Len, Matthew, Pat C
1) I don't see the relevance of the frame problem.
It is not assertions that are changing over time, but the concepts of the
2) Yes, you can specialise a concept, but then it
is no longer a primative, and now the FO has changed. The point of an FO surely
is that it is stable, since interoperability depends on it. More specifically,
if if make a translation via an FO from one ontology to another, then I expect
it not to change from one day to the next.
3) If I have a translation A to B, and a second
translation B to C, then I can calculate a translation A to C. It goes not
matter if B is an FO or some other ontology. Once I have a translation from a
new ontology into one I use, then I have a translation to every ontology that I
need to communicate with. In conventional computing, the use of a data exchange
standard provides an advantage because each translator is a special purpose
piece of software, and with N applications one needs N(N-1) translators. With an
ontology, one needs only one general purpose reasoner, which can then be applied
to any collection of translation axioms, including the concatination of multiple
sets of translations. This is a major business reasons for looking at
In theory, one could use conventional data standard
translators concatenated, however there tends to be problems with
differences in scope between the standards, and different practices with respect
to the way the same facts are represented in the different standards. These
translators are relatively inflexible, and concatinating translators is usually
a recipie for data loss.
It is likely that different scopes for ontologies
would cause the same sorts of loss along a chain of concatenated translations.
Being able to run the translation though a common ontology could reduce this
data loss. However, my understanding of Pat C's "primitives" is that they do not
construe an ontology as such, but the "primatives" from which an ontology can be
constructed, and that a latice of ontologies can be created using different
combinations of "primitives". It would not, in general, be possible to translate
from one ontology in the latice to another, although it would be possible to
identify translatable subsets.
4) While in principle I cannot prove Pat C wrong,
for a large number of reasons, I think Pat's approach is a red herring - to
repeat points I made in previous e-mails:
a) There seem to be two approaches to language -
that language talks about the world, and that language pictures the world. The
culturally dominant assumption is that language pictures the world, and I
also observed this in US SATs tests. I have noted a lack of understanding
of the differences between the two views on the forum, which is not surprising,
since it requires one to deconstruct one's cultural assumptions, something not
on the curiculum of most science and engineering courses. These two approaches
seem unlikely to agree on what primitives are useful.
b) If one looks at the use of colour terms, one
observes that the systems of terms that occur in natural languages vary and are
incompatible with one another. Having a primitive concept "colour" would not
solve this problem, because it is the set of colours that is different.
Moreover, where we cannot even agree on terms bound to our physiology, this
reinforces the suspicion that there will be substantial differences in what we
take as primitives.
c) We know from data modelling that there are some
twenty to thirty types of data model incompatibility, most of which are, in
principle, not resolvable. I do not see that these
incompatibilities have been solved by ontologies.
d) The semantics of many business terms are defined
by the processes of the business. The specification of primitives for the
behaviour of organizations would be needed to specify the derived concepts, such
as product, which are the foundation classes of exisiting
e) Most dictionaries (I do not use Longman's, as it
happens) are semiotic, in the sense they provide the starting point for
procedures to infer the meaning of words.
5) While I am sympathetic to the goal of
improving the ability to translate between applications, I do not see
that having a single FO would help. Rather, I would expect to see a
series of domain specific ontologies which provide detailed interoperation over
that domain. The domains may be as extensive as the whole of manufacturing
industry (as ISO 10303)