Lojban and LALR(1)ity

Jorge and Matthew have both brought up the question "Why should Lojban be
limited to LALR(1) parsability?"  The answer is very simple and practical.

We have a technique, namely the use of Yacc, for testing if a language meets
this specification.  Yacc and its relatives (Byacc and Bison) are available
on just about every platform.  The same cannot be said of any more powerful
parser-generator program.

History shows that Loglan did not, in fact, become unambiguously machine
parsable until the use of Yacc was introduced into the Project.  Before
that, we had various grammars which were claimed to be unambiguous but
in fact failed the test.

We already allow small extensions to LALR(1) parsability, through the
preprocessor section of the grammar.  In particular, the logical connectives
themselves (the jeks, joiks, geks, etc.) are not LALR(1).  But all of these
are essentially bounded; any unboundedness is of a simple iterative, not
recursive, kind.  (Example: the preparser allows unboundedly long numbers
to be processed, but numbers have no internal structure.)

Overusing preprocessing is dangerous: the second baseline grammar, in fact,
fails to parse "re pamai boi gerku" = "two (firstly) dogs" correctly.
This error was completely missed by Yacc, and was caught only by an actual
example.  Change 6 grew out of this: the new grammar calls for "reboi pamai
gerku", which can be parsed correctly.

-- 
John Cowan		sharing account <lojbab@access.digex.net> for now
		e'osai ko sarji la lojban.