A brief reply to two of Neil’s points:

>> [NC] It seems illogical to me to try to capture all knowledge in a single ontology, just as it is ridiculous to capture all facts about a domain in a single flat-file "database".

[PC] Agreed. That’s why I am suggesting that the most productive way to think of a foundation ontology is as the set of primitives that can be used to specify the meanings of the ontology elements used in any application. The applications can be divorced from each other, highly specialized, and highly diverse. But when the applications need to share stored information, the description of that information using a common “defining vocabulary” (the foundation ontology) will allow the machines to automatically and accurately interpret the intended meaning of the information, and use it appropriately. The set of primitives will be very much smaller than the number of datatypes or relations described using those primitives.

>> [NC] - With so many viewpoints of an ontology's construction and purpose: Pick one benefit and push the construction methodology to the limit to further that particular benefit--perhaps some other natural benefits may fall out as side effects.

[PC] This relates to the need for some “killer app” to demonstrate an ontology’s utility. I think that one such application would be a good interactive natural language interpreter that can be used for various purposes, such a NLP interface to data, help functions, semantic search. The number of potential applications for a common computer language may be comparable to the number of ways people use human languages. But developing a natural language understanding system may well take very much more effort than developing a common foundation ontology. So it is not likely to *precede* development of a common FO.

Interestingly, having a common FO could greatly accelerate the development of a powerful natural language interpreter, by serving as the common standard of meaning through which different modules of a language interpreter could communicate the results of text processing by many different methods. Statistical and rules-based methods can communicate, and also use highly specific “experts” that are tuned to specific words, phrases, or grammatical structures. By use of a common foundation ontology as the standard of meaning, a powerful system that exceeds the capabilities of individual groups could be developed by multiple groups in a collaborative fashion, where incremental improvements in replaceable modules can be developed to evolve a language understanding system from some promising starting point. But for such a process to work the foundation ontology has to come *first*, and then the app can be created. The same process could serve to permit development of other knowledge-based systems by a collaborative method allowing cumulative improvements. The absence of such a standard impedes efforts to use the collaborative technique, because *some* standard for representing information needs to be used, and in the absence of an existing broadly used standard, local groups would have to first coordinate to decide on what standard they will use for a particular project. This occurred in the CALO project. The Porter ontology used there is competent, but idiosyncratic. Who else uses it other than the participants of that project?

Another “killer app” is database integration. Cycorp is making most of their money these days from such a project (commercial, not government-funded). Ontology Works will do integration via an ontology for you, and so will Adam Pease, using SUMO. The problem is, that retrospectively mapping different databases to each other, even via an ontology, is very expensive, and is cost-justified only in some cases. To properly understand the value of agreeing on a common foundation ontology *first*, one needs to take a longer view than worrying about retrofitting existing databases. In twenty years, it is likely that the majority of databases in use at that time will have been built after 2009. So, proactively developing databases whose data elements are described ontologically by a common foundation ontology will allow all users of such DB’s to share their data accurately with each other, with development of the DB no more expensive that by a traditional method of creation of new schemas. At that point, those legacy DB’s created before the standard can still be integrated, but at a price that may not be justified. So we can envision a “triage” situation where newly created databases interoperate from the start via the FO, important legacy databases are mapped to the FO, and legacy databases whose need for interoperability is too little to justify the cost will never be able to interoperate – at least until they are refactored, in which case they may be integrated at that time for little additional cost, once there is experience using the FO for DB integration. We have been discussing interoperability issues for over fifteen years, so taking a 20-year perspective makes a lot of sense. But again, for this scenario to play out, the foundation ontology will have to come *first*.

We can solve the chicken-and-egg problem by noting that dinosaurs were laying eggs long before they evolved into chickens, so actually the egg came first. And I believe that the foundation ontology will have to come *first*, before the killer app is developed.

In any case, considering the enormous ongoing waste due to lack of any common standard of meaning for information, spending 30ドル million in a serious effort to explore one possible solution seems like a very worthwhile investment. And if other direct approaches (not waiting for some evolution over decades or centuries) are proposed, they should be funded too.

Pat

Patrick Cassidy

MICRA, Inc.

908-561-3416

cell: 908-565-4053

cassidy@xxxxxxxxx