COMMENT ⊗   VALID 00009 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	ideas[w84,jmc]		what has natural language got
C00012 00003	1984 feb 13 - Text valued expressions in editors:
C00013 00004	1984 feb 14 - Abstract actions
C00019 00005	Feb 16 - Ontology of the blocks world
C00021 00006	Feb 19	Science progress slowed by ideologic orthodoxies
C00023 00007	Feb 28 - Unique names and circumscription
C00027 00008	Feb 29 - Typology of introspection and reflection.
C00029 00009	Mar 3 - Simmon's phenomenalism
C00031 ENDMK
C⊗;
ideas[w84,jmc]		what has natural language got

1984 Feb 1

	This is in further pursuit of the greater power of natural language
and human thought over logical languages.  The advantage appears to
be in ambiguity tolerance.  Our thoughts and natural language can use
ideas involving concepts that turn out to be ill-defined when closely
examined.  It appears that the ambiguity is not merely in the words
we use, e.g. the presence of homonyms, but in the concepts themselves.
We can successfully use certain concepts to solve a problem and only
later discover that we cannot define the concept precisely.  Much of
philosophy is concerned with the fact that the concepts are ambiguous
but not much attention has been devoted to the fact that we can use
them successfully anyway.  Indeed we can fail to resolve a pointed out
ambiguity, put the problem aside, and continue to use the concept in
the case where it isn't ambiguous.

	The lack of this ability in logical languages and in their
variants used in AI shows up in the kind of inflexibility often called
brittleness.  Typically the language is developed with one or at most
a few problems in mind.  When a problem of a kind not previously
envisaged arises, almost invariably the axioms assumed turn out to
be false.  Moreover, the new problem often turns out not to be
expressable in the language that has been provided even though
its English expression may involve the words that have supposedly
been formalized.

	Moreover, no-one has developed a technique for expressing
the facts of even simple English narrative in logic, let alone
those expressed in texts like this one.

	In spite of these grumbles about logic, we plan to try to
develop a language within a logical language for expressing what
thoughts and natural language.  We distinguish between thought and
natural language, because we regard it as unlikely that all thought
is expressable in present natural language.  At least there is
a large body of common sense information that is not ordinarily
expressed in natural language.  For example, when a person is asked
to express the common sense facts about quantities of liquid and
there transfer among containers and through pipes and other channels,
he has great difficulty in doing so even though he can correctly
predict the outcome of various actions.  It seems that common sense
generalities are particularly hard to express.

	How can we proceed?  I would like to try for a methodology
somewhat different from present day linguistic and philosophical
semantics, because this methodology is already being vigorously
pursued, and I'm unlikely to add much.  Here are some possibilities:

	1. Translate simple narratives, e.g. Mr. Hug
MRHUG[S76,JMC] into logic and find out where the troubles are.

	2. Find out the problems of formalizing bits of acceptable
common sense reasoning.  It would be interesting to consider examples
of reasoning about physical situations that can be communicated over
the telephone.  Thus we avoid having to look at the physical situation
itself.  Perhaps this is too hard.  After all a sculptor or an artist
could recreate a plausible image of the scene from a description.
The rest of us can't do that, but presumably have similar abilities
in embryonic form.

**But we've digressed from ambiguity.


What is the simplest example of ambiguity tolerance?  Certainly
there is something simpler that the story about attempting to
bribe a public official.  What about mother?  As mentioned in
some previous file, the word and presumably the concept is taken
by a child variously as a proper name, a two place predicate and
a one palce predicate.  In logic we would be tempted to use separate
predicates, but of course we an also introduce an abstract entity
called "mother" and let the others be aspects of it.  Whether this
is enough is debatable.

is-a(Mary, mother)
is-r(Mary,Tom, mother)
the(mother,context)

1984 Feb 1 - Child machine

Many people have proposed to build a child-machine.  It knows little
or nothing about the world but is capable of learning everything
humans know from its experience.  Presumably it can understand
language or can learn to do so.  This is a plausible idea, but attempts
to realize it have not been successful, usually not even resulting
in published papers.  In spite of the fact that the approach has been
so far unsuccessful, trying again is worthwhile provided one recognizes
the difficulties that have been encountered and has ideas for overcoming
them. Here is my idea of what the difficulties are.

