Submitted by Steve Sawyer. Send corrections to Bob Engelmore. ATTENDEES: Bob Spillers IBM - Santa Teresa Lab Hiroshi Suematsu EDR Japan Michael Gruninger University of Toronto Bob Engelmore KSL - Stanford University Fan Hsu IBM - Santa Teresa Lab John Sharp Sandia Lab Ed Hovy ISI - University of Southern California Anthony Sarris Ontek Bart Emanuel IBM - InfoMarket John Sowa Binghamton University Takano Ogino EDR Japan Adam Farquhar KSL - Stanford University Jim Fulton Boeing Frank Olken Lawrence Berkeley National Lab Steve Sawyer Syracuse University Lee Auspitz TextWise Nancy Lawler Dept of Defense Andras Kornai IBM - Almaden Lab Tony Magliero IBM - InfoMarket Monique Sugimoto EDR Japan ------------------------------------------------------------- Minutes of the X3T2 Ad-Hoc Group for Ontology Standards Held at IBM Santa Teresa Laboratory, San Jose CA on March 4 and 5, 1996. Bob Spillers (IBM) hosted the meeting. The agenda (see attached) was broadly defined. The morning of the first day focused on outlining the purposes of this Ad-Hoc group, ontological issues to discuss, and the relationships between various ontologies. This discussion broadened to debate what standard for ontologies should be based on (and will include). The consensus was that any reasonable standard will address a class of ontologies that can exchange information through first order logic (FOL). Terminology (actually more standard concepts, versus the terms to present them) is an issue, as certain ontologies have reasons for precisely defining terms, sometimes in a specialized sense. Further, questions were raised such as: what constitutes a registerable ontology? Who registers them? How do they get shared? Discussion continued and consensus was reached that the ontology expressive mechanism should initially be limited to FOL, as with the CSMF project. Existing languages such as CGs, KIF, CycL, Loom, etc. (and to a lesser degree SQL and Express) would be candidates for conforming languages. Higher-order logics would not be precluded, but should not be part of the initial standard. Embedded in this discussion was a constant relation of these issues to the ongoing ontological work of STEP. Following a lunch break, discussion of the use of such a standard began. It was felt the standard was only valuable if it was received well by the community as this needed to be a consensus standard and consensus is a function of community. As a first step. an approach was discussed that would, initially, not start off with a set of fundamental concepts, i.e. to make no real ontological claims about the world, but limit the concepts used to define the middle level to linguistic concepts initially (i.e., predicates, nouns, etc.) and to use FOL as the syntactic mechanism. The middle level would embody a set of reference concepts. The second day began with a renewed discussion on what value an ontology standard would have. Reasons include: o Commercial vendors want a baseline for developing products, o Improve retrieval efforts, o Provide modeling support, o Support interchange between ontologies, o Provide for interoperability, and o As a basis for education. The primary target usage of the standard will be for new ontology creation (e.g., CYC was intended to solve the knowledge acquisition bottleneck problem. A second primary use is to support integration of existing 'ontologies', whether formal or informal, natural language or some other expressive form such as a database language. Finally, for information retrieval, navigation and 'mining', such as support for inferencing/pattern-recognition and interpretation. Through the day, the various sub-topics were discussed, issues were detailed, components, deliverables, data, and volunteers were related. These are presented as Appendix Two. In this discussion the roles of the standard was repeatedly discusses in the light of new perspectives. What arose from this constant focus is that the standard might be conceived as being two logical components: (1) a framework/specification (a reference model) and (2) a reference implementation that is the composite of current major mid-world linguistic-based ontologies. Use would be first as a comparison and second for interrelationship. For example: WordNet has 70K words; EDR has 400K. If a significant number