Andrea,
When I said “consistent”, I meant it in the formal logic sense:. If I propose all the axioms in the two theories, I cannot infer a contradiction.
Now, I may import an ontology that describes a single general concept and has many subclasses and properties that I can use, and other aspects I don’t care about. It is very unlikely that incorporating that ontology will produce inconsistencies, as long as I agree with the statements about the elements I intend to reuse.
But consider, for example, the case in which I import an ontology for time. It will have a very small number of undefined terms with characterizing axioms, and quite possibly a rather large set of well-defined terms for derived concepts, some of which I use. Again, I can see pretty clearly whether the parts I use are consistent. But suppose the time ontology says that a time interval can be started by an event, and it provides a number of axioms that characterize its “event” notion. In my ontology, I want to use that idea, but I have an elaborate model of events and activities. Now I need to be sure that the imported axioms for ‘events’ (or whatever symbol I equate) don’t contradict my axioms, directly or indirectly. If there is a contradiction, we don’t mean the same thing by “event”. I can still use the imported ontology if I have some class that satisfies the imported event axioms (a “mapping”), but the import is only useful if I agree that that, or some subclass of it, is the class of things I want to use as the start of time intervals. Further, if my intended use is a subclass of the imported “event” concept, I need to know that admitting instances of the imported class that I don’t intend to use as the starts of time intervals won’t contradict something I say about activities and time. In particular, consider the case where the imported ontology says that every event starts the time interval in which that event “has happened”, and implicitly all subintervals thereof that start at the same instant. If his ‘event’ includes things that my model says have a non-zero duration, and I consider those ‘events’ not to “have happened” until that duration has elapsed, there are some small intervals in which the imported model says the event ‘has happened’, but my model says the event ‘has not happened’ but ‘is happening’, which is a different state. Then there exist events e and time intervals t in which my theory says (not (hasHappenedin e t)) and the imported ontology says (hasHappenedin e t), which is a contradiction.
There is clearly an underlying inconsistency in intent, but it is only a logical inconsistency if I care about the specific behaviors of things that are accidentally included in his somewhat larger category. This is a version of Pat Hayes’ “Horatio principle” – “there are more things in (my) heaven and earth, Horatio, than are dreamt of in your philosophy” (Hamlet, Act I).
As Gary (?) pointed out, this is why you would really want these ontologies to be as modular as possible. If the imported ontology for “events and time” includes, but is separate from, the imported ontology for “time”, I can import the latter without creating the problems engendered by importing the former, even if I really wanted the kind of thing the former covers, and have to develop that part separately. And it would be a different kind of “reuse” if in that ‘separate development’ I take his “events and time” ontology and MODIFY it to be consistent with my events and activities module. I think this latter is a pretty common form of ‘reuse’. (I am reminded of Tom Lehrer: “Don’t let others’ work evade your eyes, ... but plagiarize, plagiarize, plagiarize (only be sure always to call it, please, ‘research’).”* J)
-Ed
* from “Lobachevsky”, Tom Lehrer, 1960.
Gary and Ed, There are two bullets in the reuse discussion where I (somewhat) disagree ...
- the content is consistent with the micro-theory adopted by the re-user
- the re-user is able to determine that the content is consistent with his/her theory
Yes, I guess that we might look for structural consistency which was perhaps handled in the conversion
process mentioned previously, the logical consistency (check with a reasoner?) and consistency with the
user's conceptualization.
I don't think that the content must be consistent, but the content must be mappable or translatable.
This takes us back to Hans' point about understanding the assumptions and context of the original content ... Just as more discussion showed that the events in Pascal's talk and in FIBO were semantically close (if not equivalent), it is important to somehow enable a similar line of reasoning. We need to understand how and why some model was created/defined as it was, and then other alternatives/possibilities that the model enables.
The problem here (as Cory noted) is time and money to create the information or have the dialog. Is this something that could be crowd-sourced?
--