Folks,
I would like to post the following piece, taken from a project proposal that unfortunately failed (I'll tell you the reasons why...), which lists some reasons for having axiomatic ontologies. I believe that, while answering Pat's concerns, we have to keep three things distinct enough:
1. Reasons for having axiomatic ontologies 2. Reasons for having upper-level ontologies 3. Reasons for having (a small set of) common upper-level ontologies
Note that points 1-3 above are ordered in terms of priority (or importance, if you prefer).
I don't think Mill's piece is SO bad, he clearly has a vague idea of upper-level ontologies, a vaguer idea of axiomatic ontologies, and is definitely against common UO. Explaining him the role of axiomatic ontologies is probably more important than convincing him on the utility of a common ontology....
Cheers,
Nicola
The need for axiomatic ontologies In most practical applications, ontologies appear as taxonomic structures of primitive and composite terms together with associated definitions. These are the so-called lightweight ontologies, used to represent very simple semantic relationships among the terms used by a specific community in order to facilitate partial access to the relevant information content. In this case, the intended meaning of primitive terms is assumed to be fairly well known in advance by all members of the community. Hence, the role of ontologies is that of supporting terminological services (inferences based on simple taxonomic relationships among terms) rather than explaining or fixing the intended meanings. Gradually, however, it is becoming clear that this purely terminological focus is too superficial; that the terms are not, in fact, so well-understood, and that the codification of knowledge on their basis leads to characteristic families of errors when it comes to the point where the codifications need to be processed by a computer. On the other hand, the need to establish precise agreements as to the meanings of terms becomes crucial as soon as a community of users evolves, and even more so when we need to cross the cultural and linguistic boundaries between different communities. In this case, in order to overcome terminological and conceptual ambiguities we need to express explicit logical constraints (i.e., axioms) underlying the use of terms. In this way, we can rely on model-theoretic semantics to help provide quality-assurance in relation to a proposed account of the formal structure of a domain to be represented, in ways which are independent e.g. of the intuitive meanings of the terms adopted. Finally, a further advantage of an axiomatic approach is that its intrinsically modular structure allows us to isolate those axioms responsible for incompatibilities among different ontologies, explaining therefore the reasons for non-interoperability. Recognising these reasons in advance may often be of the utmost importance. Lightweight ontologies, in contrast, may lead to simplistic data integrations where general semantic disagreements are just hidden. Building and reasoning with axiomatic ontologies is unfortunately extremely hard, both conceptually and computationally. However, this job only needs to be undertaken once, i.e. at the time of choosing the most appropriate ontology one commits to (development time). What is important in this phase is making clear to humans (besides computers) the intended meaning of terms. This is why a highly expressive language (e.g., FOL) will be desirable, independently of the computational constraints bound to the way the chosen ontology will perform at run-time once implemented in an application. Ontology-dedicated languages such as OWL may be the best choice for this latter case. In conclusion, the use of axiomatic ontologies to make explicit hidden ontological assumptions may drastically affect the trust in a computer service, but not the computational performance of the service itself. Thus, for example, a product procurement process involving multiple agents with distributed lightweight ontologies may be carried out in an efficient way by using simple terminological services, but the risk of semantic mismatch can be minimized only if the agents rely on explicit, axiomatic ontologies, ensuring mutual compatibility of the respective models in such a way as to check the extent of real agreement.
----------------------------------------------------------------------------- Nicola Guarino Co-Editor in Chief, Applied Ontology (IOS Press) Head, Laboratory for Applied Ontology (LOA), ISTC-CNR Institute for Cognitive Sciences and Technologies National Research Council Via Solteri, 38 I-38100 Trento
phone: +39 0461 828486 secretary: +39 0461 436641 fax: +39 0461 435344 email: guarino@xxxxxxxxxx web site: http://www.loa-cnr.it
|