Re: communication and representation
Ramesh S. Patil <ramesh@vaxa.isi.edu>
Message-id: <9009080002.AA04318@vaxa.isi.edu>
To: pfps@allegra.tempo.nj.att.com, Interlingua@vaxa.isi.edu
Reply-To: ramesh@venera.isi.edu
Subject: Re: communication and representation
In-reply-to: Your message of Fri, 07 Sep 90 16:43:41 -0400.
<9009072044.AA01071@vaxa.isi.edu>
Date: Fri, 07 Sep 90 17:02:36 PDT
From: Ramesh S. Patil <ramesh@vaxa.isi.edu>
Pat,
I think the issues you raise are important ones and require airing.
The differences between Ccode and Rcode are clearly motivated. But, as
messages from Peter and Mike indicate there is still a diversity of
views.
To begin with, it was never clear what the scope of interlingua
should be: if it was only a language of communication (natural language
for KR systems to talk to each other), then the issues of avoiding
paradox arising from quote discussed at length in our first meeting was
irrelevant. Secondly, I think that it was somehow assumed (without much
discussion) that the interlingua should have clearly defined semantics
independent of the two parties that engage in communication, and
independent of the order in which the sentences appeared in the message.
Finally, given that we assumed that there is a translation from the
source KR to interlingua and then from interlingua to the destination KR
system, which might have different constructs, different ways of
expressing and organizing knowledge content, the task of any translator
was then to take the knowledge expressed in interlingua and to
re-organize it in a form suitable for the KR which preserved the meaning
as much as possible (ideally without any loss of information). This I
think necessarily entails that the translator be able to reason with the
knowledge expressed in the interlingua because, the expression
available in the KR language may require that the content of the
interlingua be reformulated, i.e., decomposed into different set of
primitives and organized (or grouped) in different form so as to maintain
their original intent.
I think this sub-problem of translation requires substantially similar
reasoning capabilities that Rcode is usually expected to perform. Modulo
the fact that we may be willing to accept different computational or
real time constraint on it from that required for problem solving using
the knowledge. Thus, given the objective that the ability to translate
to and from interlingua is a consideration in the design of interlingua,
some of the issues of Rcode come into play.
Your observation that ccode is more like natural language also
suggests that the interpretation of a body of text (or message) is
likely to be context dependent, that is, depends on the perceived
intent, prior knowledge, mental model and/or cultural biases and
experiences of the speaker and listener!! NL is a marvelously
expressive in that it allows one to communicate with varying degree of
precision, allows one to express paradoxes or even nonsense but relies
on the background knowledge of the listener to interpret, understand and
extract the knowledge from it. For example, consider very carefully
written document such as US constitution, every court and legal scholar
interprets it differently and the interpretation continues to evolve and
be argued about even today, 200 years after the document first appeared.
However, if one of the goals of the writers of this document was to
assume or create a practical translator that would result in
semantically equivalent mental model in ever reader of this document,
they could not have written even one word!
Thus the issue of Ccode vs Rcode is tied to the question of "How much
reasoning capability is required for the translation component of
interlingua effort?" Also, the issue Peter raised about information
preserving vs information loss, allowing only sound translation vs.
allowing good faith effort (similar but not necessarily sound) in
translating. I think your thought provoking message has been very
useful in bringing these issues to the front.
Ramesh