Re: Machine grammar and elidables

[lojbab (Bob LeChevalier)]

On David Bowen's comments on dont's attempt to eliniate elidables:
1. David is correct that "cu" is optional, in the language and exists only to
make elision of other elidables before the selbri easy.  Indeed, John Cowan
has made this explicit in one of his proposed baseline modifications.
2. Our analysis shows a similar situation for "ku'o" - it is always optional, but using it may make other elidables easy to elide.
3. John Cowan's attempt at parser buiding led him to explore "fa'o" and to
determin that it is no longer needed or useful in the grammar.  He is proposing
its elimination as a resolution of the only significantly outstanding parser
a major difficulty in constructing a YACC 
parser.  I  am in favor of this change.
4. Most other elidables are not optionals - they truly terminate something that
has no other way of being ended, in at least one case.  Before August '89
these were only two or three terminators (mostly "ku").  They were broken out 
into separate selma'o  specifically to make it easier to elide them, and to tell
what structure a given terminator was eliding.  Before then, it was trivial to
construct sentences that required 4, 5, or 6 "ku"s in a row to terminate things.
properly and to make sure things grouped the way you wanted them to.  The c
change to this was approved by all active students of the language at the time,
and was among the reasons for the Lesson4/Lesson 4A rewrite, and my soon t
thereafter decision that the entire textbook would need rewriting before it
could be published.
By making the language more speaker-and-listener friedndly by expanding the
terminators, we made it even harder to do what you are trying to do, I think.
It is extremely difficult in the current grammar to mess with the elidables
- minor change can lead to major conflicts.

The primary requirement you need is thorough familiarity with the verbose
listing output by YACC, listing all states, and the go-tos/shifts/reduces
that take place depending upon the look-ahead token.  The ultimate Lojban
formal grammar IS that set of states produced by YACC, since we regard the YACC
grammar (and NOT the E-BNF) as the legitimate standard of the Lojban grammar.

We have some pragmatic rules for many of the elidables; i.e. times that you
cannot elide "ku" at the end of a sumti include any time it is followed by
any token that can start off a selbri (the list is large so expanding this
one elidable would multiply the grammar by an order of magnitude in my belief
- and it maight then no longer be LALR 1).  "kei" is even tougher, since what
can follow depends on what came before.  In "le nu broda [kei]" the "kei"
cannot be elided if the next tokenm is the possible start of either a selbri
OR a sumti; in "le nu broda ko'a" elision is forbidden if the next token
starts a term (i.e. a sumti or tcita sumti).  But "ku" can be used instead of
"kei" in this instance and it will always work and always allow "kei" to
be elided, since it terminates the "le" sumti that the "nu broda ...[kei]"
abstraction is inside.

Another problem for would-be grammar experimenters.  The Lojban grammar is
not much larger than most computer languages, but its degree of recursive
and overlapping constructions leads to a lot of states in the state table.
We needed a specially designed version of YACC, produiced by Abraxas (PCYACC),
with elevated limits for some table size limits and compiled under "Huge"
model, in order to even YACC the last few grammars.  John Cowan reports that
he is having less trouble on UNIX systems, but even there you have to watch
your table limits, I'm sure, especially if you want to try something like don
is.

In short, I do not believe the effort will bear much fruit, and I would only
recommend it as an excuse to learn the grammar - you WILL have to do that in
order to make any progress at all.  And remember that we are going to be
very prejudiced against changes that in some way restrict usages that we have
already taught or used in the language.  Almost all proposed changes now either
have no effect on most day-to-day usage, or expand the capabilities thereof.
This makes an elidable eliminating effort even more difficult than it might
otherwise be.

-lojbab