Problem

In our book, we carry out a systematic study on reference processing from the angle of binding constraints over anaphoric links.

Binding constraints on anaphora are an important research issue in the field of semantic processing. Empirically, these constraints stem from quite robust generalizations and exhibit a universal character, given their parameterized validity across natural languages. Conceptually, the relations between them involve a certain kind of symmetry, lending them a modular nature as a consistent set of conditions on reference.

Inasmuch as they appear as non-trivial restrictions on the antecedents an anaphor can take — syntacticians usually refer to them as binding principles —, they are of critical importance in enhancing accuracy and reducing the search space in the process of anaphor resolution.

In contrast to this, however, the formal and computational handling of binding constraints has persistently presented non-negligible resistance when it comes to their integration into the course of reference processing.

From the point of view of anaphor resolution, binding constraints have been casually accepted either on the side of preferences or on the side of filters — which undoubtedly seems to be their natural place —, but in any case in a haphazard and partial fashion.

The way in which they have been typically integrated into grammar, in turn, raises also many questions. In its mainstream formulation, the basics of which date back to the early eighties, the complete process of verification of binding constraints involves an exponential algorithm inclusively requiring a series of extra grammatical parsing steps. More recently, prominent constraint-based frameworks for grammatical knowledge representation and processing either require special purpose extensions of the description formalism for handling these constraints (e.g. Lexical Functional Grammar) or offer no integration for them into the grammatical setup (e.g. Head-driven Phrase Structure Grammar / HPSG).
 
 

Goal

The primary goal underlying the research reported in our book is thus to contribute to dissolving the gap between, on the one hand, the conceptual elegance, empirical adequacy and practical relevance of binding constraints, and on the other hand, their thorough integration into grammar and the course of reference processing.

Binding constraints appear as grammatical restrictions on possible anaphoric links between anaphors and their antecedents. Accordingly, the essential thread of our research is directed towards clarifying whether and how these constraints can be fully integrated into reference processing, by fully integrating them into natural language grammar and by discussing how this can be practically achieved in a prominent constraint-based framework such as HPSG.
 
 

Overview

In chapter 1 Introduction: Anaphora and Reference Processing and in chapter 2 Heuristics and Constraints, we provide an overview of the state of the art in reference processing and anaphor resolution. We discuss how binding constraints are to be properly integrated into the whole setup of factors impinging on reference processing.

In chapter 3 Empirical Universality, we provide an introductory description of what is at present a quite stable set of facts and central results that any account of binding constraints must be able to handle. After that, we discuss recent challenges on the universality of binding constraints and try to show that these are but the effects of the conjoined occurrence of other collateral linguistic constraints together with binding constraints proper.

In chapter 4 Symmetries and Duality, we study long-distance reflexives and argue for a formal account of a fourth binding constraint, associated with this class of nominals. We discuss how this new constraint consistently fits into the existing binding theory, based on three constraints, and explore the far-reaching consequences of this integration both for a radically new conception of binding as obliqueness quantification and for new insights into the formal semantics of nominal expressions.

In chapter 5 Formal Semantics, we elaborate on how to accommodate obliqueness quantification in grammar. We also propose how the grammatical representation of binding constraints may be coordinated with the processing of reference, in general, and the resolution of anaphors, in particular.

In chapter 6 Constraint-based Specification, building on clarification from previous chapters, we discuss the integration of binding constraints into the constraint-based framework for grammatical knowledge representation and processing provided by HPSG.

In chapter 7 Computational Implementation, after presenting a brief survey of the computational systems available for the implementation of HPSG particular grammars, we discuss the major aspects of implementing the supporting grammar fragment we used. Subsequently, we report on the integration of the previously specified binding constraints into that core grammar.

Finally, in chapter 8 Conclusions: Summary and Outlook, we make a detailed summary of the results obtained in the book. We also briefly elaborate on how the research reported in this book can be extended to other themes that seem to offer themselves as relevant areas for continuation. Our study here will be exploratory and will point out the seemingly important aspects of integrating binding constraints into successively broader domains, where larger sets of data, processing factors or practical issues are taken into account.
 
