From cbmvax!uunet!ctr.columbia.edu!shoulson Thu Jul 11 13:29:21 1991 Return-Path: From: "Mark E. Shoulson" Date: Thu Jul 11 13:29:21 1991 Message-Id: <9107111550.AA19082@relay2.UU.NET> To: lojban-list@snark.thyrsus.com In-Reply-To: Chris Dollin's message of Thu, 11 Jul 91 12:11:21 bst <9107111111.AA08794@tigger.hpl.hp.com> Subject: the grammar Status: RO I was thinking about the grammar last night, in particular in reference to the new changes. I was giving some thought to the change which makes "cu" optional rather than elidable. At first, I thought it made eminent sense. It would cause a shift-reduce conflict, but that's okay, since at least YACC handles those by shifting, which is what we want. But then I realized that this can lead to trouble. If "cu" is optional, then that means that the sentence "le mlatu klama" is grammatical. The usual reading of it would be as a sumti, "the cat-like comer." Since a bare sumti isn't a grammatical utterance (is it?) the only possible reading is "le mlatu klama," "the cat comes." OK, now we have a break from the accepted reading caused by the parser, but that's not a real biggie. Yet. The trouble comes with something like this: le mlatu klama le ctuca cukta the cat come the teacher book. With "cu" mandatory, this utterance is ungrammatical, as it is just a bare pair of sumti. With "cu" elidable, things are murky, since we haven't formalized elidability, but we "know" that it's not elidable in this case, so we know what it should be. With "cu" optional, this sentence is clearly grammatical. Trouble is, it's *ambiguous* (there, I said it. Lojban's bugbear word). It's either: le mlatu klama le ctuca cukta the cat comes to the teach-book Or: le mlatu klama le ctuca cukta the cat-comer [to the] teacher comes (SOV order.) You can mandate when elision is permissible (and presumably will at some point), but once you make the change in the grammar, without some snazzy rules the parser/generator is free to do as it pleases. YACC will pass that "sentence" as a valid utterance. That means it's not a pair of sumti. So it's a bridi/jufra. There are two choices. Which will you take? This is *syntactic* ambiguity. I guess there'll be some disambiguation rule like right-branching or something, or preference to SVO order, but that seems contrary to what the parser is about (though I know similar rules are used to disambiguate other Lojban constructions, e.g. tanru). Does this make sense? Do these little, usually incorrect comments of mine get on people's nerves and make me look silly overmuch? Sorry if they do. ~mark