An Integrated Computational Approach to Binding Theory

Roberto Bonato, PhD Thesis, 2006

University of Verona, Computer Science Department
University of Bordeaux I, LaBRI, INRIA Futurs, "Signes" team

Abstract

In this thesis I explore how different approaches to Binding Theory issued from the last thirty years of linguistic enquiry may be effectively integrated into a computational framework. My purpose is to enrich the current framework of computational semantics in order to inductively compute semantic representations of a sentence which incorporate the principles of Binding Theory.

The original formulation of Binding Theory presents principles A, B and C as syntactic conditions that indexed Determiner Phrases must fulfill in order for the sentence in which they occur to be well-formed. Indexes are a formal device halfway between syntax and semantics that was introduced to encode coreferential relations between DPs in a sentence. They basically act as filters that discard every structure whose indexing violates any of A, B, or C principles. However, Determiner Phrases that occur in a phrase-marker issued from generative parsing of a sentence do not come with indexes associated. Principles A, B and C provide a procedure to verify that a given indexing for a sentence is BT-compliant, but they are not constructive: no effective procedure to associate correct indexing to DPs in a sentence is provided. This is both a theoretically and a practically challenging issue. How do human beings come to associate the correct indexing (i.e. to establish the correct mutual denotational relationships) to the DPs occurring in a sentence? And how can we devise a computational procedure to mimic this process in order to obtain a semantic representation for the sentence which encodes the additional information provided by the constraints of Binding Theory?

In this thesis I tackle the problem of integrating in a computational semantics framework the mechanisms needed to encode the principles of Binding Theory into the semantic representations computed for a sentence. Different interpretations that have been given to Binding Theory ask for different implementations of such mechanisms. Eventually, I propose an integrated approach that incorporates some of the basic features of the approaches described into a framework which is both computationally effective and linguistically well-grounded. We believe this to be the first accomplished effort to integrate within a single coherent computational framework some of the basic achievements and insights in Binding Theory issued of the last 30 years of linguistics and formal semantics enquiry.

Download

Thesis (pdf, 1.05 Mb)

Contact

roberto dot bonato at gmail dot com