A Formal Approach to Discourse Anaphora

Bonnie Lynn Webber, Bonnie Lynn Webber

• Publisher: Routledge
• Year: 2016
• ISBN:
  9781315403328
• Edition: 1
• Topic: Languages & Linguistics
• Subtopic: Linguistics
• Index: Languages & Linguistics

First published in 1979, this book starts from the perspective that dealing with anaphoric language can be decomposed into two complementary tasks: 1. identifying what a text potentially makes available for anaphoric reference and 2. constraining the candidate set of a given anaphoric expression down to one possible choice. The author argues there is an intimate connection between formal sentential analysis and the synthesis of an appropriate conceptual model of the discourse. Some of the issues with the creation of this conceptual model are discussed in the second chapter, which follows a background to the thesis that catalogues the types of anaphoric expression available in English and lists the types of things that can be referred to anaphorically. The third and fourth chapters examine two types of anaphoric expression that do not refer to non-linguistic entities. The final chapter details three areas into which this research could potentially be extended. This book will be of interest to students of linguistics.

Chapter 1. Introduction

Table of Contents

1. Statement of the Problem

2. The Range of Discourse Anaphora

3. Historical Background

3.1 Factors Influencing Anaphor Resolution

3.1.1 Number/gender agreement

3.1.2 Backwards Anaphora Constraint

3.1.3 Theme

3.1.4 Role Inertia

3.1.5 Semantic Sectional Restrictions

3.1.6 Recency and Scene Shifts

3.1.7 Implicit Causality

3.1.8 Possible Worlds

3.2 Methods of Simplifying Anaphor Resolution

3-3 Previous Research on Verb Phrase Ellipsis

4. The Range of Antecedents and Referents

4.1 Individuals

4.2 Sets

4.3 Stuff

4.4 Generics

4.5 Prototypes

4.6 Actions, Events, States, Propositions,

4.7 Descriptions

4.8 Predicates

5. Fundamental Assumptions

6. Thesis Organization

1. Statement of the Problem

This thesis follows from a desire to make natural language man/machine communication more robust, by providing an ability to deal with anaphoric language in a perspicuous, transportable non-ad hoc way. Without such an ability, there is no hope for the extended natural language discourse required in problem solving and information seeking situations. This thesis starts from the perspective that dealing with anaphoric language can be decomposed into two complementary tasks: (1) identifying what a text potentially makes available for anaphoric reference and (2) constraining the candidate set of a given anaphoric expression down to one possible choice. In the past however, it has only been the second of the two (usually called the "anaphor resolution" problem) that has stimulated research in psychology and artificial intelligence (AI) natural language understanding.

This research in psychology and AI has produced a host of interesting examples which demonstrate the range of syntactic, semantic, social and factual knowledge that may be critical in choosing among the possible antecedents or referents for a given anaphoric expression (or alternatively, in predisposing the listener to one particular candidate). These examples have also served to demonstrate techniques for constraining that choice to only the most probable candidate(s) or creating a predisposition towards it. (I shall have more to say about this in Section 3.2.) Unfortunately, these examples have also been such as to obscure the other problem I mentioned above - that of identifying what the text makes available for anaphoric reference and how it does so. That is the focus of this thesis.

The ability to identify what the text makes available for anaphoric reference is not a trivial one, and in this thesis I will make two strong claims:

1. None of the three types of anaphoric expressions that I have studied - definite anaphora, "one"-anaphora and verb phrase deletion - can be understood in purely linguistic terms. That is, none of them can be explained without stepping out of the language into the conceptual model each discourse participant is synthesizing from the discourse.
2. On the other hand, if a discourse participant does not assign each new utterance in the discourse a formal representation in which, inter alia,
   1. quantifiers are indicated, along with their scopes;
   2. main clauses are distinguished from relative clauses and subordinate clauses;
   3. clausal subjects are separated from clausal predicates; (cf. Chapter 2, Section 2; Chapter 3, Section 2; Chapter 4, Section 2), then that discourse participant cannot identify all of what is being made available for anaphoric reference.

In either case - i.e., if either the form of the discourse sentences or the conceptual discourse model is ignored - there will be anaphoric expressions which cannot be resolved correctly. The reason is that that which is to serve as the intended antecedent or referent of that anaphor was never recognized in the first place.

2. The Range of Discourse Anaphora

In the next section (Section 3), I will be surveying significant previous research on discourse anaphora. However, in order to set this background material itself into perspective vis a vis the broad extent of the phenomenon, this section lists the types of discourse anaphora which linguists have catalogued to date.