of highly useful equivalent words in EDR were aligned with WordNet, it would be easier for people who use either EDR or WordNet to relate at some level to another dictionary ontology that contains different definitions for some of the same words, but also has other words. This led to a discussion of technical' versus 'natural language' ontologies. There were enough differences in the form to consider these as important subsets of ontologies. However, a natural language ontology was considered as the more encompassing (and thus, target) aspect of any reference? It was also agreed that, in conducting the work on developing an ontology standard, there should be two parallel efforts: a short term and a long term. The short term focus is on practical use while the long term is on theoretical fulfillment. The difference between short-term and long-term thrusts is degree of completeness. They may be precise in what they specify, but they do not specify all that could be specified about any given concept. NOTES: Adam Farquar (from Stanford KSL) has put his set of ontology tools on the web, along with Gruber's previous Ontolingua ontology. Bob Engelmore of Stanford KSL is acting as web/mail host. Next meeting of the Ad Hoc group is June 27-28, 1996 (tentative), followed by another meeting along with rest of X3T2 (in late August or September). The project could be formally approved at any of these meeting, but the work can go on in the interim anyway. The plan is have a draft by the next X3T2 meeting. Appendix One -- The Flip Chart Topics of Day One Purpose and Scope Why an Ontology Standard 1) First Order Logic 2) Why : Commercial forces Interchange Interoperability Education Retrieval Modeling support Education Topics for an Ontology Standards Committee's Interests (who's interested) 1. THEORY Lawler, Sowa, Hovy, Gruninger, Olken, Spillers, Farquhar, Kornai, Auspitz, Fulton, Sarris. EDR, Skuce Principles of Ontology Creation What is an ontology Extension / intension (re: Sowa) General methodology of ontology building / maintenance Principals of extensions Methodology/principles of conflict resolution - for resolving between ontologies 2. LOGIC AND FORMALISM Kornai, Ausptiz, Sowa, Farquhar, Gruninger, Emanuel Some theory lists + what's there 3. CONCEPT LIBRARY EDR, Sowa, Engelmore, Gruninger, Spillers, Hovy, Farquhar, Sarris, Kornai, Olken. Skuce, Auspitz 4. TESTING Farquhar, Hovy, Lawler 5. TOOLS Farquhar, Engelmore, Skuce, Kornai, Auspitz 6. STANDARDS Spillers, Sharp, Fulton, Sarris, Hovy Appendix two: The Flip Chart Topics of Day Two (Task Assignments) Standards Document What it is How it talks to other ontologies Instance(s) of conforming standards Approach (difference is not rigor, but completeness) The long term The short term LOGIC and FORMALISM deliverables 1. Produce Spec. for KIF/CG (etc) Sowa. Genezeth - draft now, "final" June 96 2. Tutorial / description Sowa, Farquhar, Genereth, Olken, Grunefeld - before Sept 96 examples KIF. CG desiderata: (1) Not to preclude higher-order logic (2) Able to be extended to handle probability, confidence, etc. (3) Encoding for other langauges (EUC, ?, future extension) (4) Partial protection mechanisms CONCEPT LIBRARY Content upper structure: 300 language middle structure: 80,000 technology domain models (specialized ontologies): variable Examples 1. upper (Pangloss, UC, CYC) 2. middle (wordnet, EDR, CYC) 3. technology (STEP, SIC, ...) Deliverables (Spillers, Hovy, CYC, EDR) 1. upper Mar 96 (Hovy) 2. upper <-> WN Mar 96 (Hovy) WN <-> CYC ? (Spillers/Hovy) CYC <-> upper ? (Spillers/Hovy) EDR <-> upper ? (Spillers/EDR) all by Jun 96 3. Sample link up tech ontology - middle ontology (Farquhar/Hovy) TESTING & VALIDATION Components particular strategies for conflict resolution requirements (size, number, org.) compliance tests Examples Work of others (Xerox, MIT, Berkely, USC, others) Deliverables Prototypes(Farquhar / Kornai) Measuring tools/insturments Formal validation checker TOOLS Components (examples) 1. Browsing/ web services (Farquhar, ISI, Skuce...) 2. Editing / extension (Farquhar, Skuce...) 3. Cross-ontology alignment (Knight - ISI) 4. Consistency checking 5. Validation tool Deliverables 1. Browsing/editing tool (Farquhar) Mar 96 2. Candidate API's for interchange (Farquhar, Kornai) 3. Draft API for tools