Proposed standards for KIF and CGs (John F. Sowa)
Date: Wed, 29 Jun 94 21:45:37 -0400
From: (John F. Sowa)
Message-id: <>
To:, interlingua@ISI.EDU, kr-advisory@ISI.EDU, srkb@ISI.EDU
Subject: Proposed standards for KIF and CGs
Precedence: bulk
At last month's meeting of the ANSI OMC (Operations Management Committee),
the recommendations from the April meeting of X3T2 were approved:

 1. The name of the X3T2 committee has been changed to "Information
    Interchange and Interpretation."  The new name reflects the concerns
    for both information sharing (Interchange) and the semantics of the
    information (Interpretation).

 2. The two new work items for developing standards for the Knowledge
    Interchange Format (KIF) and for conceptual graphs (CGs) were
    approved.  Draft proposed standards for both of them have been
    promised by December 1994.  Mike Genesereth is the editor for the
    KIF proposal, and John Sowa is the editor for the CG proposal.

Besides these two new work items, X3T2 continues to have responsibility
for other work on conceptual schemas:

 1. The conceptual schema for data interchange, for which X3T2 has been
    authorized for several years.

 2. The conceptual schema for IRDS (Information Resource Dictionary
    Systems), which was transferred from X3H4 to X3T2 earlier this year.

 3. As a result of the ANSI reorganizations earlier this year, X3T2
    is now the official US TAG (Technical Advisory Group) to the ISO
    Working Group on Conceptual Schema Modeling Facilities (CSMF).

At the OMC meeting, some concerns were raised about authorizing new
work items at a time when there are declining resources -- i.e. many of
the companies that have traditionally been members of ANSI committees
have been cutting back on both travel and personnel.

The response to that concern was that KIF and CG standards are needed
to support the other assignments.  Both of them are sufficiently stable
now that draft proposals for them can be written on schedule.  Software
for both of them has been developed, and more is in progress.  The
proposed standards should ensure compatibility between KIF and CGs and
facilitate the development of prototypes that apply them to the more
ambitious task of defining and developing the CSMF.

Other concerns have been raised by groups that have been using and
developing tools based on other paradigms, including Entity-Relationship
diagrams, SQL, EXPRESS, IDEF1X, and various notations for representing
object-oriented designs and specifications.  Some of these notations
have been used for many years for modeling and specifying data.  And
most of them are currently being extended to support broad ranges of
applications and techniques.  People who have been using tools and
techniques based on these notations would be very unhappy if they were
forced to change to some new standard that would make their current
systems obsolete.

The response to that concern is that KIF and CGs are NOT being proposed
as replacements that would obsolete any system or notation that has
proved to be useful for any purpose.  Instead, they are simply a graphic
(CGs) and a linear (KIF) form for a de facto standard that has been
around longer than either ANSI or ISO -- namely, first-order logic.
Both of them contain the barest minimum needed for defining and
representing all existing paradigms:

 1. The basic Boolean operators:  and, or, not, implies, equivalence,
    and any others that are definable in terms of them, such as nand,
    nor, and exclusive or.

 2. The two basic quantifiers:  the existential and universal.

 3. Definitional mechanisms based on lambda calculus.

 4. Metalanguage capabilities that are necessary for defining themselves
    and all other formal languages, logics, and computing paradigms.

 5. Syntax, rules of inference, and model-theoretic semantics.

Everything else goes into the ontologies, which strictly speaking are
not part of either KIF or CGs, but which are definable in terms of them.
That includes ontologies for time, procedures, formal grammars, abstract
automata, Petri nets, algebraic structures, contexts, object-oriented
encapsulations, distributed agents, etc.  Defining ontologies for every
one of these systems is NOT part of the KIF and CG work items, but some
of them will have to be defined for the knowledge sharing effort and for
the ANSI and ISO CSMF project.

Advocates of various special-purpose systems, such as SQL, EXPRESS, etc.,
keep saying "Oh, but we are planning to extend our language to include
all that!"  But comments like that miss the whole point:  those languages
are already much too big, much too complex, and much too inflexible to
be extended further without becoming totally unwieldy.  Our approach is
not to add more, but to take out everything but the barest essentials.

Basic point:  Logic with definitional mechanisms and metalanguage has
all good things -- modularity, flexibility, extensibility, and object-
oriented structure.  Anything and everything else can be defined using
the extensibility features.

That does not mean that all the problems of the world have been magically
solved.  Instead, it means that logic provides the foundation in terms
of which any solvable problems can be solved.  All we are proposing to
standardize are the foundations, and we will work with anyone and everyone
who wants to build their systems on top of the logic-based foundations.

Other people have asked why we are proposing two different languages,
KIF and CGs.  The answer is that we are NOT proposing two languages.
We are proposing a single abstract system of logic.  Then we are offering
two different syntaxes for that logic:  KIF with an easy-to-parse linear
notation that is convenient for knowledge sharing between heterogeneous
systems; and conceptual graphs with a more readable graphic syntax
that has been designed for simplifying the mapping to and from natural
languages.  But anyone who prefers any other notation for any other
purpose -- ease of implementation, human factors, or whatever -- is free
to use it as an alternative syntax as long as it has a formally defined
mapping to and from the same abstract logic.

Proponents of SQL, EXPRESS, and other languages have asked why the
syntax of their language is not being proposed as the basic conceptual
schema language.  The answer is that anyone who prefers such a syntax
is ENCOURAGED to develop it to express the same abstract logic as KIF
and CGs.  As long as they can support exactly the same semantics, they
can use their favorite syntax as an embodiment of the abstract logic
on exactly the same footing as KIF and CGs.

In fact, the new work items for KIF and CGs will simplify the problem
of defining conceptual schema modeling facilities in terms of SQL,
EXPRESS, and any other language.  The main thing that the developers
of those languages need to do is to provide a suitable syntax that can
be mapped to and from KIF and/or CGs.  Then  the definition of that mapping
would also serve as a definition of the CSMF for the new language.

John Sowa