Joint Workshop for the Use of Models that Define the Data and Processes for Information Systems

Sarris Abstract

Dear Jim(s):

Attached is a contribution for the joint workshop coming from me as an industry expert. This is based on the call for participation you issued in late March. I also intend to ensure that an ISO CSMF position is officially drafted during our upcoming (May 13-24) meeting, and that an international expert is appointed to represent ISO CSMF.

X3T2, as the US national body (TAG) to ISO CSMF, and the ANSI ASC standards body for domestic work on KIF, CGs and a linguistic ontology [proposed] *may* chose to submit a separate national body position. We met just before your calling notice came out and there was considerable preliminary interest in both the topic and potential contributions to the meeting. However, I have recommended that we as a committee wait until after the ISO meeting to prepare any possible US national body positions, if that is acceptable to you. There are a number of potential developments that could affect that position and I would like to ensure that whatever direction we take, we are as current and well-organized as possible.

I would also be happy to review papers if I could be of service in that regard.

Please feel free to contact me if you have any questions or comments.

Abstract for Expert Presentation to Joint Workshop on Data and Process Modelling

The subject of this workshop---primarily approaches, techniques and tools for describing, integrating and utilizing enterprise models---is complex exactly because it represents the intersection of several different information standards and technologies. From ways to conceptually model enterprise semantics in the form of data, processes and rules, to neutral languages and fundamental constructs for integrating heterogeneous models, to CASE and repository/data dictionary technologies for effective model management and reuse, the problems are different aspects of the same underlying problem: knowledge representation. Artificial intelligence is not a panacea in the sense that some claimed in the 1980s. But advancements in modeling require a concerted effort to focus resources on flexible, expressive languages and plug and-play sets of domain concepts---'ontologies' in the current terminology. Adaptable, integrated tools are required to support such efforts.

A brief survey of current standards efforts in this area indicates that while there is some direct overlap, more often than not the projects can placed relative to one another within a common framework. They can be shown to be heading at different speeds and with specialized thrusts toward an overall goal that reflects the following key characteristics:

1. Heterogeneous modeling forms that excel in their specialized focus, but which can also be connected to other forms to produce a unified model, or at least to interchange semantics that have meaning in more than one modeling context. This requires syntactical interchange as a first step. The syntactic level of interchange, as well as some static semantics (based on E-A-R modeling) has been largely addressed by standards efforts such as CDIF and IDEF IDL. However, more powerful, expressive languages (based on symbolic logic) are needed for process/dynamic semantics and constraints or rules.
2. Distribution of enterprise models across intra- and internets. The location and source of the models can no longer be predetermined. Basic distribution (in terms of protocols for messaging and some application interoperation across networks) is being handled through COBRA IDL, OLE and other object-based networking approaches.
3. Natural language interfaces that allow users to describe their enterprise in a way that appeals to them as domain experts, as well as enabling models produced in specialized technical formats to be presented in formats more oriented to natural language. With the exception of some work involving NIAM, CGs and linguistic ontologies, this area has suffered from the problem of trying to take on too much at once: namely, the issue of natural language interpretation, which is far from being solved, and which may well never be fully automated.
4. Registries of formal and informal ontologies available for access in the public domain and for use in various levels of model (semantic) integration, i.e., model mapping, alignment and unification. This presupposes tools for knowledge representation, model development and model management, including CASE tools and repositories/data dictionaries.

   The two major efforts in the repository field are IRDS and PCTE. These need to be closely coordinated to ensure that the structured data/DBMS community and the software/scientific and technical data communities have compatible standards. A major concern in this area is that repository standards should not be dictating their own content model format; rather they should store and manage KR-based models using a neutral storage format (e.g., OODBMS/multi media). Repositories and CDIF-based CASE tools must work closely together.

   The only major standards effort addressing the issue of knowledge representation/ontology (in general) is the CSMF project. It does not simply deal with model form, but rather with the underlying/common constructs used to represent all entities/objects, processes and constraints/rules. The current effort addresses an initial 'starter set' of these constructs. The CSMF work is based on the ANSI/SPARC and ISO three-schema data management architecture, which was originally developed many years ago, but is currently being 'rediscovered' by a number of communities. STEP is, of course, the largest standards project targeted at producing an application domain model. As a forerunner in the modelling and KR/ontology fields, the STEP community has had to develop some of its own approaches and techniques in cases where there were none available to utilize at the time.

5. Inferencing techniques, paradigms and tools ('engines'). While the techniques may be commonly found in specialized technical circles, practical application of the techniques in the form of COTS inference engines is lacking. Advancements in this area depend on languages and ontologies, as well as implementation environments in the form of CASE tools and repositories/data dictionaries.

This paper will explore the characteristics described above in more detail, assessing the state-of the-art in each area and proposing future directions. Emphasis will be on the KR/ontology aspects of the ISO Conceptual Schema Modelling Facilities (CSMF) project, which the author proposes is the farthest reaching of the projects in terms of addressing the overall knowledge representation issue.

Cheers,

"The problem with Data Mining is that with all the current hype, you might end up just getting the shaft."