RE: multiple inheritance
Bill Andersen wrote:
--------------------------------
Patrick Cassidy wrote:
> The point of an ontology is to create a compact and easily maintainable
> representation of knowledge, and multiple inheritance is a means to make
> the representation more compact than it would be with single inheritance.
That may be the point of "an ontology", but it's not the purpose of
"ontology". The purpose of ontology is to try to get the answers to
be "right" to the extent they can be. Neither compactness nor ease of
maintenance are players in this definition.
Frame systems, DLs and the like were introduced to make knowledge "more
compact" and "easily maintainable", as you suggest. I believe that both
have done more harm than good, not from a computational standpoint, but
from a methodological standpoint - they have seduced us into believing
that *all* we need are frames or DL. Another version of this is that we
*must* have multiple inheiritance. Why would someone make such an
(ontologically) non-obvious assertion? Because they intend subsumption
to do a lot of work and they can't do that work unless they have MI.
> The other main reason for multiple inheritance is to make it easier to
> generate agreement on a standard upper ontology. In merging existing
> ontologies, it will probably be necessary for some groups to give up
> some features that they consider useful, but if using multiple inheritance
> can allow different *consistent* views to be include in the same
> ontology, it will ease the pain of transition and increase the likelihood
> of being able to create a standard that can be widely used.
This is totally unsubstantiated. It makes agreement harder if
anything, by introducing more and more complex interpretations for
the subsumption relation.
> Does anyone actually classify a castle under "bunch of bricks"? ;-)
Sure - have you ever seen a vase classified as a "lump of clay" or
some such?
...bill
--------------------------------------------------
Pat Cassidy replied:
Concerning your points on multiple vs. single inheritance:
> The purpose of ontology is to try to get the answers to
> be "right" to the extent they can be. Neither compactness nor ease of
> maintenance are players in this definition.
It goes without saying that accuracy and correctness are the
main criteria. It does not follow that ease of maintenance is
incompatible, or even that it gets in the way. In most
cases, ease of maintenance would be expected to help
in maintaining accuracy by allowing one to make changes only
once for certain purposes, and avoid forgetting to make them
in places where they are needed.
> Frame systems, DLs and the like were introduced to make knowledge "more
> compact" and "easily maintainable", as you suggest. I believe that both
> have done more harm than good, not from a computational standpoint, but
> from a methodological standpoint - they have seduced us into believing
> that *all* we need are frames or DL. Another version of this is that we
> *must* have multiple inheiritance. Why would someone make such an
> (ontologically) non-obvious assertion? Because they intend subsumption
> to do a lot of work and they can't do that work unless they have MI.
Two commentators have suggested that multiple inheritance
may cause sloppy ontology builders to make more mistakes than
they would otherwise. But it can also be expected that the improved
compactness of ontologies with multiple inheritance can make such
systems easier to build and understand, and therefore less prone to error.
Is there in fact any evidence that large ontologies using multiple
inheritance will have more *total* errors than those without?
>> The other main reason for multiple inheritance is to make it easier to
>> generate agreement on a standard upper ontology.
> This is totally unsubstantiated. It makes agreement harder if
> anything, by introducing more and more complex interpretations for
> the subsumption relation.
It will be quite impossible to "substantiate" such an expectation,
without building comparable upper ontologies with and without multiple
inheritance and then seeing which becomes more widely used --
an unlikely scenario. This expectation is based mainly on discussions with
individuals who have built ontologies with different structures.
If everyone else would accept *my* notions of how an inheritance tree
might be structured, I too might be willing to accept a single-inheritance
system, but . . .
>> Does anyone actually classify a castle under "bunch of bricks"? ;-)
> Sure - have you ever seen a vase classified as a "lump of clay" or
> some such?
NO, except metaphorically or in jest. Such linguistic games will
have to be handled by linguistic interpreters, not by the ontology.
Pat
cassidy@micra.com