What's in a Name: Current Methods, Applications, and Evaluation in Multilingual Name Search and Matching

Names of people, places, and organizations have unique linguistic properties, and they typically require special treatment in automatic processes. Appropriate processing of names is essential to achieve highquality information extraction, speech recognition, machine translation, and information management, yet most HLT applications provide limited specialized processing of names. Variation in the forms of names can make it difficult to retrieve names from data sources, to perform co-reference resolution across documents, or to associate instances of names with their representations in gazetteers and lexicons. Name matching has become critical in government contexts for checking watchlists and maintaining tax, health, and Social Security records. In commercial contexts, name matching is essential in credit, insurance, and legal applications. This tutorial will focus on personal names, with special attention given to Arabic names, though it will be clear that much of the material applies to other languages and to names of places and organizations. Case studies will be used to illustrate problems and approaches to solutions. Arabic names illustrate many of the issues encountered in multilingual name matching, among which are complex name structures and spelling variation due to morphophonemic alternation and competing transliteration conventions.