The way I was taught Wittgenstein was that the Philosophical Investigations
were a repudiation of Tractatus, and particularly of the idea that language
<b>pictures</b> the world, to replace it by the idea that language is what
we use to talk to each other and is founded on the "forms of life" we engage
in. The opening section of Investigations contrasts a passage where St
Augustine describes learning the names of things he naturally recognises
with a story about buying five red apples. In the latter - which is not
about Natural Language Processing, what ever Pat Hayes says - he explicitly
denies the need for pictures in the head. The remainder of the
Investigations can been seen as a series of therapies aimed to wean us off
the "pictures in the head" model of language. (01)
IMO the "pictures in the head" theory has its roots in Plato's theory of
forms. From the information processing view, its paradigmatic algorithm is
is pattern recognition, in which a signal produces a response in a matched
filter (the pattern recogniser) which fires when the convolution of signal
and filter crosses a pre-set threshold. The temptation of "pictures in the
head" arises from the way our brain presents patterns ready-recognised to
our conscious, so that when I look out of my window at an arrangement of
red, white and black, I see a house, with no need for further deliberate
investigation. (02)
In contrast, the paradigmatic algorithm for language is classification. I
have a collection of arbitrary symbols (words) whose forms in no way relate
to their usage - so "Kangaroo" might equally mean "Go Away" as naming a
hopping animal. In using a language to name something, I use classification
criteria that choose between alternative categories (and their names). For
example, I would use the term "bird" of an animal which had feathers and
walked on two legs. (03)
There remains a difficult problem of relating pattern recognition to
classification (i.e. I have no idea what the answer is). If there were a
fixed set of distinct patterns, then there would be no problem (and, for
example, machine vision would have been solved long ago). However, that
would also imply that there are a fixed set of "forms of life". The problem
of industrial data exchange is that local organizations evolve their own
vocabularies according to their peculiar business processes, so that
information sharing between companies is treacherous. The differences in
systems of military ranks between services and countries also exhibits the
same problem. (And, by implication, ontologies based on general language
usage provide a poor guide to the problems of developing ontologies precise
enough interoperation between professional organizations.) (04)
I note in this forum a recurring argument between on one hand the "one
upper ontology"/"finite set of basic concepts" school and the "no single
ontology" school. That is, in terms of the above, the "pattern recognition"
and the "classification" schools of meaning. Since ontology languages seem
oriented to supporting classification, it seems the former are confused. (05)
Sean Barker
Bristol (06)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Subscribe/Config: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx (07)
|