 

Chief contributions

The research reported in our book aims to bridge the persistent gap between the conceptual elegance, empirical adequacy and practical relevance of binding constraints, on the one hand, and their integration into grammar and the course of semantic processing, on the other. While meeting this goal is the primary contribution of our book, in the process of meeting it several other results were also achieved. A complete list of such instrumental results is provided below. We would like to emphasize here the following two:

An innovative understanding of binding constraints, as an alternative to the mainstream view envisaging them as well-formedness conditions on syntactic representations: Under this new conception, binding conditions are the effect of phase quantifiers over reference markers arranged in obliqueness hierarchies.

A new algorithm for the verification of binding constraints, as an alternative to the mainstream one, which is exponential and, moreover, has the drawback of requiring extra grammatical parsing steps: The new algorithm is polynomial, is integrated into the grammatical parsing process, has a full-fledged constraint—based specification, and is practically tested by means of an implementation in a computational grammar.
 
 

Results

On a par with these two chief contributions, we obtained the following results:
 

While circumscribing the convenient theoretical and practical setup for the study of binding constraints, we carried out:

        • an extensive overview of reference processing and anaphor resolution;

        • an overview and critique of the adequacy of centering theory for the purpose of anaphor resolution and anaphora understanding.
 

When addressing problems and possible counterexamples to the generalizations underlying binding constraints recently raised, we ensured the preservation of the hypothesis about their universality by uncovering new parameterization factors, namely:

        • linear vs. non-linear obliqueness hierarchy option for predicators;

        • possibility for binding constraints to apply on the basis of (i) derived, (ii) non-derived, or (iii) derived and non-derived obliqueness hierarchies, depending on specific languages.
 

Concentrating on long-distance reflexives, which have appeared in the literature as a particularly problematic class of anaphors, we argued that:

        • Portuguese has long-distance reflexives;

        • the widespread, mainstream prediction concerning the correlation between long-distance reflexives, on the one hand, and subject-orientedness and morphological simplicity, on the other, has no empirical ground;

        • there is a fourth binding constraint for long-distance reflexives.
 

Building on the existence of a fourth binding constraint, we have shown that:

        • binding constraints can be organized as corners of a logical square of opposition and new symmetries among them can be uncovered;

        • binding constraints can be analyzed as the effect of phase quantifiers over obliqueness orders.
 

When addressing the proper place of binding constraints inside the architecture of grammar, we proposed:

        • a thorough integration of phase quantification over obliqueness in grammar, either in the semantic representation or in the syntax-semantics interface component (using DRT);

        • neat interface points between grammars with binding constraints and anaphor resolvers;

        • a systematic classification of types of anaphora and a clarification of the coordination between semantic representation and reference processing.
 

When looking for accommodating binding constraints in a constraint-based setup for grammatical knowledge representation and processing, we provided:

        • a specification of a grammar with binding constraints without resorting to special purpose description devices (using HPSG);

        • a definition of a new polynomial algorithm for verifying binding constraints embedded in their constraint-based specification, as an alternative to the exponential, mainstream one.
 

Finally, we carried out:

        • a demonstration by example of the practical viability of the specification developed by providing an implementation of a support grammar and of the specified binding constraints in that grammar (using Gregor Erbach's ProFIT);
 

These results form a sort of serendipitous chain in the sense that each led to the discovery of the next, although they do not form an implicative chain.

Developing an HPSG/ProFIT grammar for binding constraints does not require a full-fledged DRT account of obliqueness quantification, although this account is highly useful in illuminating how to proceed when developing such a grammar. By the same token, having a DRT account of obliqueness quantification does not require the existence of a fourth binding principle, although the latter was very stimulating for hypothesizing an analysis of binding constraints in terms of phase quantification over obliqueness hierarchies. And for obtaining an empirical generalization that shows up as the fourth binding constraint, it is not indispensable for Portuguese to have a long-distance reflexive, which happens, nevertheless, to be particularly telling with respect to falsifying the hypothesis that long-distance reflexivity can be reduced to a cyclic effect of short-distance reflexivity.

As they are logically independent, all these major results are of chief relevance for enhancing the empirical, formal and computational plausibility of each other, and together they appear as a set of coherent results.