SRILM is a collection of C++ libraries, executable programs, and helper scripts designed to allow both production of and experimentation with statistical language models for speech recognition and other applications. SRILM is freely available for noncommercial purposes. The toolkit supports creation and evaluation of a variety of language model types based on N-gram statistics, as well as several related tasks, such as statistical tagging and manipulation of N-best lists and word lattices. This paper summarizes the functionality of the toolkit and discusses its design and implementation, highlighting ease of rapid prototyping, reusability, and combinability of tools. 1.

In memory of my grandfather, LAW Yuk Kau, who passed away several months before I started this study. He gave me strength, kept me determined in my research and helped me realize my career goal. He will always live in my heart. iv ACKNOWLEDGMENTS I would like to express my deepest thanks to my supervisor Professor Dekai Wu. Not only has he shared with me his insightful ideas and comments in my research, he has also supported me throughout the whole program of my study. From the time I approached him showing an interest in pursuing a postgraduate degree to now as I graduate, he has always been positive about my ability and work. His encouragement and indulgence has enabled me to complete my study. I would also like to thank Professor Pascale Fung. This thesis work has been developed on top of her research group’s product, C-ASSERT. Her generosity in sharing research ideas has been a key factor in the completion of this thesis. I am also grateful to her and Professor Brian Mak for sparing their valuable time to join my defense committee. Thanks go to Ms. Shuana Dalton and Miss Joanne Ng for helping me to review my thesis.