Corpus-driven learning of Event Recognition Rules
Roberto Basili, Maria Teresa Pazienza, Michele Vindigni · 2007
In this paper a complex framework for adaptaing IE systems to changing domains and users is described. The proposed methodology is based on the integration of different learning methods over a corpus (i.e. example-driven conceptual clustering, corpusdriven probabilistic learning and terminological reasoning) and on an available general-purpose ontology. First experiments have been carried out on domain-specific corpora (the annotated portion of the PennTree Bank and the Reuters TREVI collection 2 ) and used Wordnet [16] as the reference ontology. However, the methodology is independent from the specific domain as well as from the adopted ontology. Early evaluation is promising and first results will be presented and discussed.