Mike Calcagno (calcagno@sfs.nphil.uni-tuebingen.de)
Thu, 25 Jul 1996 17:35:19 +0200 (MET DST)
my two cents: > Coming back to HPSG, the situation is not that clear. I know what > the general architecture is: a grammar is specified using totally > well-typed feature structures over a signature that specifies the type > hierarchy and the appropriateness specification. But what do rules > look like? I think this is slightly inaccurate. The *objects* which are being modelled are something like totally well-typed feature structures (or interpretations of feature and species names, or whatever your favorite structure is), but a grammar is specified as a set of *constraints* on those objects. A legal rule, then, is any well-formed formula is the formal language you define for expressing those constraints. > Let's assume that a grammar consists of a set of > principles, a set of rules, a set of ID schemata and a lexicon. Two > questions pop up: (1) What is the formal structure of each > constituent? They would most likely be, as you suggest, implicative constraints within a certain feature logic... id schemata, for example, have certain objects in their denotation, so you could imagine having something like an ID-principle which says: phrase --> id1 V ... V idn similarly for the lexicon, or any universal principles... i don't know what you mean by "rules"... lexical rules? (2) What other requirements have to hold for a well-formed grammar in order for it to be an HPSG grammar? i don't think i can answer this question, but see below... > The second question is even more puzzling. Suppose we agree upon the > formal structure of the rules (including principles, schemata and > lexicon). Is every grammar thus specified an HPSG grammar? i would think not... in King's SRL, for example, you can write grammars about lists, or ancestral relations, or a lot of things that would seem to fall outside the scope of HPSG grammars... maybe it has to be "about" objects that are taken to model linguistic knowledge? > To sum up, let me re-phrase the question I asked in the first > paragraph: Given some grammar that is supposed to be an HPSG grammar, > how can one verify this claim? a different question: why is such a claim important, unless in order to be invited to some party or other you had to have an "hpsg" account of some phenomena, as opposed to an lfg account *snicker* or something else? i would think we'd be better off asking ourselves whether or not the predictions of such a grammar could be precisely determined, and whether or not those predictions reflect the empirical phenomena the grammar was supposed to account for... obviously, it should not be incompatible with other enlightening accounts of different phenomena, so it should be connected to current work in *some* way... but if, for example, you changed the head feature principle, i don't think a valid criticism would be "that's not hpsg because you don't assume the head feature principle!"... more constructive would be "your proposal makes predicitions incompatible with these other (equally precise) proposals"... or "can we now modify existing proposals to reflect your change to the head feature principle without making incorrect predictions?" mike calcagno
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:28 PST