Parallel Natural Language Parsing: From Analysis to Speedup Parallel Natural Language Parsing: From Analysis to Speedup

Preface The research presented in this thesis done in the context of the " Parallel Natural Language Interfaces " subproject of the IMPACT project. The IMPACT project was headed by the ING bank and founded by the organization for High Performance Computing and Networking (HPCN), itself partly funded by NWO. The goal of this subproject was to improve the performance of natural language interfaces. Considering the context of the research, there was a strong focus on achieving speedup by means of parallel processing. Although the research goals aim at improving the performance of natural language interfaces in general, the research in this thesis focuses on parallel parsing. The reason for this was that the parsing component of the Deltra system, which initially was the target platform of our research, was by far the most computationally expensive component. The parsing component is often the most computationally expensive component for other natural language processing applications as well. To broaden the scope of our research, we also investigated a parser for a different grammar (LinGO) at a later stage of the project. Although a dissertation often appears to be an individual achievement, there are a number of people to whom I am indebted. First and foremost, I would like to thank my promotors Henk Sips and Anton Nijholt and added promotor Ruud Sommerhalder for the guidance and support they have given me. I am especially grateful to Ruud Sommerhalder for his involvement with the successful completion of my dissertation. Also, the contents of this thesis would not have been what it is had I not had the right people around me. In this respect, I would like to thank several people. Firstly, I would like to thank Job Honig, for the many discussions often related to natural language parsing, Stephan Oepen, for providing me with the opportunity to get on track with the LinGO grammar, and Ulrich Callmeier for the help and support with the porting effort to make my parser compatible with the LinGO grammar. Secondly, I would like to thank all the people that enabled me to run my experiments on the right machinery: Jan Hol, for allowing me to use the 8 processor UltraSparc of the Faculty of Aerospace Engineering, and Takashi Ninomiya and Makino Takaki of the Department of Information Science at the University of Tokyo, for running experiments on their 64 processor UltraSparc. In computer science research, often …