1. Definite Pronoun Anaphora
   "Today I met a man with two heads. I found him very strange.
   him = the just-mentioned man with two heads whom I met today
2. Definite Noun Phrase Anaphora
   "Today I met a man who owned two talented monkeys. The monkeys were discussing Proust."
   the monkeys = the two just-mentioned monkeys owned by the just-mentioned man with two heads I met today
3. "One(s)" Anaphora
   "Wendy got a blue crayon for her birthday and I got a purple one."
   one = crayon
4. Verb Phrase Deletion <*1>
   "Whenever Wendy buys herself a new hat, Phyllis does 0 too."
   0 = buy herself a new hat
5. "Do it" Anaphora
   "Although the cat had to be taken to the vet, Wendy refused to do it."
   do it = take the cat to the vet
6. "Do so" Anaphora
   "If you won't take the cat to be spayed, I will do so."
   do so = take the cat to be spayed
7. Null Complement Anaphora [Hankamer & Sag, 1976]
   "Although the cat had to be taken to the vet, Wendy refused 0.
   0 = to take the cat to the vet
8. "Sentential It" Anaphora
   "Although Marilyn's cat ate a hole in Fred's coat, it didn't bother him.
   it = the fact that Marilyn's cat ate a hole in Fred's coat
   "When did Marilyn's cat eat a hole in Fred's coat? I think it happened on New Year's Eve."
   it s the event in which Marilyn's cat ate a hole in Fred's coat
9. "Sluicing" [Ross, 1969]
   "Someone asked after you, but I don't remember who 0."
   0 = asked after you
   "John attended MIT, but I don't remember when 0."
   0 = John attended MIT
10. "Gapping" [Ross, 1967]
    "Bruce eats cottage cheese on Wednesdays, and Harry 0, on Thursdays."
    0 = eats cottage cheese
11. "Stripping" [Hankamer, 1971]
    "Wendy eats half sour pickles, but 0₁ not 0₂ in her own apartment."
    0₁ = Wendy (does)
    0₂ = eat half sour pickles
12. "Such" Anaphora
    "When Mary kicked her cat, she was punished since such behavior is deplorable in little girls."
    such behavior =? kicking one's cat, kicking pets, attacking small animals,???

3. Historical Background

As I mentioned in Section 1, the problem of anaphor resolution i.e., choosing the correct antecedent or referent from among several possible candidates - has received a great deal of attention in artificial intelligence, most of it directed at resolving definite anaphora (definite pronouns and definite noun phrases). One practical reason for the general interest in dealing with anaphora arises from wanting to provide comfortable and "habitable" [Watt, 1968] natural language man/machine communication over a typewriter channel. If a person is forced to make explicit what would normally be ellipsed in communicating with another human being, s/he may find the dialogue too time-consuming and burdensome to be of benefit, especially given the additional burden of written rather than spoken communication. Moreover as [Balzer et al., 1977] point out, making such information explicit is liable to lead to more errors due to the extensive bookkeeping involved in keeping referring expressions consistent and complete.

An additional reason for this concern with anaphora is that for definite anaphora, there is no practical substitute in natural language. Naming, the alternative used in both mathematics, logic and programming, e.g.

"Let G be a barber who shaves everyone who does not shave himself." (SETQ MYLIST (MAPCAR S (...)))

has the disadvantage of requiring the speaker to know a priori what will be talked about later so that s/he can assign it a name right off. This is impractical, if not impossible, in natural language discourse, whether between people or between a person and a machine.

Anaphor resolution has received a significant amount of attention from both linguists and psychologists as well. The former have looked at it primarily in terms of "co-reference restrictions" - structural constraints within a sentence that prevent two noun phrases from being interpreted as referring to the same thing. Psychologists, on the other hand, have looked at anaphor resolution in terms of memory and processing strategies. They hope to characterize people's behavior vis a vis anaphor resolution and then use that characterization in turn as evidence for how discourse information is organized and accessed in memory. Again this work has primarily involved definite anaphora.

In order to give the reader a feeling for this research on definite anaphora, in the next section (Section 3.1) I shall present a short piece of text containing several anaphoric expressions. For each one, I shall describe various factors that have been proposed as applicable to its correct resolution. In the following section (Section 3-2), I will discuss some techniques that have been proposed by AI researchers for simplifying anaphor resolution - i.e., for predisposing the listener to one particular candidate or for constraining the reasoning that might be necessary for choosing among several possible candidates. I shall also point out that none of these techniques addresses the complementary problem I mentioned earlier - that of identifying what the text makes available for anaphoric reference. Finally in Section 3.3, I shall briefly describe the hitherto most adequate approach to verb phrase ellipsis and its remaining deficiencies that the approach to verb phrase ellipsis presented here attempts to address.

3.1 Factors Influencing Anaphor Resolution

This section is organized around a short piece of text containing several anaphoric expressions. After presenting the text, I shall describe various factors that have been proposed as applicable to the correct resolution of each expression. In many cases, the examples may not seem to justify hypothesizing these factors as an appropriate level of explanation. So interested readers are advised to consult the original sources referenced here in order to discover the range of phenomena each is meant to account for.

• (a) Fred left his niece at home and headed for the 200 with Mary and John.
• (b) Since the zoo was far away, they first asked a man down the block who owned a car whether they could borrow it.
• (c) When they got to the zoo, Fred heard that a black mamba had just escaped.
• (d) Suddenly near John he saw the snake.
• (e) The girl saw it too, as did John.
• (f) Fred admired John because he was able to catch the snake.
• (g) Fred regretted not having a stick, since he could have used it to help John.
• (h) Luckily, the friends had each brought a bottle of wine.
• (i) John volunteered to drink them all in order to forget the black mamba.

3.1.1 Number/gender agreement

One simple factor influencing the choice of an referent for "they", "he", "it", etc. is that in English, most pronouns are marked for number and gender. So in sentences (b), (c) and (i), "they" must refer to something interpretable as a se...

Table of contents

• Cover
• Half Title
• Title
• Copyright
• Original Title
• Original Copyright
• Acknowledgements
• Table of Contents
• Synopsis
• Chapter 1. Introduction
• Chapter 2. Definite Pronouns
• Chapter 3. "One" Anaphora
• Chapter 4. Verb Phrase Ellipsis
• Chapter 5. Conclusion
• Bibliography