1. Representation.  Consider the very first thing the child-machine
is to learn.  Whatever is learned, whether it be a fact or the appropriate
response to a class of stimuli, it must be represented somehow in the
memory of the machine.  Therefore, the designer must provide for this
representation.  Here he faces a dilemma.  If he wants some sophisticated
fact to be representable, it seems that he must build in quite a few
concepts.  But this is the problem the proposers of the child-machine
want to avoid.  The extreme avoiders of this difficulty have a tendency
to build stimulus-response machines.  They have two problems.  First,
they are usually forced to admit only fixed stimuli and this makes the
machine quite far from a real world child who never sees exactly the
same stimulus twice and must begin with a complex stimulus classifier.
Second, it seems that little that children learn takes a simple
stimulus-response form.  Children mainly learn facts which they subsequently
use more-or-less intelligently, i.e. with the aid of inference involving
other facts.

Well, maybe I only know that one difficulty.

1984 feb 13 - Text valued expressions in editors:

	Besides being able to search in various ways to move the pointer,
it would be valuable in editors to be able to designate text-valued
expressions in various ways.  This is also worthwhile when driving
programs from the editor.  What set this off was seeing
"not_successor_zero" referred to in an EKL proof.  There it was
the proper name of a certain line of the proof.  However, it would
have been more convenient to have been able to refer to the line
by its partial content.

1984 feb 14 - Abstract actions

Abstract performatives or maybe we should just call them abstract actions.

	To a speech act theorist, a performative is a sentence that
does something by its utterance other than state a fact or command.
Perhaps they have a better definition.  An example is a promise, whose
utterance is not merely a statement of intention but which creates
an obligation to carry out the promise.

	It seems useful in the Elephant style of programming to consider
abstract performatives, e.g. reserving a seat on Flight 177 for
John Searle.  We imagine this as an action the reservation system can
do, but we don't associate with any particular modification of any
database.  Instead it is related to subsequent actions by the
statements,  "If a seat has been reserved for someone, and the
reservation hasn't been cancelled, and he arrives at the gate
at the time of the flight, then he should be allowed on the airplane".
Of course, cancelling a reservation is another abstract performative,
and, perhaps, so is allowing someone on a plane.  The idea is that
the reservation program can be written in terms of abstract performatives,
and the program can refer to past performative events.  It is up
to the compiler to make these abstract actions concrete, e.g. to
store an item corresponding to the reservation in a database.

	All this is related to a comment I made on Moshe Vardi's
New York paper at the conference on logic and computer science.
Vardi discussed how one should update a database in terms of how
a new fact causes changes in the database.  I suggested to him
that one should think of it in terms of events rather than just
in terms of facts, e.g. firing a person and deciding that he had
never worked for IBM at all are distinct events even though they
may result in the same change in a particular database.  If the
structure of the database is modified, they may come to result
in different changes.

	When I told Carolyn about this, she commented something
to the effect that if one thinks about it entirely in terms of
events and their histories, then one doesn't really need the
database as a primary object, and I promptly related this to
Elephant.

May 24

	Let's call them abstract actions rather than abstract
performatives.

	Another example of an abstract action is for the program
to commit itself to some future action or even to achieve a future
goal.  This commitment need have no effect whatsoever.  However,
the correctness of a program depends on commitments being carried
out.  This came out in a discussion with Brian Smith who substantially
agrees.

May 26

	"A program is correct if it carries out its commitments" is
worth exploring as a definition, because we can consider the simple
case of having initial commitments.  However, the idea may be worth
elaborating, since it may turn out to be impossible for some reason
to give every passenger the seat reserved for him.  Therefore, instead
of considering only  correctness,  it may sometimes be worthwhile to
use the concept of  success.  A program is  successful  to the extent
that it succeeds in carrying out its commitments.

Feb 16 - Ontology of the blocks world

	Our ontology of the blocks world is still too limited to
do planning properly, i.e. it isn't heuristically adequate.  Here
are some examples of additional useful concepts.

	1. What goes under block  A in the final tower.
	   What goes above ... .
In our previous work we have imagined these as lists of blocks in order.
We recognized that there might be more complicated structures but
wanted to postpone having to define them.  The key is to use the
concept before defining it precisely and maybe never doing so.  Like
a situation it has an indefinite number of properties.  Perhaps we
need a name and a discussion of entities with indefinite number of
properties.  Remember to regard this as a case of finitization.

	2. The preconditions for moving A.  We think of this as an
entity to be constructed, e.g. we may think of new preconditions.
Likewise, but vaguer: what moving A will make possible.

