The Talking Heads Experiment: Origins Of Words And Meanings


The Talking Heads Experiment: Origins Of Words And Meanings (2015) .. by Luc Steels


Contents

Preface iv

I The 1999 Talking Heads book 1

1 Introduction 5
1.1 The Talking Heads experiment 5
1.2 The main hypotheses 7
1.3 A bottom-up approach to artificial intelligence 9
1.4 History of the project 10
1.5 Beyond Turing 15
1.6 The book 17

2 Preview 19
2.1 The main components 19
2.1.1 Teleporting 19
2.1.2 The robots 21
2.1.3 The agents 23
2.1.4 Interactivity 24
2.2 The Guessing Game 26
2.2.1 Rules of the game 26
2.2.2 Nature of the game 27
2.2.3 The semiotic square 28
2.2.4 Processes involved in language communication 30
2.2.5 Knowledge sources and competences 31
2.3 Perception and categorisation 32
2.3.1 Scene and topic selection 32
2.3.2 Sensory channels 34
2.3.3 Making distinctions 35
2.4 Lexicalisation 37
2.4.1 Same meaning, same referent 37
2.4.2 A new word 38
2.4.3 Competition between words 39
2.4.4 Disambiguation 40
2.4.5 Same meaning, different referent 42
2.4.6 Situated grounded semantics 43
2.5 The origins of grammar 45
2.6 Conclusions 47

3 Perception 49
3.1 What sensors sense 50
3.1.1 Artificial sensors and actuators 50
3.1.2 Natural sensing 51
3.1.3 Behaviours 51
3.2 Segmentation 53
3.2.1 Feature extraction 54
3.2.2 Divergent perception 56
3.2.3 The sieve architecture 56
3.3 Sensory channels 58
3.3.1 Example channels 58
3.3.2 Conceptual spaces 59
3.3.3 Perceptual constancy 60
3.3.4 Transformations 61
3.3.5 Scaling 61
3.3.6 Saliency 62
3.4 Methodology 65
3.4.1 Putting up scaffolds 66
3.4.2 Idealisation and realism 67
3.5 The geom world 70
3.6 Conclusions 72

4 The Discrimination Game 75
4.1 The paradoxes of meaning 76
4.1.1 The empiricist’s stance 77
4.1.2 The rationalist’s stance 78
4.1.3 Arguments for and against rationalism 78
4.1.4 Arguments for and against empiricism 80
4.2 Selectionism 81
4.2.1 Principles of selectionism 81
4.2.2 Selectionist cognitive systems 82
4.2.3 The tree metaphor 83
4.2.4 Deriving new sensory channels 84
4.2.5 Comparing approaches 84
4.3 Discrimination trees 85
4.3.1 Making distinctions 85
4.3.2 Categorisers 86
4.3.3 The Discrimination Game 88
4.3.4 The Pachinko machine 89
4.3.5 Competition between categories 90
4.3.6 Variations on discrimination 91
4.3.7 The Discrimination Game in action 91
4.3.8 The importance of scaling and saliency 94
4.3.9 Combinations of categories 95
4.3.10 A real world scene 95
4.4 An ecology of distinctions 97
4.4.1 Growth dynamics 97
4.4.2 Pruning dynamics 98
4.4.3 Average discriminatory success and repertoire size 98
4.4.4 Adaptivity in categorisation 100
4.4.5 Real world scenes 102
4.5 Conclusions 102

5 The Naming Game 105
5.1 Inventing a lexicon 106
5.1.1 Representing lexical associations 107
5.1.2 Updating the score 108
5.1.3 Constructing and acquiring words 109
5.1.4 The Naming Game in action 110
5.1.5 Characterising the lexicon Ill
5.1.6 Monitoring 113
5.1.7 Measuring lexical coherence 115
5.2 Scaling up 116
5.2.1 Coping with new meanings 116
5.2.2 Lexicon acquisition by virgin agents 119
5.2.3 Preservation in changing populations 120
5.3 Self-organisation 122
5.3.1 Winner-take-all processes 123
5.3.2 Collective behaviour and self-organisation 123
5.3.3 Increasing-returns economics 125
5.3.4 Lessons from nature 126
5.4 Lexical dynamics 126
5.4.1 Spatially distributed naming games 127
5.4.2 Language contact 129
5.5 Conclusions 131

