RE: SUO: Peer-to-peer interoperability of ontologies
>That view of ontologies is consistent with the traditional practice
>of lexicographers. Even as large a dictionary as the OED is considered
>to be *descriptive* rather than *prescriptive*. It records all the
>word senses that have been used, and now that it has been computerized,
>it continues to add new word senses as they develop.
>
>
>> The solution I advocate, at a technical level, is to view each
>> ontology as an evolving repository continuously incorporating new
>> meanings from interaction with other parties just as needed to make
>> sense of the ongoing exchanges.
>
>
>That is consistent with the practice of lexicographers who develop
>dictionaries for human use. The ontologies used by computers must
>have more precisely stated definitions, but they must be able to grow
>and adapt to current usage by both their human and their computer users.
Are we not neglecting something here? When the human uses a dictionary, they
use it in the context of vast amounts of prior information. When a computer
uses an ontology it can use it in a context of FOL (in which case the
definitions would most certainly have to be precisely stated). But this
doesn't have to be the case. Jean-Luc appears to be advocating morphing,
maybe 'organic' ontologies, but why not let the applications do the
morphing? In the context of the semantic web I would suggest that this will
be particularly useful from where we stand now. Contradictions will
sometimes occur (unless the domain is shrunken below 2 statements), then
what do you do? Maybe relax, look somewhere else, find the conclusions
suggested by each case... whatever.
The early colour theories were a way away from what is accepted as
scientific fact nowaday, but the painters that followed these sciences were
still capable of good work. Even if the ontological palette isn't God's own
RGB or CMYK (both essentially arbitrary choices that happen to be good with
human vision)... ok, SUO has burnt umber instead of red, this isn't a major
drawback as long as the applications form their opinions based on good
information (the painter uses her eyes).
There are ontologies and applications. If both are rigid then the system
will find a locked groove and stay there (ok, perhaps chaotically). If
either can be fuzzy (using a flexible interpretation) or woolly (using a
flexed term) then we can get somewhere. Don't get me wrong, I see no reason
not to try Jean-Luc's path, I simply can't think offhand how one might
implement such a system. Rigid definitions, flexible interpretation is
pretty easy to do using any number of certainty factor/probabilistic etc etc
approaches.
Cheers,
Danny.