Re: alternative interlingua

Danny Bobrow <bobrow@parc.xerox.com>
Message-id: <4akq3ekB0V0wM1RDFi@nero.parc.xerox.com>
Date: Fri, 10 Aug 90 18:35:38 -0700 (PDT)
From: Danny Bobrow <bobrow@parc.xerox.com>
To: bobrow@parc.xerox.com,
        pfps@allegra.tempo.nj.att.com (Peter F. Patel-Schneider)
Subject: Re: alternative interlingua
Cc: interlingua@vaxa.isi.edu
In-Reply-To: <9008101725.AA03130@arisia.Xerox.COM>
References: <9008101725.AA03130@arisia.Xerox.COM>

Excerpts from mail: 10-Aug-90 alternative interlingua P.
Patel-Schneider@alleg (3567)


> I have the following assumptions about how the interlingua is to be used:
> 1S/ No information is to be lost when translating into the interlingua.
> 2S/ When translating into the interlingua, no knowledge of the system that
>    will eventually be translating out can be used.
> 3S/ When translating out of the interlingua, no knowledge of the system that
>    the information was initially represented in can be used.

This is really useful because it highlights a basic difference in our
assumptions.  Let me repeat some of what I said last night, contrasting
my assumptions with yours (labelled above, 1S, 2S and 3S)

1B/ When outputting into the interlingua, it should be possible to use
forms close to those used by the native language; call these NL
extensions.  Information about how to interpret these NL extensions with
respect to an agreed kernel  be publically available (e.g. by being in a
named library or put in a preamble to the file).  Because the
interlingua output is in a form close to the NL original, no information
need be lost.  However, the kernel translation of the output may not be
fully equivalent to NL original.  The kernel translation must of course
be implied by the interlingua output (i.e. necessary conditions when the
output is true), but need not include all sufficient conditions, so
recognition may not be possible. 

2B/ There is knowledge of the minimal capabilities of any system that is
to use the Interlingua.  This includes knowledge both of the logic
supported and some the inferencing capabilities (this is implicit in the
design of the Interlingua)

3B/ When translating out of the intelingua, there can be arbitrary
amounts of knowledge of the system (or sets of assumptions) that went
into creating the Interlingua file.  Therefore, when moving a KB from
one Loom system to another, it is possible to recapture completely what
was meant using the Interlingua.  Any particular receiver, can build
special purpose translators to capture Loom expressions but need not.

As we get experience with various special forms and their translations,
we might decide to add cabilities to the kernel to ensure that some
expressions can have complete (necessary and sufficient) translations
into the kernel.  This supports an incremental style in the evolution of
the Interlingua Kernel.