DAML Intent of Work March 2001
University of West Florida/IHMC
The original UWF/IHMC SOW discussed the need for developing protocols for transactions between 'Semantic AGEnts', ie systems whose primary function is concerned with meanings: translators, authenticators, inference validity checkers and so on; and with the need to extend conventional model-theoretic semantic models to cope with these developments. Experience with the development of DAML+OIL over the last year suggests that this need is not so pressing, however, and that the 'semantic web' is moving towards a rather more limited scenario, where namespace conventions for example are standardized to avoid the potential complexities that would require interventions by SAGEs of the kind discussed. We therefore envision work during the next year with a sharper focus: somewhat more limited theoretical scope, but possibly in more technical depth.
The effort will fall into three main areas.
1. Web Semantics and trust brokering.
Contrary to widespread belief, we do not see the use of ontologies on the web as needing any major advances in logic itself. Conventional notions of truth, assertion and validity apply just as well to multi-agent situations as to single-agent ones, and any semantic complexities that may arise can be handled using conventional metamathematical ideas (for example, see 3. below). However, there are other issues which need to be addressed.
The chief challenge which the 'semantic web' poses to conventional model-theoretic accounts of meaning, is that these have paid no attention to issues of trust, warranty and security, and the dual problem of assigning blame for failure: what might be called tort logic. Logic has traditionally been wholly concerned with simple assertions of truth, and has very few mechanisms available for describing any nuances of 'degrees of warranty' other than modelling these, inappropriately, as degrees of truth. There are tensions already in the discussions surrounding RDF which clearly arise from a dissatisfaction with the simple use/mention distinction which is assumed in most logical notations, where the only alternative to asserting something is to quote it (in one form or another). This seems like the most pressing issue for the next year in extending logical semantics for the web, and it is also of course central to many 'agent' systems.
We propose to attempt to make progress here by extending the ontology rather than the logic. That is, we will create (in DAML+OIL) a 'brokering' ontology concerned with the issues of agency, trust and warranty. The concepts described in the brokering ontology should enable other ontologies to refer issues of responsibility to a DAML+OIL reasoning engine so that it could decide, using conventional logic, where blame should be attributed for failure of a warranted decision. (This work will be done in informal consultation with Jeff Bradshaw and his colleagues, who are now at IHMC; and in informal consultation with the DAML-funded group at MIT.)
We expect that this will require creating several other ontologies of general utility, including ontologies for temporal relationships and time-tagging, ontologies for events and processes, and ontologies for the notions of agency, responsibility and permission.
2. Translations from industrial standards
There are a several large-scale 'industrial' ontology standards (usually not called by that name) in existence, and we propose to spend some effort providing translators (implemented in LISP or Python) from some of the relevant 'data modelling' languages into DAML+OIL. In particular, we will translate the European Process Industries STEP Technical Liason Executive (EPISTLE) Core Model ECM V3.1 (http://www.stepcom.ncl.ac.uk/epistle/ ) into DAML+OIL from EXPRESS-G. This is a very general-purpose basic ontology providing a set of approximately 20,000 concepts used in many process industries and of widespread utility to industrial applications.
In the same spirit, we will try to locate other pre-existing 'standard' ontology frameworks which are likely to be of broad utility and which have been developed by industrial consortia, standards organizations and so on (*not* by academics) and attempt to translate their underlying formalisms into DAML+OIL. (We plan to do this work in informal consultation with the DAML-funded sites at Stanford, SRI and CMU, as well as the relevant external organizations.)
All of this ontology development will be to some extent a deliberate test of DAML+OIL expressiveness, and we expect it to reveal expressive deficiencies in the current (March 2001) release, and help put them into sharper relief. However, our primary goal for the next year is to work around any deficiencies by the best means that can be found.
The Knowledge Interchange Format (KIF) is widely used as a 'standard' format for first-order axiomatisations. It has, however, never been fully standardised. As part of a volunteer effort to create an up-to-date version of KIF and present it for IEEE standardisation, I have been working with Chris Menzel (Texas A&M) on a revised version of KIF which overcomes many of the perceived expressive problems of KIF 3.0, and has a uniform, rigorously defined semantics. The language (which in its full form - rarely used - is quite dauntingly complex) will be redefined as a series of 'extensions' each one adding extra syntactic and semantic complexity, to allow users to choose the sublanguage appropriate for their intended use. The 'core' language will be simple first-order logic; extensions introduce a class heirarchy, syntactic sorting, namespaces and URIs, sequence quantifiers, and meta-descriptions.
IEEE-KIF development has been influenced by DAML+OIL in several ways. It will have DAML+OIL-compliant provision for namespaces and ontology importing, and will incorporate a class-heirarchy sublanguage similar in expressive power to DAML+OIL. (The current draft does not include XML datatypes.) We intend to submit this document for formal IEEE approval during the next year, probably in the fall of 2001. I include it here because of the relevance of KIF to the DAML+OIL effort.
One example of this relevance has already arisen in the IEEE-KIF development. KIF uses sequence quantifiers which provide an essential expressive ability which enables it to describe its own syntax. However, traditional accounts of KIF do not provide a fully worked-out semantics for sequence quantification, and we have discovered that when this is done rigorously, the full language is equivalent in semantic expressiveness to a sublanguage of an infinitary extension of first-order logic in which countably infinite conjunctions and disjunctions can be considered legal lexpressions. This means that KIF is strictly more expressive than first-order logic; and moreover, this extra expressiveness is crucial to its reflexive powers of self-description. This poses a logical problem, since the extra expressiveness cannot be supported by a complete mechanical inference process in its full form. We are investigating computationally feasible sublanguages to find those which provide the most useful meta-expressive functionality which could support something lke the RDF notion of "reification".