Re: heterogeneous DBs
phayes@cs.uiuc.edu
Message-id: <199309131833.AA01739@dante.cs.uiuc.edu>
Sender: phayes@dante.cs.uiuc.edu
Date: Mon, 13 Sep 1993 13:36:27 +0000
To: interlingua@ISI.EDU
From: phayes@cs.uiuc.edu
Subject: Re: heterogeneous DBs
Cc: brayman@zuben.ca.boeing.com (Bill Brayman)
Bill Brayman writes, in response to Jeffrey Van Baalen:
>The semantics are implicity in
>the applications that use the databases. Our approach is to try to
>factor out the semantics whereever they are. However, representing any
>semantics beyond simple data structures, though, takes more
>representational power than any declarative language yet devised with
>the possible exception of KIF and Conceptual Graphs.
>
>The real issue is that these so-called semantics must be encoded in an
>executable form today, else databases wouldn't be practical.
>Consequently, we find that in practice, most semantics must be captured
>in software procedures, not the database.
I'm not speaking for Jeffrey Van Baalen, but I suspect that there is
a misunderstanding over 'semantics' here. Brayman uses 'semantics' to refer
to what in AI is often called knowledge representation, ie some implemented
system which somehow encodes knowledge in a formalism: semantics-1. Others
- perhaps Van Ballen - use 'semantics' to refer to something like model
theory, ie an account of how these formalisms could relate to a possible
world: semantics-2. The former is a computational achievement, the latter
is a theoreticians' tool. In the latter perspective, declarative languages
such as KIF and CG and database languages don't constitute semantics: they
are things that have, or can be given, a semantics. And semantics, in this
sense, are not things that one would usually expect to find implemented in
executable form. (Although interesting things might be achieved by so
doing, that would not be doing knowledge representation. For example,
inference can be usefully viewed as a process of showing that models don't
exist by trying to construct them sufficiently systematically.)
A reason that many AI people are 'shocked' by a lack of semantics-2 is that
unless such an account is provided, there is no exact way to say what the
expressions of a representational formalism might mean, so talking about
translation between one and onother, for example, becomes difficult to make
precise enough for their tastes. This applies just as well to the
formalisms used in the DB itself as to those proposed to encode their
semantics-1; in fact to anything in a computer which is supposed to somehow
encode knowledge.
My aim is not to start another quarrel, but to try to avoid confusion. Can
we find a terminology which we can use to all communicate with one another
without misunderstanding? Let me tentatively suggest that we might speak of
'semantic representations' for semantics-1 and 'semantic interpretations'
for semantics-2. This allows me to slip naturally into 'representation' and
'interpretations'. Is it similarly acceptable to someone with a database
background?
Pat Hayes
----------------------------------------------------------------------------
Beckman Institute (217)244 1616 office
405 North Mathews Avenue (217)328 3947 or (415)855 9043 home
Urbana, IL. 61801 (217)244 8371 fax
hayes@cs.stanford.edu or Phayes@cs.uiuc.edu