基于组合模式的语法检查 ——短语结构语法与依存语法之结合

论文摘要

合语法性检查是计算语言学研究领域中一个值得深入研究的问题。本研究主要讨论英语的合语法性检查。对于该问题,国外已有一些研究,如Kukich(1992)、Atwell and Elliott(1987)、Douglas and Dale(1992)、Golding and Schabes(1996;Kukich,1992)、Bustamante and León(1996)、Gojenola and Oronoz(2000)等都从不同的角度展开了讨论。此外,还有一些语法检查系统研究项目也值得我们回顾并参考。其中有Kuenning的Ispell、Atkinson的Aspell、Bernth的EasyEnglish、Jensen et al.的Critique、Bustamante and León的GramCheck以及Bredenkamp et al.的FLAG。虽然语法检查系统的开发可以从确定性模型和非确定性模型两种不同的模式展开,但直接运用已有的句法分析系统来开发语法检查系统值得我们探讨。这样,想要设计一个比较成功的语法检查器,必须要有一个非常健壮的句法分析系统。目前,基于不同理论框架、应用不同方法构建的句法分析系统已有不少。其中,基于统计学模型的语法分析系统,如Collins(1999),准确率高,但速度不够快,应用于机辅适应性教学的服务器对学习者的输出进行实时处理不够理想。而链语法模型(Sleator&Temperley,1991;Lafferty et al.,1992)不但有较高的准确率,更重要地是,它在速度方面也有其过人之处。基于组合模式的英语语法检查系统正是基于链语法分析器的。因为链语法分析器的理论基础是链语法,它是一种与依存语法有着很近关系的语法模

论文目录

Table of Contents

List of Tables

List of Figures

Declaration

Acknowledgements

Curriculum Vitae

Major Publications

Abstract

摘要

1 Introduction

1.1 Introduction

1.2 Research overview

1.3 Outline of the dissertation

2 Literature Review

2.1 Introduction

2.2 Starting from tagging and parsing

2.2.1 Taggers

2.2.1.1 Linguistic models

2.2.1.2 Statistic models

2.2.1.3 Machine-learning models

2.2.2 Parsers

2.2.2.1 Context-free grammar models

2.2.2.2 Context-free grammar models with context sensitivity

2.2.2.3 Models with lexical sensitivity

2.2.2.4 History-based models

2.3 Grammar checkers

2.3.1 Research review

2.3.2 Projects review

2.3.2.1 Ispell and Aspell

2.3.2.2 Style and Diction

2.3.2.3 EasyEnglish

2.3.2.4 Critique

2.3.2.5 CLAWS as a Grammar Checker

2.3.2.6 GramCheck

2.3.2.7 Grammar Checker by Park and his colleagues

2.3.2.8 FLAG

2.3.2.9 Style and grammar checker by Naber

2.4 Grammar formalisms

2.4.1 Dependency grammar

2.4.1.1 Lucien Tesniére

2.4.1.2 Post-Tesniére development

2.4.1.3 Link grammar

2.4.2 Link grammar: a detailed discussion

2.4.2.1 Links

2.4.2.2 Basic assumptions

2.4.2.3 Dictionary and linking requirements

2.4.2.4 Post-processing

2.4.2.5 Speed and robustness features

2.4.2.6 The cost system

2.4.2.7 Differences between Link grammar and other dependency-related grammars

2.4.3 Transformation-based machine learning algorithm

2.4.3.1 Initial state of the parser

2.4.3.2 Structural Transformations

2.4.3.3 Learning Transformations

2.5 Summary

3 Research Questions

3.1 Introduction

3.2 Researches related to link grammar

3.3 Detailed study of the Link Parser

3.3.1 Merit of link grammar for the purpose of grammar checking

3.3.2 An evaluation of the parser

3.3.3 A closer examination

3.3.4 Problems to be addressed

3.4 Research questions

3.5 Objectives of this research

3.6 Summary

4 Theoretical Framework

4.1 Introduction

4.2 Phrase structure grammars vs. dependency grammars

4.3 On combined models

4.4 Grammar checking in the combined model: Link Parser incorporated with phrase structure information

4.4.1 Phase One

4.4.2 Phase Two

4.4.3 Phase Three

4.5 The Grammar checker

4.5.1 The interface

4.5.2 Speed of the grammar checker

4.5.3 Evaluation of the grammar checker

4.6 Summary

5 Research Design

5.1 Research methodology

5.2 Implementation of the combined model

5.2.1 Preparing the training corpus

5.2.2 Training in transformation-based machine learning algorithm

5.2.2.1 The comparison algorithm

5.2.2.2 Training process

5.2.3 Applying the knowledge to the post processing stage in the Link Parser

5.3 Evaluation of the combined model

5.4 Summary

6 Research Outcome: Evaluating the Combined Model

6.1 Introduction

6.2 Experiment Ⅰ

6.2.1 Objectives of the experiment

6.2.2 Test set

6.2.3 Result and analysis

6.2.4 Case study

6.3 Experiment Ⅱ

6.3.1 Objectives of this experiment

6.3.2 Test set

6.3.3 Result and analysis

6.4 Performance of the grammar checker

6.5 Summary

7 Conclusion

7.1 Conclusion

7.2 Remaining issues and future work

7.2.1 Error and error judgment

7.2.2 Adaptiveness to CLEC

7.2.3 Code-level refinement

A The C source code of the main part of the grammar checker

B The Perl 5.6 code of the comparison algorithm

C The rule-set learned for the transformation-based machine-learning algorithm

D Test set of the 83 sentences in Experiment Ⅰ

E Re-ranking of the 16 linkages of Sentence 40 in Experiment Ⅰ

F Test set of the 427 sentences in Experiment Ⅱ

G Downloading address of the Grammar Checker for evaluation

Bibliography

Index

基于组合模式的语法检查 ——短语结构语法与依存语法之结合

论文摘要

论文目录

相关论文文献

猜你喜欢