Error-toleran t nite-state lo okup for trademark searc h

Error-toleran t lo okup of w ords in large v o cabularies has man y p oten tial uses, b oth within and b ey ond natural language pro cess- ing (NLP). This w ork 3 describ es a generic library for nite-state-based lexical lo okup, originally designed for NLP-related applications, that can b e adapted to application-sp ecic error metrics. W e sho w ho w this to ol can b e used for searc hing existing trademarks in a database, using or- thographic and phonetic similarit y . W e sk etc h a protot ypical implemen- tation of a trademark searc h engine and sho w results of a preliminary ev aluation of this system. 1