FASTUS: A Cascaded Finite-State Transducer for Extracting Information from Natural-Language Text

• Published in: FINITE-STATE LANGUAGE PROCESSING, Roche, Emmanuel, & Schabes, Yves [Eds], Cambridge: Massachusetts Instit Technology Press, 1997, pp 383-406
• Main Authors: Hobbs, Jerry R, Appelt, Douglas, Bear, John, Israel, David, Kameyama, Megumi, Stickel, Mark, Tyson, Mabry
• Format: Book Chapter
• Language: English
• Published: 01.01.1997
• ISBN: 0262181827; 9780262181822

In light of the slow processing time of a text understanding system called TACITUS, developed by Hobbs et al, an information extraction system named FASTUS has been designed for implementation in CommonLisp on a Sun workstation. FASTUS uses a cascaded finite-state approach, breaking the task of information extraction down into five stages; earlier domain-independent stages apply linguistic knowledge to recognize syntactic elements & provide input to later stages that search for domain-dependent patterns. Stage 1 identifies complex words & names, dates, times, & locations; stage 2 identifies noun & verb groups & functional word classes. Complex noun & verb groups are processed at stage 3, where modalities are associated with verb groups. Stage 4, corresponding to the basic clause level, scans input phrases for common domain events; stage 5 merges structures across sentences in a text. The history of the FASTUS system is sketched, & its advantages are itemized: conceptual simplicity, effective performance, high run-time speed, & fast development time. 21 References. J. Hitchcock