Phoneme-based statistical transliteration of foreign names for OOV problem.

Gao Wei.Thesis (M.Phil.)--Chinese University of Hong Kong, 2004.Includes bibliographical references (leaves 79-82).Abstracts in English and Chinese.Abstract --- p.iAcknowledgement --- p.iiiBibliographic Notes --- p.vChapter 1 --- Introduction --- p.1Chapter 1.1 --- What is Transliteration? --- p.1Chapter 1.2 --- Existing Problems --- p.2Chapter 1.3 --- Objectives --- p.4Chapter 1.4 --- Outline --- p.4Chapter 2 --- Background --- p.6Chapter 2.1 --- Source-channel Model --- p.6Chapter 2.2 --- Transliteration for English-Chinese --- p.8Chapter 2.2.1 --- Rule-based Approach --- p.8Chapter 2.2.2 --- Similarity-based Framework --- p.8Chapter 2.2.3 --- Direct Semi-Statistical Approach --- p.9Chapter 2.2.4 --- Source-channel-based Approach --- p.11Chapter 2.3 --- Chapter Summary --- p.14Chapter 3 --- Transliteration Baseline --- p.15Chapter 3.1 --- Transliteration Using IBM SMT --- p.15Chapter 3.1.1 --- Introduction --- p.15Chapter 3.1.2 --- GIZA++ for Transliteration Modeling --- p.16Chapter 3.1.3 --- CMU-Cambridge Toolkits for Language Modeling --- p.21Chapter 3.1.4 --- Re Write Decoder for Decoding --- p.21Chapter 3.2 --- Limitations of IBM SMT --- p.22Chapter 3.3 --- Experiments Using IBM SMT --- p.25Chapter 3.3.1 --- Data Preparation --- p.25Chapter 3.3.2 --- Performance Measurement --- p.27Chapter 3.3.3 --- Experimental Results --- p.27Chapter 3.4 --- Chapter Summary --- p.28Chapter 4 --- Direct Transliteration Modeling --- p.29Chapter 4.1 --- Soundness of the Direct Model一Direct-1 --- p.30Chapter 4.2 --- Alignment of Phoneme Chunks --- p.31Chapter 4.3 --- Transliteration Model Training --- p.33Chapter 4.3.1 --- EM Training for Symbol-mappings --- p.33Chapter 4.3.2 --- WFST for Phonetic Transition --- p.36Chapter 4.3.3 --- Issues for Incorrect Syllables --- p.36Chapter 4.4 --- Language Model Training --- p.36Chapter 4.5 --- Search Algorithm --- p.39Chapter 4.6 --- Experimental Results --- p.41Chapter 4.6.1 --- Experiment I: C.A. Distribution --- p.41Chapter 4.6.2 --- Experiment II: Top-n Accuracy --- p.41Chapter 4.6.3 --- Experiment III: Comparisons with the Baseline --- p.43Chapter 4.6.4 --- Experiment IV： Influence of m Candidates --- p.43Chapter 4.7 --- Discussions --- p.43Chapter 4.8 --- Chapter Summary --- p.46Chapter 5 --- Improving Direct Transliteration --- p.47Chapter 5.1 --- Improved Direct Model´ؤDirect-2 --- p.47Chapter 5.1.1 --- Enlightenment from Source-Channel --- p.47Chapter 5.1.2 --- Using Contextual Features --- p.48Chapter 5.1.3 --- Estimation Based on MaxEnt --- p.49Chapter 5.1.4 --- Features for Transliteration --- p.51Chapter 5.2 --- Direct-2 Model Training --- p.53Chapter 5.2.1 --- Procedure and Results --- p.53Chapter 5.2.2 --- Discussions --- p.53Chapter 5.3 --- Refining the Model Direct-2 --- p.55Chapter 5.3.1 --- Refinement Solutions --- p.55Chapter 5.3.2 --- Direct-2R Model Training --- p.56Chapter 5.4 --- Evaluation --- p.57Chapter 5.4.1 --- Search Algorithm --- p.57Chapter 5.4.2 --- Direct Transliteration Models vs. Baseline --- p.59Chapter 5.4.3 --- Direct-2 vs. Direct-2R --- p.63Chapter 5.4.4 --- Experiments on Direct-2R --- p.65Chapter 5.5 --- Chapter Summary --- p.71Chapter 6 --- Conclusions --- p.72Chapter 6.1 --- Thesis Summary --- p.72Chapter 6.2 --- Cross Language Applications --- p.73Chapter 6.3 --- Future Work and Directions --- p.74Chapter A --- IPA-ARPABET Symbol Mapping Table --- p.77Bibliography --- p.8