Section: New Results

Dynamic extension of a French lexical resources based a text stream

Participants : Damien Nouvel, Benoît Sagot, Rosa Stern, Virginie Mouilleron, Marion Baranes.

Lexical incompleteness is a recurring problem when dealing with natural language and its variability. It seems indeed necessary today to regularly validate and extend lexica used by tools processing large amounts of textual data. This is even more true when processing real-time text flows. In this context, we have introduced two series of techniques for addressing words unknown to lexical resources, and applied them to French within the context of the EDyLex ANR project:

• Extending a morphological lexicon We have studied neology (from a theoretic and corpus-based point of view) and developed modules for detecting neologisms in AFP news wires in real time and inferring information about them (lemma, category, inflectional class) [24] . We have shown that we are able, using among others modules for analyzing derived and compound neologisms, to generate lexical entries candidates in real time and with a good precision, to be added in the Lefff lexicon.

• Extending an entity database We have also extended our previous work on named entities detection and linking in order to be able to extract new named entities from AFP news wires and create candidate entries for the Aleda entity database.