The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

PhD Thesis #9031

Title:An Adaptable Formalism for the Computational Analysis of English Noun Phrase Reference
Date: 1990
Abstract:This thesis addresses issues of analysis and representation of the meaning of human language on computers. The notion of adaptable representations, which are designed to encode ambiguity implicitly, is introduced, and placed in a context of related work on representation and reference analysis. A parsing formalism is presented which allows the analysis and translation of a subset of English into a representation language which encodes not only the truth-functional semantics of the input discourse but also some of its surface form. The parsing algorithm works strictly word-by-word, always maintaining a self-contained partial representation of the entire discourse so far, in order to provide a stringent framework for examination of the idea of adaptability. It is shown that the use of adaptable representations which implicitly encode local lexical ambiguity can improve the efficiency of parsing by reducing non-determinism. A basic noun phrase reference analysis algorithm is presented, which allows inferences to be made about the reference of the input discourse. Following the common belief that noun phrase reference analysis in humans takes place as information is received, processing is performed word-by-word, in parallel with parsing. Part of the reference analysis algorithm constitutes a new view of the distinction between restrictive and non-restrictive noun phrase modifiers. it is shown that this new view and the adaptability of the system in general does not compromise the ability of the reference analysis algorithm to produce correct results. Finally, a new approach to the generation of readings of sentences containing many-ways ambiguous or vague set reference is presented. The approach relies heavily on adaptability, and it is suggested that without (an equivalent of) adaptability any approach is likely to be combinatorially explosive and therefore in general computationally intractable. This presented as one specif

[Search These Pages] [DAI Home Page] [Comment]