Translation, Brains and the Computer: A Neurolinguistic Solution to Ambiguity and Complexity in Machine Translation (Machine Translation: Technologies and Applications, 2)

This book is about machine translation (MT) and the classic problems associated with this language technology. It examines the causes of these problems and, for linguistic, rule-based systems, attributes the cause to language’s ambiguity and complexity and their interplay in logic-driven processes. For non-linguistic, data-driven systems, the book attributes translation shortcomings to the very lack of linguistics. It then proposes a demonstrable way to relieve these drawbacks in the shape of a working translation model (Logos Model) that has taken its inspiration from key assumptions about psycholinguistic and neurolinguistic function. The book suggests that this brain-based mechanism is effective precisely because it bridges both linguistically driven and data-driven methodologies. It shows how simulation of this cerebral mechanism has freed this one MT model from the all-important, classic problem of complexity when coping with the ambiguities of language. Logos Model accomplishes this by a data-driven process that does not sacrifice linguistic knowledge, but that, like the brain, integrates linguistics within a data-driven process. As a consequence, the book suggests that the brain-like mechanism embedded in this model has the potential to contribute to further advances in machine translation in all its technological instantiations. Dedication 6 Preface 8 Acknowledgements 10 Contents 12 Part I 16 Chapter 1: Introduction 17 References 25 Chapter 2: Background 26 2.1 Logos Model Beginnings 26 2.2 The Advent of Statistical MT 29 2.2.1 Pattern-Based Processes in SMT and Logos Model 30 2.3 Overview of Logos Model Translation Process 33 2.4 Psycholinguistic and Neurolinguistic Assumptions 38 2.4.1 First Assumption 38 2.4.2 Second Assumption 39 2.4.3 Third Assumption 39 2.5 On Language and Grammar 40 2.5.1 The Origin and Nature of Grammar 41 2.5.2 Language, Grammar and Associative Memory 42 2.5.3 In Principio Erat Verbum 43 2.6 A Note About Neural MT (NMT) 45 2.7 Conclusion 46 Postscripts 47 Postscript 2-A 47 Postscript 2-B 50 Postscript 2-C 50 References 51 Chapter 3: Language and Ambiguity: Psycholinguistic Perspectives 53 3.1 Levels of Ambiguity 53 3.2 Language Acquisition and Translation 56 3.2.1 Linguistic Processes Involved in Second Language Acquisition 57 3.2.2 On Learning French 61 3.3 Psycholinguistic Bases of Language Skills 63 3.3.1 Analytical Basis 63 3.3.2 Empirical Basis 64 3.3.3 Analogical Basis 64 3.4 Practical Implications for MT 66 3.4.1 Semantico-Syntactic Solutions to the Problem of Ambiguity in MT 67 3.4.1.1 Some Classic Examples 67 3.5 Psycholinguisitcs in a Machine 70 3.5.1 Generate Target Translation16 73 3.5.2 OpenLogos and Logos Model 73 3.6 Conclusion 75 References 75 Chapter 4: Language and Complexity: Neurolinguistic Perspectives 76 4.1 On Cognitive Complexity 76 4.2 A Role for Semantic Abstraction and Generalization 81 4.3 Connectionism and Brain Simulation 84 4.4 Logos Model As a Neural Network 87 4.5 Language Processing in the Brain 90 4.5.1 Cortical Circuits and Logos Model 91 4.5.2 Hippocampus and Logos Model 92 4.6 MT Performance and Underlying Competence 100 4.7 Conclusion 102 Postscripts 102 Postscript 4-A 102 Postscript 4-B 103 Postscript 4-C 105 Postscript 4-D 105 Postscript 4-E 105 Postscript 4-F 106 Postscript 4-G 106 Postscript 4-H 107 References 107 Chapter 5: Syntax and Semantics: Dichotomy Versus Integration 110 5.1 Syntax Versus Semantics: Is There a Third, Semantico-­Syntactic Perspective? 110 5.2 Recent Views of the Cerebral Process 117 5.3 Syntax and Semantics: How Do They Relate? 