Seminarium NLP 2008-2009 @ ZIL IPI PAN - Finite-state cascade system CASSYS and its application to rule-based proper name extraction <meta http-equiv="content-type" content="text/html; charset=utf-8" /> <link rel="OWNER" href="mailto:adamp_at@at_ipipan.waw.pl" /> </head> <body> <pre> AUTOR: Nathalie Friburger AFILIACJA: Université François-Rabelais de Tours TYTUŁ: Finite-state cascade system CASSYS and its application to rule-based proper name extraction STRESZCZENIE: Natural Language Processing uses statistical and/or rule based techniques to cut texts in syntactic groups (chunking), to extract information or for other purposes. Rules are a good way to describe linguistic phenomena: Finite State Transducer (FST) formalism is particularly efficient to write rules describing patterns in texts. An FST is a special automaton in which the tags in input can be associates to output; those outputs can replace the pattern recognized or can be merge with it. The purpose of this presentation is to present our Finite State Cascade System: CasSys, whose the very first version was developed in 2002 to extract named entities in written texts. In recent years, Cassys has evolved to be sufficiently general to manipulate texts (even xml texts) in order to do chunking or to transform texts with the use of transducers. CasSys permits to pass FSTs on texts in a special way, called "cascade of transducers", improving the possibilities offered by an FST alone. First of all, we present principles of a cascade of transducers and the types of manipulations we can do on texts. Secondly the options offered by CasSys and the results produced and finally, the example of named entities extraction. </pre> </body> </html>