From cbmvax!uunet!mcnc.org!aurs01!aurw31!waugh Tue Mar  5 13:17:40 1991
Return-Path: <cbmvax!uunet!mcnc.org!aurs01!aurw31!waugh>
	for lojban-list@snark.thyrsus.com
Date: Tue, 5 Mar 91 10:47:08 -0500
From: cbmvax!uunet!mcnc.org!aurs01!aurw31!waugh (Jack Waugh)
Message-Id: <9103051547.AA01181@aurw31.local>
To: aurs01!lojban-list@snark.thyrsus.com@mcnc.org
Subject: Grammar Patterns to Aid Parsing
Status: RO

What I meant by "hard" little words was ones that the listner
would have to know in order to parse a sentence.  Let me refine
that proto-proposal a little (forget about "hard"):

An adequate model to use for the first stages of listening
would be to say that the listner parses first and then
starts semantic processing.  Parsing is casting the
utterance into a tree, so as to know the grouping relationships.
Once this is done, the listener turns to the semantics
of the words.  For example, in Lisp, you might have an
expression:

	(CONS A B)

The parentheses say that CONS is to operate on A and B; this
can be determined without any semantic information on CONS.
Once the parse is done, the interpreter can start to ask
"what does B mean", what does A mean, and what does CONS
do with these two operands.

Of course, the parsing rules for Lisp would not be appropriate
for a spoken language for use between humans.  However,
maybe we could design a set of conventions which would
require the parsing operation to act on less information
than the whole cmavo vocabulary, just as a Lisp parser
doesn't have to know the behavior of the functions.

An approach would involve words in phonological series.
The predicate words and names already conform to this.
It seems to me quite a strength and argument for Loglan that
the parser can easily recognize a predicate word, and the
thinking about what it means can be delayed so as not to
stand in the way of rapid parsing.  Perhaps we could buy
the same facility for the rest of the language.  Some
of the meanings that today reside in cmavo (little words)
could be put in words in such series.  There would be one
series per parsing-category.

When a category is not frequently used and has a lot of
semantic choices, the way to make the series would be
to always start with a marker syllable (or string of
syllables) (I would write the marker as a separate
"word", although it would have no semantics without
what follows it) and follow the marker with some more
syllables to specify the semantic choice.  The semantic part
would of course have to follow some rules that would let
you know where it ended.