Section: New Results

Information extraction from corpora parsed with FRMG

Participants : Yayoi Nakamura-Delloye, Rosa Stern, Éric Villemonte de La Clergerie, Benoît Sagot.

Following previous experiments, in particular in the context of the FUI-funded project Scribo that ended in 2010 (NAKAMURADELLOYE:2010:HAL-00511541:1,NAKAMURADELLOYE:2010:HAL-00511481:1), work has been achieved for extracting information from corpora parsed with FRMG.

In the context of the EDyLex project, we have proposed two pattern-based named entity extraction methods for ontology enrichment [36] , [35] . The proposed methods are characterized by the use of entity relation patterns obtained by our unsupervised extraction method. These patterns correspond to syntactic paths that connect two named entities in dependency trees produced by FRMG. This work aims to take advantage of parsing benefits and also offers solutions for parsing disadvantage. The proposed methods are characterized by the use of entity relation patterns obtained by our unsupervised extraction method. These patterns correspond to syntactic paths that connect two named entities in dependency trees. This work aims to take advantage of parsing benefits and also offers solutions for parsing disadvantage.

We also developed a mechanism for integrating the results into an domain ontology, namely the ontology under deployment at the Agence France-Presse [37] .