Parsing Parallel Grammatical Representations

Parsing Parallel Grammatical Representations Derrick Higgins Department of Linguistics University of Chicago 1050 East 59th Street Chicago, IL 60626 dchiggin (~midway.uchicago.edu Abstract ~:aditional accounts of quantifier scope em- ploy qualitative constraints or rules to account for scoping preferences. This paper outlines a feature-based parsing algorithm for a gram- mar with multiple simultaneous levels of repre- sentation, one of which corresponds to a par- tial ordering among quantifiers according to scope. The optimal such ordering (as well as the ranking of other orderings) is determined in this grammar not by absolute constraints, but by stochastic heuristics based on the de- gree of alignment among the representational levels. A Prolog implementation is described and its accuracy is compared with that of other accounts. 1 Introduction It has long been recognized that the possibility and preference rankings of scope readings de- pend to a great degree on the position of scope- taking elements in the surface string (Chomsky, 1975; Hobbs and Shieber, 1987). Yet most tra- ditional accounts of semantic scopal phenomena in natural language have not directly tied these two factors together. Instead, they allow only certain derivations to link the surface structure of a sentence with the representational level at which scope relations are determined, place constraints upon the semantic feature-passing mechanism, or otherwise emulate a constraint which requires some degree of congruence be- tween the surface syntax of a sentence and its preferred scope reading(s). A simpler and more direct approach is sug- gested by constraint-based, multistratal theo- ries of grammar (Grimshaw, 1997; Jackendoff, 1997; Sadock, 1991; Van Valin, 1993). In these models, it is possible to posit multiple represen- tational levels for a sentence without according ontological primacy to any one of them, as in all varieties of transformational grammar. This allows constraints to be formulated which place limits on structural discrepancies between lev- els, yet need not be assimilated into an overrid- ing derivational mechanism. This paper will examine the model of one of these theories, Autolexical Grammar (Sadock, 1991; Sadock, 1996; Schiller et al., 1996), as it is implemented in a computational scope gen- erator and critic. This left-corner chart parser generates surface syntactic structures for each sentence (as the only level of syntactic represen- tation), as well as Function-Argument seman- tic structures and Quantifier/Operator-Scope structures. These latter two structures together determine the semantic interpretation of a sen- tence. It will be shown that this model is both categorical enough to handle standard gener- alizations about quantifier scope, such as bans on extraction from certain domains, and fuzzy enough to present reasonable preference rank- ings among scopings and account for lexical differences in quantifier strength (Hobbs and Shieber, 1987; Moran, 1988). 2 A Multidimensional Approach to Quantifier Scoping 2.1 The Autolexical Model The framework of Autolexical Grammar treats a language as the intersection of numerous inde- pendent CF-PSGs, or hierarchies, each of which corresponds to a specific structural or functional aspect of the language. Semantic, syntactic, morphological, discourse-functional and many other hierarchies have been introduced in the …

Content truncated to 500 words