Bob Carpenter (carp@research.bell-labs.com)
Wed, 29 Jan 1997 16:03:18 -0500 (EST)
I wholeheartedly agree with Gosse's message and want to clarify the point about type-logical categorial grammars. In TLG, you want to define the copula disjunctively as: VP/(NP or PP or VP or ADJP) This allows coordination of unlike categories because with the classical logical of disjunction and the usual type-raising logic for slashes (ala Lambek) you derive: NP --> (VP/(NP or ... or ADJP)) \ VP and you can derive the same category for any other element on the disjunctive list of complements for the copula. By classical logic again (noting that slash in type-logical grammar is just directed implication), you derive: VP/(NP or ADJP) <--> VP/NP and VP/ADJP But it is crucial to point out here, as Bayer and Johnson did in their recent Lanuage paper, that conjunction is *not* the same as lexical ambiguity. If you have two lexical entries: word --> Cat1 word --> Cat2 there is no way to derive word --> (Cat1 and Cat2) at least given the way the lexicon is typically linked up to the rest of the grammar. This has the intended effect (again as Bayer and Johnson point out) that you don't get know --> VP/S know --> VP/NP leading to know --> VP/(S or NP) <--> VP/S and VP/NP which would have the unwanted effect of admitting 'knows Mary and that it will rain today' as being grammatical. [Note: for those wondering how disjunction becomes conjunction and vice-versa here, it is the same as in classical logic: (A or B) -> C is logically equivalent to (A -> C) and (B -> C) HPSG is different in that the subcat list is not implicational, or in other words, is not a logically negative environment, so that C[SUBCAT:(A or B)] is equivalent to C[SUBCAT:A] or C[SUBCAT:B]. EndNote] - Bob ------------------------------------------------------------------------------ Bob Carpenter Email: carp@research.bell-labs.com Lucent Technologies Bell Labs WWW: http://macduff.andrew.cmu.edu/carpenter 600 Mountain Avenue, 2D-329 Voice: 908 582-5790 Murray Hill, NJ 07974 Fax: 908 582-3306 ------------------------------------------------------------------------------
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:36:03 PST