●導讀
Preface
Acknowledgements
Introduction
1.1 What This Book Is About
1.1.1 Why Do Spoken Language Translation?
1.1.2 What Are the Basic Problems?
1.1.3 What Is It Realistic to Attempt Today?
1.1.4 What Have We Achieved?
1.2 Overall System Architecture
1.3 An Illustrative Example
1.4 In Defence of Hand-Coded Grammars
1.5 Hybrid Transfer
1.5.1 The Need for Grammatical Knowledge
1.5.2 The Need for Preferences
1.6 Speech Processing
1.7 Corpora

Part 1 Language Processing and Corpora
Translation Using the Core Language Engine
2.1 Introduction: lti-Engine Translation
2.2 Word-to-Word Translation
2.3 Quasi Logical Form
2.3.1 Introduction
2.3.2 Structure of QLF
2.3.3 QLF as a Transfer Formalism: Examples
2.3.4 Head-Head Relations in QLF
2.4 Unification Grammar and QLFs
2.4.1 The CLE Unification Grammar Formalism
2.4.2 Unification Grammar Example: French Noun Phrases
2.4.3 Example 2a: Clauses in Swedish
2.4.4 Example 2b: Relative Clauses in Swedish
2.5 Orthographic Analysis and the Lexicon
2.6 Transfer Rules
2.6.1 Pre- and ttransfer
2.7 The QLF-Based Processing Path
2.7.1 Linguistic Analysis
2.7.2 Transfer and Transfer Preferences
2.7.3 Generation
2.8 Summary

Grammar Specialisation
3.1 Introduction
3.2 Explanation-Based Learning for Grammar
Specialisation
3.2.1 A Definition of Explanation-Based Learning
3.2.2 Explanation-Based Learning on Unification Grammars
3.2.3 Category Specialisation
3.2.4 Elaborate Cutting-Up Criteria
3.3 An LR Parsing Method for Specialised Grammars
3.3.1 Basic LR Parsing
3.3.2 Prefix Merging
3.3.3 Abstraction
3.4 Empirical Results
3.4.1 Experimental Setup
3.4.2 Discussion of Results
3.5 Conclusions

Choosing among Interpretations
4.1 Properties and Discriminants
4.2 Constituent Pruning
4.2.1 Discriminants for Pruning
4.2.2 Deciding Which Edges to Prune
4.2.3 Probability Estimates for Discriminants
4.2.4 Relation to Other Pruning Methods
4.3 Choosing among QLF Analyses
4.3.1 Analysis Choice: An Example
4.3.2 Further Advantages of a Discriminant Scheme .
4.3.3 Numerical Metrics
4.4 Choosing among Transferred QLFs
4.5 Choosing Paths in the Chart

The TreeBanker
5.1 Motivation
5.2 Representational Issues
5.3 Overview of the TreeBanker
5.4 The Supervised Training Process
5.4.1 Properties and Discriminants in Training
5.4.2 Additional Functionality
5.5 Training for Transfer Choice
5.6 Evaluation and Conclusions

Acquisition of Lexical Entries
6. 1 The Lexical Acquisition Tool, LexMake
6.2 Acquiring Word-to-Word Transfer Rules
6.3 Evaluation and Conclusions

Spelling and Morphology
7.1 Introduction
7.2 The Description Language
7.2.1 Morphophonology
7.2.2 Word Formation and Interfacing to Syntax
7.3 Compilation
7.3.1 Compiling Spelling Patterns
7.3.2 Representing Lexical Roots
7.3.3 Applying Obligatory Rules
7.3.4 Interword Rules
7.3.5 Timings
7.4 Some Examples
7.4.1 ltiple-Letter Spelling Changes
7.4.2 Using Features to Control Rule Application
7.4.3 Interword Spelling Changes
7.5 Debugging the Rules
……
Part 2 Linguistic Coverage
Part 3 Speech Processing

計算語言學研究濫觴於上世紀五六十年代的機器翻譯研究。中文的相關研究也幾乎同步開始，1960年起在柏克萊加州大學研究室、鄒嘉彥、C.Y.Dougherty等人已開始研究中英、中俄機器翻譯。他們的中文計算語言學研究，可說是與世界最很好科技同步的。

    棒性和繫統的自適應能力。同書面語相比，口語的聲學特性有一定的特殊性，這類語音的基頻、時長、幅度等特征都隨表達內容、感情色彩等不同，變化的範圍比朗讀語音大得多，同時還有非語聲信號和噪聲，充分研究這些特性，建立精細的聲學模型非常重要。而且，講話人往往是在較強的背景噪聲或多講話人環境下發音的，如果是電話自動語音翻譯繫統，還存在通訊干擾等其他因素的影響，因此，提高語音識別在不同說話人、不同聲學環境及通道條件下的魯棒性，在口語翻譯繫統中尤其重要。另外，在語言學層面，口語句子中含有大量的修正、重復、口頭語、省略等非規範語言現像，研究這些特征，對語言模型進行完善，包括建模、算法和訓練等各個方面，將有助於提高語音識別的正確率。
    （2）翻譯方法有待於進一步研究。盡管統計翻譯方法具有較高的魯棒性，但是，對等