Victor Sadler

WORKING

WITH

ANALOGICAL SEMANTICS:

Disambiguation Techniques in DLT

 

1989

FORIS PUBLICATIONS

Dordrecht Holland/Providence RI U.S.A.

 

(Distributed Language Translation 5)

 

[copyright Mouton de Gruyter]

 

Table of Contents

Acknowledgements 8
Introduction 9

Part I: Prototype R&D

Chapter 1. The field: Types of ambiguity relevant to machine translation 15

1.1 Types and terms . 17

1.1.1           Lexical ambiguity 18

1.1.2           Relational ambiguity 19

1.2 The scale of difficulty 25

Chapter 2. The background: Development of the prototype architecture 27

2.1          Dumb syntax, smart semantics 30

2.2    The knowledge sources 33

 

2.2.1            Potential sources 33

2.2.2    Knowledge sources in the prototype 36

Chapter 3. The theory: Assessing plausibility 39

3.1          What constitutes plausibility? 40

3.2    Methods of assessing plausibility 41

 

3.2.1            The conventional approach: features and primitives 41

3.2.2    The DLT prototype: collecting the evidence 43

3.2.3            The first DLT prototype: making meaning explicit 46

3.2.4    The second DLT prototype: leaving meaning implicit 50

Chapter 4. The techniques: Reasoning by analogy 53

4.1           The word match 55

4.2    Disambiguating the source language 59

 

4.2.1            Lexical choice: the word pair match 59

4.2.2    The Y match 61

4.2.3    The X match 62

4.2.4    Combining the X and Y matches 64

4.2.5    Combining word pair matches 64

4.2.6    Resolving relational ambiguities 66

4.2.7    The disambiguation dialogue 68

4.3 Disambiguating the intermediate language 71

4.3.1    Contextual cues in the bilingual dictionary 72

4.3.2    Word match and expectancy match 74

4.3.3    Method of the expectancy match 76

4.3.4    Combining the match scores 77

4.3.5    Functional disambiguation 78

Chapter 5. The evaluation: Tests and limitations 81

5.1 Some tests of SWESIL 82

5.1.1            Testing the word match function 82

5.1.2    Testing the word pair match function 86

5.1.3    The Meijby Test 89

5.1.4    Applications for referential disambiguation 98

5.2 The limitations of SWESIL 99

5.2.1            Strategic problems 99

5.2.2    Limitations on existing functions 102

5.2.3    Adding new functions 104

Part II: Design for a production system

Chapter 6. The Bilingual Knowledge Bank: An integrated knowledge source for machine translation 109

6.1 Rationale for a BKB 110

6.1.1            Objectivity: the corpus as primary knowledge resource 110

6.1.2    Dictionary building: the need for a bilingual corpus 111

6.1.3    Reversibility: the use of bilingual context 112

6.1.4    Consistency: discarding the LKB 114

6.1.5    Quantity: structuring the corpus 116

6.1.6    Scope: increasing the breadth of knowledge 119

6.1.7    Specificity: processing encyclopaedic knowledge 122

6.1.8    Sensitivity: distinguishing word senses 124

6.1.9    Productivity: towards full alignment 126

 

6.1.10    Dynamicity: updating the knowledge sources 131

6.1.11             Probability: relativising the frequencies 132

6.1.12    The conclusion: the need for an on-line corpus 132

6.1.13    Comparison with other research 133

6.2 Constructing the BKB 137

6.2.1             Structural disambiguation 137

6.2.2    Identifying translation units 139

6.2.3    Identifying referents 140

6.3 Advantages and spin-offs of the BKB approach 143

6.3.1            Basic advantages 143

6.3.2    The spin-offs 145
6.3.
3 Possible objections 146

 

Chapter 7. Disambiguation with a BKB: Something old and something new, something borrowed... 149

7.1 Towards a new process architecture 153

7.1.1    Selecting the translation units 155

7.1.2    Challenging the selections 156

7.1.3    Backtracking 157

7.2 Examples of (simulated) BKB-based disambiguation 158

7.2.1            An example of lexical disambiguation 159

7.2.2    Choosing between TL synonyms 182

7.2.3    Structural disambiguation 186

7.2.4    Functional disambiguation 188

7.2.5    Referential disambiguation 190

7.3 Principles and techniques . 193

7.3.1            TU selection by the Metataxor 193

7.3.2    Semantic coherence and the Examiner 195

7.3.3    Evaluating referential relations 198

7.3.4    Evaluating functional relations 198

7.3.5    Backtracking and semantic feedback 200

7.4 A new look at the disambiguation dialogue 202

7.4.1    Structural ambiguity 203

7.4.2    Referential ambiguity 204

7.4.3    Functional ambiguity 204

7.4.4    Lexical ambiguity 205

Chapter 8. Towards intelligent disambiguation: Inferencing over the BKB 207

8.1 Inferring indirect referential relations 209

8.1.1            Inference based on explicit references 209

8.1.2    Inference in the absence of explicit references 212

 

8.2    Recognizing contradictions and inconsistencies 218

8.3          Exploring deeper implications 223

8.4    The BKB as a basis for inference procedures 227

 

8.4.1            Discovering inference rules 227

8.4.2    Applying inference rules 229

8.4.3            The knowledge representation 230

Chapter 9. Spin-off: BKBs, MKBs and other animals... 235

9.1           Other translation environments 236

9.2    Monolingual knowledge banks 241

9.3           Database applications 243

In conclusion 246
References 247
Index 252