Technical Reports (CIS)

Document Type

Technical Report

Date of this Version

December 1990


University of Pennsylvania Department of Computer and Information Science Technical Report No. MS-CIS-90-94.


This thesis presents Structure Unification Grammar and demonstrates its suitability as a framework for investigating natural language from a variety of perspectives. Structure Unification Grammar is a linguistic formalism which represents grammatical information as partial descriptions of phrase structure trees, and combines these descriptions by equating their phrase structure tree nodes. This process can be depicted by taking a set of transparencies which each contain a picture of a tree fragment, and overlaying them so they form a picture of a complete phrase structure tree. The nodes which overlap in the resulting picture are those which are equated. The flexibility with which information can be specified in the descriptions of trees and the generality of the combination operation allows a grammar writer or parser to specify exactly what is known where it is known. The specification of grammatical constraints is not restricted to any particular structural or informational domains. This property provides for a very perspicuous representation of grammatical information, and for the representations necessary for incremental parsing.

The perspicuity of SUG's representation is complemented by its high formal power. The formal power of SUG allows other linguistic formalisms to be expressed in it. By themselves these translations are not terribly interesting, but the perspicuity of SUG's representation often allows the central insights of the other investigations to be expressed perspicuously in SUG. Through this process it is possible to unify the insights from a diverse collection of investigations within a single framework, thus furthering our understanding of natural language as a whole. This thesis gives several examples of how insights from investigations into natural language can be captured in SUG. Since these investigations come from a variety of perspectives on natural language, these examples demonstrate that SUG can be used as a unifying framework for investigating natural language.



Date Posted: 05 August 2007