Computational models of referring : a study in cognitive science / Kees van Deemter.

Format:

Book

Author/Creator:

Deemter, Kees van, author.

Language:

English

Subjects (All):

Reference (Linguistics).

Presupposition (Logic).

Computational linguistics.

Physical Description:

x, 339 pages ; 24 cm

Place of Publication:

Cambridge, Massachusetts : The MIT Press, [2016]

Summary:

To communicate, speakers need to make it clear what they are talking about. The act of referring, which anchors words to things, is a fundamental aspect of language. In this book, Kees van Deemter shows that computational models of reference offer attractive tools for capturing the complexity of referring. Indeed, the models van Deemter presents cover many issues beyond the basic idea of referring to an object, including reference to sets, approximate descriptions, descriptions produced under uncertainty concerning the hearer's knowledge, and descriptions that aim to inform or influence the hearer. The book, which can be read as a case study in cognitive science, draws on perspectives from across the cognitive sciences, including philosophy, experimental psychology, formal logic, and computer science. Van Deemter advocates a combination of computational modeling and careful experimentation as the preferred method for expanding these insights. He then shows this method in action, covering a range of algorithms and a variety of methods for testing them. He shows that the method allows us to model logically complicated referring expressions, and demonstrates how we can gain an understanding of reference in situations where the speaker's knowledge is difficult to assess or where the referent resists exact definition. Finally, he proposes a program of research that addresses the open questions that remain in this area, arguing that this program can significantly enhance our understanding of human communication. Book jacket.

Contents:

I First Part: Setting the Stage 5

1 Aims and Scope of This Book 7

1.1 Aims and Main Thesis 8

1.2 Reference in Practical Applications of Computing 12

1.3 Computational Models of Reference Production 14

1.4 Determining the Information Content of an RE 16

1.5 Focus on Speakers or Hearers? 18

1.6 Referring in One Shot 19

1.7 A Perspective on Reference: Information Sharing 21

1.8 Summary of the Chapter 23

2 Theories of Reference 25

2.1 What Makes a Referring Expression? 25

2.2 Knowing What Something Is 28

2.3 Denotation and Connotation 30

2.4 The Russell-Strawson Debate 32

2.5 Intensional Contexts 35

2.6 Attributive Descriptions and Misdescriptions 37

2.7 Proper Names 39

2.8 The Gricean Maxims and Relevance Theory 41

2.9 Summary of the Chapter 43

3 The Psychology of Reference Production 45

3.1 Common Ground 45

3.2 Audience Design and the Egocentricty Debate 50

3.3 Rationality and the Gricean Maxims 55

3.4 Intrinsic Preference for Certain Attributes 60

3.5 Comparing Preference with Discrimination 63

3.6 Insights from Dialogue 66

3.7 Ecological Validity of Experiments 68

3.8 Summary of the Chapter 69

II Second Part: Solving the Classic Reg Problem 71

4 Getting Computers to Refer 73

4.1 Computational Pre-history of REG 73

4.2 The California School 77

4.3 The Classic REG Task 80

4.4 Assumptions Behind the Classic REG Task 83

4.5 Exploring the Gricean Angle Computationally 86

4.6 The Incremental Algorithm 90

4.7 Logical (In)completeness 94

4.8 Computational Tractability of REG Algorithms 97

4.9 Salience 99

4.10 Summary of the Chapter 102

5 Testing REG Algorithms: The TUNA Experiment 105

5.1 Why the TUNA Experiment? 107

5.2 How to Test a REG Algorithm? 109

5.3 The TUNA Corpus and Its Annotation 111

5.4 Analysis of the Furniture Corpus 117

5.5 Analysis of the People Corpus 120

5.6 Modelling a Plurality of Speakers 122

5.7 Lessons from the TUNA Experiment 124

5.8 Lessons from the TUNA Evaluation Challenges 125

5.9 A Note on Alternative Metrics 127

5.10 Summary of the Chapter 128

6 Probabilistic and Other Alternatives to the Classic REG Algorithms 129

6.1 Variations in Language Production 130

6.2 Bayesian Models of Reference 133

6.3 Probabilistic Referential Overspecification: the PRO Algorithm 136

6.4 Constraint Satisfaction for REG 144

6.5 Krahmer et al.'s Cost-Based Approach 149

6.6 Appek's Heirs: Reference as Part of a Wider Problem 152

6.7 Summary of the Chapter 156

III Third Part: Generating a Wider Class of Res 159

7 First Extension: Using Proper Names 161

7.1 Why Have Proper Names Been Neglected in REG? 162

7.2 Incorporating Proper Names into REG 163

7.3 Reifying Properties 166

7.4 Challenges for REG Posed by Proper Names 167

7.5 Summary of the Chapter 169

8 Second Extension: Referring to Sets 171

8.1 Purely Conjunctive References to Sets 171

8.2 Negation and Disjunction 175

8.3 Satellite Sets and Their Use in REG 178

8.4 Generating Boolean Logical Forms Incrementally 181

8.5 Optimization of Generated REs 185

8.6 Issues Raided by the Algorithms Proposed 186

8.7 Lexical Coherence in Conjoined RES 187

8.8 Avoiding Surface Ambiguities 192

8.9 Beyond Sets of Objects 197

8.10 Summary of the Chapter 198

9 Third Extension: Using Gradable Properties 201

9.1 The Semantics of Vague Descriptions 202

9.2 Pragmatic Constraints on What Can Be Said 204

9.3 Empirical Grounding 205

9.4 Computational Generation of Vague Descriptions 206

9.5 Puzzles for Incremental Content Determination 212

9.6 A Case Study: Real-World Objects and Their Sizes 214

9.7 Can We Ever Be Clear? Salience as a Gradable Property 220

9.8 Summary of the Chapter 222

10 Fourth Extension: Exploiting Modern Knowledge Representation 225

10.1 Knowledge Representation and REG 226

10.2 Description Logic: a Primer 228

10.3 Applying Description Logic to Familiar REG Problems 230

10.4 Exploiting the Full Power of Di. 234

10.5 Using SROTQ⁺ to Generate Complex REs 238

10.6 Rethinking REG: Using Shared Knowledge That Is Not Atomic 242

10.7 Why Study the Generation of Logically Complex Res? 246

10.8 Summary of the Chapter 249

11 The Question of Referability 251

11.1 Revisiting the Logical Completeness of REG 251

11.2 Limitations of SROIQ⁺ and the GROWL Algorithm 257

11.3 Even More Expressive Algorithms? 259

11.4 Summary of the Chapter 260

IV Fourth Part: Generalizing Reference Generation 261

12 First Challenge: Large Domains 263

13 Second Challenge: Breakdown of Common Knowledge 273

14 Third Challenge: Approximate Reference 281

15 Fourth Challenge: Going Beyond Identification 285

Summary of Part IV: Complexities of Information Sharing 292

V Epilogue 293

16 Epilogue 295

16.1 REG Algorithms as Cognitive Models 296

16.2 The Gricean Maxims and the Principle of Intrinsic Preference 300

16.3 Future Research: The Way Ahead 304.

Notes:

Includes bibliographical references and index.

ISBN:

9780262034555

0262034557

OCLC:

927241516