	We need some formalized examples of such concepts.

	They provide good examples of vague concepts requiring
ambiguity tolerance.  We need to figure out how a person or machine
can use such a concept for what it is worth and not get bogged down
in haggling over definitions.

Feb 19	Science progress slowed by ideologic orthodoxies

	To what extent is the progress of science today slowed by ideological
orthodoxies?  That it was so slowed in the past seems unquestionable,
and this question occurred to me while reading Daniel Boorstin's
"The Discoverers".  It may be, of course, that historians of science
tend to exaggerate the ideological, since that is more readily
expounded than the scientific ideas themselves.  It does seem to
me that progress in computing, computer science and AI have been
somewhat slowed by ideology, but it isn't so easy to say what
the offending ideology is.  If one could identify it, then some
propaganda against it might help, and my presidential essays
should be devoted to that end.

Feb 28 - Unique names and circumscription
	Reiter Mailed that he has a student who claims to have
proved that unique names cannot be done by circumscription.
Since my attempts all involve trying to cheat in some way, he
may be right.  It occurs to me that we can always get unique
names by circumscription provided we circumscribe  eqdenot(name1,name2)
before taking any other facts into account and then hold to the
result of this circumscription when other facts come along.
Thus our interpretations may not be minimal models of all the
facts.  This has the undesirable feature that the results of
reasoning depend on the order in which facts are taken into
account.  However, we may have to suffer it.

	(We need a good
example where unique names plays a role, i.e. a conclusion is
reachable with unique names that is not reachable without it.
Possibly we can get one by putting  y ≠ top(x) ∧ y ≠ left(x)
as a precondition for  move(x,top(y)).  Remark: Actually
the condition should be that the  p  in  move(x,p)  not involve
x.  There is both a precondition and a result.  move(x,top x)
is impossible.  However,  move(x, left x)  possible but doesn't
result in  at(x,left x).  If we regard  top(x)  and  left(x)
as abbreviations for  top(x,s)  and  left(x,s)  respectively
we get out of this trouble.)

	Returning to the main point, all this suggests using
circumscription or some other form of non-monotonic reasoning to
explain words like "but" and "however".  Normally, the listener
circumscribes after every sentence including the conclusions
of the previous circumscription in his axiom.  (What does he
circumscribe?  Aha, he circumscribes  ab(z), but presumably
keeping the results about  ab  previously proved.  We need
to check whether this works properly).  The words "but" and
"however", however, constitute signals that the previous
circumscription (if any) has to be undone, and the new fact
included with the previous facts in a new circumscription.

Feb 29 - Typology of introspection and reflection.
We need such a typology; the different phenomena of introspection aren't
all the same.  It is a common error to try to build a complete theory
of some general phenomenon on a few over-specialized examples.

1. Brian Smith's example of a person whose raft has capsized thinking
about what he did wrong and how to do better next time.

2. The introspection involved in re-arranging a Prolog program using
the postponement heuristic.

3. The Kempe heuristic.

4. Simply recognizing oneself as a certain material object analogous
to other material objects.

5. Answering questions like "Do I know whether Reagan is standing
or sitting?".

6. Do I want dinner now?

7. A program counting its own jump instructions.

8. "A reflective computational calculus, similarly, must be able to obtain
fully instantiated explicit descriptions of its own state in a language
and setting appropriate to their examination and manipulation". - Brian Smith
in Reflection and Semantics in LISP, 1983 Aug.

Mar 3 - Simmon's phenomenalism

	Simmons's Computations from the English, Prentice-Hall 1984,
advances a "phenomenalist" view of the world that "disposes of the
distinction between objects and awareness of objects".  I have always
found such views philosophically objectionable, but now it seems to
me that they lead to scientific error as well.  The fact that such
views are widely prevalent, perhaps dominant, in the cognitive science
community may be responsible for the neglect of common sense knowledge
of the world.  Cognitive scientists try to discuss natural language
understanding solely in linguistic terms, not bothering to try to
formulate the common sense facts about the world knowledge of which
lies at the basis of human ability to understand what was said.
An analogous situation existed in logic until Goedel and Tarski
made a sharp break with the "positivist" view that truth cannot
be distinguished from provability, i.e. from the perception of
truth.  Kreisel informs me that Frege suffered from an unwillingness
to make this distinction and therefore could not formulate
any notion of completeness of his formal system.  Question: was
Goedel the first to formulate this notion?