6 The Guessing Game 133
6.1 Defining the Guessing Game 134
6.1.1 Example of a coupled game 135
6.1.2 Input-output coupling 140
6.1.3 Updating the scores 141
6.1.4 Repair processes 143
6.2 Synonymy 144
6.3 Ambiguity 149
6.3.1 How words may still get the same meaning 150
6.3.2 How words get different meanings 151
6.3.3 Competition between word meanings 155
6.3.4 Lexical and ontological development 157
6.4 Scaling up 159
6.4.1 Increasing the population size 159
6.4.2 Lexicon acquisition by new agents 163
6.5 Conclusions 165

7 Grounding 167
7.1 A first grounding experiment 168
7.1.1 Integrating perception and action 168
7.1.2 Concept acquisition 170
7.1.3 Generalisation without learning 172
7.1.4 The influence of the environment 174
7.1.5 Coping with perceptual anomalies 179
7.2 Semiotic dynamics 183
7.2.1 Tracking language evolution 183
7.2.2 Semiotic landscapes 183
7.2.3 Competition diagrams 185
7.2.4 RMF coherence 186
7.3 The ideal language 187
7.3.1 Total coherence 187
7.3.2 Communicative success despite incoherence 188
7.4 Damping synonymy and ambiguity 189
7.4.1 The story of fepi 190
7.4.2 The story of xu 192
7.4.3 The entry of 03 194
7.5 Rousseau’s paradox 195
7.5.1 Universality versus relativism 195
7.5.2 Ontological coherence 197
7.6 Conclusions 198

II Installations and experiments 201

8 The first series (1999) 203
8.1 The Laboratorium exhibition 203
8.2 The installation 205
8.3 Start up of the experiment 208
8.4 Results of the experiment 222
8.5 Conclusions 224

9 The second series (2000-2001) 231
9.1 The NOISE exhibition 231
9.1.1 The exhibition 231
9.1.2 Installation at Kettle’s Yard in Cambridge 234
9.1.3 Installation at the Wellcome Gallery in London 238
9.2 Iconoclasm 242
9.3 Installation at the Palais de la Decouverte in Paris 258
9.4 The portable Talking Heads 258
9.5 Look into the Box 260
9.6 Conclusions 264

III Beyond the Talking Heads 267

10 Beyond the Talking Heads experiment 269
10.1 Experiments with the aibo robots 269
10.1.1 aibo’s first words 270
10.1.2 The Perspective Reversal experiment 273
10.2 Scaling up to grammar 278
10.2.1 Early syntax experiments 278
10.2.2 The Case Grammar experiments 282
10.3 Conclusions 294

11 Language strategies for humanoid robots 295
11.1 The Proper Naming Game 296
11.1.1 Challenges 297
11.1.2 Semiotic networks 300
11.2 Action Games 304
11.3 The Colour Description Game 306
11.3.1 Compositional procedural semantics with IRL 311
11.3.2 Building blocks for natural language semantics 313
11.3.3 Strategies for colour 314
11.3.4 Translation to grammar 316
11.3.5 Influence of embodiment 320
11.4 Conclusion 321

12 Language evolution 323
12.1 Culture-driven language evolution 323
12.2 Fitness landscapes 329
12.2.1 Fitness landscapes of language systems 330
12.2.2 The fitness landscape of language strategies 335
12.3 Selection and alignment of language strategies 338
12.4 Generation of new strategies 343
12.4.1 A meta-strategy for generating new conceptualisation strategies 344
12.4.2 Meta-strategies for generating new lexicogrammatical strategies 347
12.5 Conclusions 354

Bibliography 357

Index 369
Name index 369
Subject index 373

(Visited 7 times, 1 visits today)