each. In our opinion, the ability of these primarily closed solutions to succeed on a global web scale is limited. In general, a unilateral solution to the problem on a global level seems unlikely in the foreseeable future. We review and contrast our own research experiments AReXS and CASA and suggest that as yet unaddressed issues should be considered, such as reconciling implicit ontologies and evolving ontologies and task-oriented analysis. We also consider the role of semantic interoperation in multi-agent systems and describe strategies for achieving this via the ROADMAP methodology, with emphasis on building and assuring knowledge models.

Keywords. Ontology translation/mapping, Ontology maintenance/evolution, Data standardisation.

1. Introduction

The much talked about goal of building a new Internet that is comprehensible to machines as well as humans is generally considered to involve enhancing content and information sources with semantic markings and explicit ontologies. A number of approaches to this goal have been proposed, and these generally involve a new representation for semantically enriched data. Something that seems to be often overlooked, however, is that a single solution is unlikely to be usefully applicable to the entire world wide web. It is obvious that business needs are generally quite di.erent to the needs of individuals, and that even within the business community di.erent areas will require solutions of varying sophistication, accuracy and scale. The widespread success of the world wide web and its underlying technologies, HTML and HTTP, has been due in no small part to their simplicity and ease of adoption. By providing a simple architecture that anyone could learn and use with minimal overhead, content .ourished on the web. Other information technologies that arguably provided more e.ective methods for locating and retrieving data failed to take o. in the same exponential way that the web did.

Where the web infrastructure itself doesn’t even contain the most rudimentary searching and resource location features, Gopher, WAIS and a large number of proprietary online databases that predated the world wide web all provided automated indexing, searching, hypertextuality and other information management capabilities. But despite their apparent advantages, all of these technologies were overtaken by the web. In fact, in many cases proprietary databases and indexes have had their interfaces replaced with web-based solutions, to the point that the actual technology is largely hidden. It is more than a coincidence that where the world wide web succeeded and grew to become a de facto standard, the more complex alternatives faltered and missed out on popular adoption.

Similarly, we consider that the next generation of semantically-capable global information infrastructure will necessarily be relatively simple in order to achieve the same scale of acceptance. That is not to say that sophisticated technologies have no place - on the contrary, they will be vital for the areas of industry that require them, and their advances will no doubt drive other research e.orts even further. Also, the intelligent agents that roam this infrastructure will themselves be very sophisticated. However, there remains a fundamental role for simple, .exible and adaptive technologies that do not demand strict adherence to formal standards and protocols and the development and publishing costs that follow. By leaving the majority of the intelligence for semantic comprehension in the interpreting applications rather than the medium itself, we will develop technologies that can operate in any information environment, not just those that are sophisticated and semantically enhanced. There is no suggestion that semantically rich environments are not useful and desirable, but it is not practical to expect the entirety, or even the majority, of the information landscape of the future to be uniformly structured, as current research seems to imagine.

2. Current projects toward a semantic web

Discussions of the problems of semantic operability on the web have a tendency to become discussions of the problem of managing and integrating ontologies. The reasons for this are not obscure: ontologies are widely regarded as a critical element of the next generation of data integration solutions, and the world wide web is a heterogeneous environment in which foreign data (and therefore ontologies) are regularly juxtaposed. What is less clear is how such data can be combined. A number of new technologies have been proposed that extend or replace existing web technologies, prominent among these are RDF, SHOE, PROMPT and Chimaera.