COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	logica[e82,jmc]		Logic and ordinary language
C00011 ENDMK
C⊗;
logica[e82,jmc]		Logic and ordinary language

	These remarks are triggered by the first few pages of
Barwise and Perry "Situations and Attitudes", but the issues were
in my mind previously.

	When we propose to express formally information normally expressed
in natural language, we need to make a distinction between expressing
what is communicated and what the people know.  This is because only
a part of what is known is communicated.  Moreover, only part of what
is known can be communicated with present linguistic habits, and only
a part of what could be communicated is actually communicated.
As humanity and its cultures evolve, we learn to communicate more.  Most
likely the original languages could communicate much less than
modern languages, and it is worth investigating whether some primitive
languages have deficiencies in what can be expressed in them.

	What saves communication is that the recipient and the sender
have much knowledge in common which serves as the context of the
communication.  Moreover, if the recipient doesn't understand the
message correctly, there is often an opportunity to elaborate.

(Insert remarks about why we can't speak mentalese).

	Barwise and Perry begin their book

	"This book has a simple thesis: that the standard view
of logic derived from Frege and Russell and work in mathematics
is completely inappropriate for many of the uses to which it has
been put by philosophers, linguists, computer scientists and others,
that it is infected by ideas that may be compatible with mathematics
but which are not appropriate for the way we ordinarily use language.
But our form of argument is largely positive. We present an alternative,
what we have come to call "Situation Semantics," that we think does better."

	We don't continue to cite their arguments but instead defend
the thesis that first order logic is adequate to express any facts
expressed in ordinary language or storable in the brain.
The Barwise-Perry situation semantics, whatever it may be, can be
handled by using a predicate  true(statement,situation)  where the
first argument is whatever they use for assertions, and the second
argument is whatever they call a situation.  There is the objection
that ordinary language doesn't surround every assertion by such a
predicate, but this isn't a serious question for AI.  Besides, natural
language is ready to refer to a conext at the drop of a hat.  The only
problem is that the contexts explicitly referred to in natural language
are always partial or inner contexts and can themselves be taken in
more general outer contexts and these in still more general contexts.

	All this means for our theory is that
we must reify  true(Assertion,context)  to
True1(assertion,context1)  which again is an assertion that can
appear in

	true2(True1(assertion,context1),context2),

which can be further reified if this is required.

	Likewise with  value("my wife",s),  further reification
may be required in some contexts.

Some properties they allege natural language to have:

objectivity: assertions about the world are made

productivity: the meanings of complex expressions are systematically
related to the meaning of their constituent parts.

compositionality: the meanings of complex expressions are an explicit
function of the meanings of their constituent parts.

efficiency:
indexicality, connections, and resource situations

From the AI point of view, the key requirement for any formalism
is the ability to express facts about the consequences of events
including actions.  Examples include

1. When a block is moved from  A  to  B  it ceases being at  A 
and starts being at  B.  Formulating this requires some notion of
time or of succession of situations, so perhaps this is more basic.

2. When a person is informed of some fact, he comes to know the
fact.  When a person forgets a fact, he ceases to know it.

3. The answers to the usual reading comprehension test should follow
formally from the formal expression of a narrative.

4. When person  A  trusts person  B,  and  B  tells  A  that  p,
then  A  comes to believe  p.

	Along with most philosophers and linguists (and AI people
who blindly follow them), Barwise and Perry seem to ignore the
problem of expressing the rules that determine the consequences of
events.  At least the first chapter gives no indication that they
are going to be treated at any point in the multi-volume work.
Indeed most philosophers and linguists ignore the question of
inference entirely except in so far as their translations into
a logical formalism imply that some inferences are possible.
They usually don't investigate whether the conclusions a person
would ordinarily draw from the assertions they express follow
in their formalisms.

	The result, or perhaps the starting point,
 is a total emphasis on translation from English
into the formalism.  This leads to controversies that seem
sterile to me.  One that comes to mind from reading a recent
paper by Michael Bratman is the controversy about whether
intending an action
 can be identified with believing one will perform the action.
None of the participants to the controversy offers any formalization
of events that cause a person to intend or the events that can be
expected to follow from the fact of an intention (and other
preconditions).