Finite-State Transducer Cascade to Extract Proper Names in Texts

This article describes a finite-state cascade for the extraction of person names in texts in French. We extract these proper names in order to categorize and to cluster texts with them. After a finite-state pre-processing (division of the text in sentences, tagging with dictionaries, etc.), a series of finite-state transducers is applied one after the other to the text and locates left and right contexts that indicates the presence of a person name. An evaluation of the results of this extraction is presented. 1 Motivation Finite-State Automata and particularly transducers are more and more used in natural languages processing [13]. In this article, we suggest the use of a finitestate transducer cascade to locate proper names in journalistic texts. In fact, we study names because of their numerous occurrences in newspapers (about 10 % of a newspaper) [3]. Proper names have already been studied in numerous works, from the Frump system [5] to the American programs Tipster1 and MUC2 ....

• Type: Conference
• Year: 2001
• Where: WIA
• Authors: Nathalie Friburger, Denis Maurel
• Keywords: Artificial Intelligence, Person Name, Proper Names, WIA 2001, finite-state Cascades