119 5.4 Conclusion 124 Postscripts 125 Postscript 5-A 125 Postscript 5-B 126 Postscript 5-C 126 Postscript 5-D 127 Postscript 5-E 128 Postscript 5-F 129 Postscript 5-G 130 References 134 Chapter 6: Logos Model: Design and Performance 137 6.1 The Translation Problem 137 6.1.1 Five Fundamental Design Decisions 139 6.2 How Do You Represent Natural Language? 140 6.2.1 Effectiveness of SAL for Deterministic Parsing 141 6.3 How Do You Store Linguistic Knowledge? 143 6.3.1 General Remarks 143 6.3.2 Logos Model Lexicon 143 6.3.3 The Pattern-Rule Database 144 6.4 How Do You Apply Stored Knowledge to the Input Stream? 147 6.4.1 Modules RES1 and RES2 (R1 and R2) 148 6.4.1.1 Homograph Resolution 148 6.4.1.2 Garden Path Resolution 149 6.4.1.3 Clausal Ambiguity Resolution 150 6.4.2 Module PARSE1 150 6.4.2.1 Simple Noun Phrases 150 6.4.2.2 Complex Noun Phrases 151 6.4.2.3 Adjective Scoping 151 6.4.2.4 Auxiliary Verb Phrases 152 6.4.3 Module PARSE2 152 6.4.3.1 Prepositional Phrase Complementation 152 6.4.3.2 Relative Clause Complementation 153 6.4.4 Module PARSE3 155 6.4.5 Module PARSE4 159 6.5 How Do You Effect Target Generation? 163 6.5.1 Target Components 163 6.6 How Do You Cope with Complexity? 163 6.6.1 A Final Illustration 164 6.7 Conclusion 167 Postscripts 167 Postscript 6-A 167 Postscript 6-B 168 References 172 Chapter 7: Some Limits on Translation Quality 173 7.1 First Example 174 7.2 Second Example 176 7.3 Other Translation Examples 177 7.4 Balancing the Picture 178 7.5 Conclusion 179 References 181 Chapter 8: Deep Learning MT and Logos Model 182 8.1 Points of Similarity and Differences 183 8.2 Deep Learning, Logos Model and the Brain 188 8.3 On Learning 190 8.4 The Hippocampus and Continual Learning 195 8.5 Conclusion 202 8.6 A Final Demonstration 205 References 209 Part II 212 Chapter 9: The SAL Representation Language 213 9.1 Overview of SAL 213 9.2 SAL Parts of Speech 214 9.2.1 Open Classes (Table 9.1) 214 9.2.2 Closed Classes (Table 9.2) 214 9.3 SAL Nouns (WC 1) 215 9.3.1 Aspective Nouns 217 9.3.2 Concrete Nouns 219 9.3.3 Animate Nouns 220 9.3.4 Abstract Nouns 221 9.3.5 Measure Nouns 222 9.3.6 Place Nouns 223 9.3.7 Mass Nouns 224 9.3.8 Information and Time Nouns (Fig. 9.10) 225 9.4 SAL Verbs (WC 2) 226 9.4.1 The Intransitive-Transitive Verb Spectrum 227 9.4.1.1 Intransitives (Overview) 227 9.4.1.2 Simple Transitives (Overview) 228 9.4.1.3 Preverbal Transitives, Simple and Complex (Overview) 228 9.4.1.4 Preclausal Transitives (Overview) 229 9.4.2 Intransitive Verbs 230 9.4.3 Subjective Transitive Verbs 231 9.4.4 Reciprocal Transitive Verbs 232 9.4.5 Ditransitive Verbs 233 9.4.6 Objective Transitive Verbs 234 9.4.7 Pre-process Verbs 235 9.4.8 Simple Preverbal Verbs 236 9.4.9 Preverbal Complex Verbs 237 9.4.10 Preverbal-Preclausal Verbs 238 9.4.11 Preclausal Verbs 239 9.5 SAL Adjectives (WC 4) 241 9.5.1 Preclausal/Preverbal Adjectives 242 9.5.2 Preverbal Adjectives 243 9.5.3 Adverbial Adjectives 244 9.5.4 Non-adverbial Adjectives 245 9.6 SAL Adverbs (WC 3 and WC 6) 246 9.6.1 Locative Adverbs have the Following Supersets 246 9.6.2 Non-locative Adverb have the Following Supersets 247 Postscript 248 Postscript 9-A 248 Front Matter ....Pages i-xvi Front Matter ....Pages 1-1 Introduction (Bernard Scott)....Pages 3-11 Background (Bernard Scott)....Pages 13-39 Language and Ambiguity: Psycholinguistic Perspectives (Bernard Scott)....Pages 41-63 Language and Complexity: Neurolinguistic Perspectives (Bernard Scott)....Pages 65-98 Syntax and Semantics: Dichotomy Versus Integration (Bernard Scott)....Pages 99-125 Logos Model: Design and Performance (Bernard Scott)....Pages 127-162 Some Limits on Translation Quality (Bernard Scott)....Pages 163-171 Deep Learning MT and Logos Model (Bernard Scott)....Pages 173-202 Front Matter ....Pages 203-203 The SAL Representation Language (Bernard Scott)....Pages 205-241