Specific meanings of words

[cbmvax!uunet!MATH.UCLA.EDU!cuvmb.cc.columbia.edu!jimc]

There has been recently a conversation between And Rosta and Lojbab in
which And (and others?) asks for the gismu definitions to be made more
specific and Lojbab says that such specificity is impossible, because
first the meanings of the words are dynamically negotiated between the
users; second the meanings depend greatly on context, as in the hammer
example; and third because even without these problems gismu X would
have to be defined in terms of gismu Y which, inevitably, would be
defined circularly in terms of X.

As is well known, I prefer to think of words as well defined.  However,
I do not exactly disagree with Lojbab's positions.  Here is an approach
that I have found somewhat useful.

It is probably hopeless to prepare text definitions of the gismu that
avoid circular usages.  However, think of the referent set of a
predicate: a list of sets of thus-related objects.  For example, the
referent set of the predicate "eat" includes:

        betsy (an elephant) eats peanut #325
        betsy eats straw #116432
        willie (a monkey) eats peanut #326
        willie eats stolen sno-cone #58
        and so on for a large but finite number of records in the
                database.  (Some predicates have infinitely many
                records.)

The doctrine is that the definition of "eat" is no more and no less than
this referent set (set of records).

Any one language user knows a subset of the whole definition.  The
purpose of language behavior is to transfer records from the speaker to
the listener.  For example, the kid might not know it, but "that monkey
is reaching for your sno-cone".  Now the record is in the kid's copy of
the database, possibly in time to prevent the theft.

Humans are very good at recognizing observed relations as being
instances of a particular word.  If you see a snail on a leaf it might
not be obvious what is happening, but when you see the holes left
behind you recognize that the relation you observed belongs in the
"eat" referent set.  Also, by exchanging phrases (database records),
the speakers keep up to date on how each other is grouping events
under words.  This is the essence of the dynamic renegotiation of
meaning that Lojbab discusses.

How many places (cases) does each referent set record contain?  I see
the records as containing many modal places such as time of occurrence,
authority, speaker, and so on, which introduce a fairly complete
context.  It is via context introduced this way that something might
normally be just a rock, but might be in the hammer referent set in
certain events.

By considering referent sets as definitions, we can accomplish several
goals:
    1.  Each user has a specific definition for the word.
    2.  Dynamic negotiation and individual differences are recognized.
    3.  Context is recognized explicitly.
    4.  No circular definitions; no incomprehensible text definitions.

Needless to say, the text definitions are still needed as training aids,
so they should be as clear as feasible.

                -- jimc