Mark Johnson (mj1@lx.cog.brown.edu)
Mon, 25 Nov 1996 13:23:32 -48465656 (EST)
Regarding Robert Borsley's posting, I had the good fortune to work with Sam Bayer on this while he was at Brown. That work convinced me that it is worthwhile thinking some more about the _linguistically_ desirable properties of features and how directly a given _formal_ treatment captures them. Sam's analyses seems more directly compatible with a resource-sensitive conception of features inspired by Categorial Grammar, rather than the unification-based conception underlying HPSG and LFG. We wrote a couple of papers on this, M. Johnson and S. Bayer (1995) <A HREF="ftp://lx.cog.brown.edu/pub/mj/barcelona.ps.Z"> Features and Agreement in Lambek Categorial Grammar</A> Proceedings of the 1995 Formal Grammar Workshop. S. Bayer and M. Johnson (1995) <A HREF="http://xxx.lanl.gov/abs/cmp-lg/9506007"> Features and Agreement</A> Proceedings of the 33rd Annual Meeting of the Association for Computational Linguistics. Jochen Doerre and Suresh Manandhar have analysed the properties of a class of feature systems that includes the feature system Sam Bayer and I proposed. @incollection{DoerreManandhar95, author = {Jochen D{\"o}rre and Suresh Manandhar}, title = {On Constraint-based {L}ambek Calculi}, editor = {J. D{\"o}rre}, booktitle = {Computational Aspects of Constraint-based Linguistic Description III, DYANA-2 deliverable R1.2.C}, publisher = {ESPRIT, Basic Research Project 6852}, month = {September}, year = 1995, note = {To appear in Blackburn, de Rijke (eds.), Specifying Syntactic Structures, papers from the Logic, Structures, and Syntax workshop, Amsterdam, Sept. 1994} } While I was on sabbatical at Rank Xerox, Grenoble, I looked into the possibility of restructuring a unification grammar so that its basic mechanisms would use a resource-conscious treatment of features, rather than the traditional cosatisfiability approach. Although my work was fairly preliminary, I think it has some promise (e.g., it provides a simple model of some of the non-monotonic devices of LFG). I worked on LFG (well, they were paying) but I that one could do something similar for HPSG as well (indeed, I think that how the type hierarchy interacts with resource sensitivity would be very interesting to investigate). M. Johnson (1996) <A HREF="ftp://lx.cog.brown.edu/pub/mj/rslfg.ps.Z"> Resource-sensitivity in Lexical-Functional Grammar</a> to appear in the Proceedings of the 1996 Roma Workshop. By the way, I am not claiming that a resource-sensitive formulation of LFG is better than the standard unification-based approach. Rather, I think that both are capable of expressing the basic linguistic insights of the theory, but that they cut up the phenomena in varying ways. Until recently I thought that there was essentially only one way of formalizing feature dependencies in a grammar, but now there seems to be a wide range of possibilities to explore. Comments? (I've probably said enough here to be flame-bait for the next few weeks ... ) Mark -- Mark Johnson, Cognitive & Linguistic Sciences Box 1978 Brown University, Providence, RI 02912 telephone (401) 863 1670, telefax (401) 863 2616. preferred email: Mark_Johnson@Brown.edu > Has anyone thought about the implications for HPSG-type theories of the > observations and ideas about coordination in Bayer's recent Language > paper on coordination? It seems to me that he presents quite strong > evidence for including some sort of disjunction and conjunction within > categories, but also that there are objections to some features of his > approach. As I understand it, he assumes that the work that is done in > HPSG-type grammars by underspecification should be done entirely by > disjunction (join). This would seem to entail that object position would > have to be identified as a position allowing an [NPdefsing or NPindefsing > or NPdefplur or NPindefplur] (at least). This seems pretty undesirable. > Unlike an underspecification appproach, this approach also does not lead one > to expect the same range of options in different positions, e.g. object > of a verb and object of a preposition. This must be a disadvantage. A > further point is that his join introduction, which allows an X to count > as an [X or Y], must surely lead to computational problems. It would be > better I would have thought, if disjunction was a consequence of conjoining > unlike categories and not a prerequisite for such coordination. I assume too > that his antecedent strengthening, which allows an element requiring an X > to count as an element requiring an [X and Y] must also lead to > computational problems. Again, it would seem better if conjunction were a > consequence of coordination of categories with different requirements and > not a prerequisite. > > Does anyone have any thoughts on any of this? > >
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:30 PST