Section: New Results

Towards a French FrameNet

Participants : Marie-Hélène Candito, Marianne Djemaa, Benoît Sagot.

The ASFALDA project (https://sites.google.com/site/anrasfalda/ ) is an ANR project coordinated by Marie Candito. 5 partners collaborate on the project, on top of Alpage : the Laboratoire d'Informatique Fondamentale de Marseille(LIF), the Laboratoire de Linguistique Formelle (LLF), the MELODI team (IRIT - Toulouse) and the CEA-List. It is a three-year project which started in October 2012, with the objective of building semantic resources (generalizations over predicates and over the semantic arguments of predicates) and a corresponding semantic analyzer for French. We chose to build on the work resulting from the FrameNet project [57] , (https://framenet.icsi.berkeley.edu/ ) which provides a structured set of prototypical situations, called frames, along with a semantic characterization of the participants of these situations (called frame elements). The resulting resources will consist of :

1. a French lexicon in which lexical units are associated to FrameNet frames,

2. a semantic annotation layer added on top of existing syntactic French treebanks

3. and a frame-based semantic analyzer, focused on joint models for syntactic and semantic analysis.

In 2014, we first finished the work on the lexicon, which was started in 2013 [19] . The step 2 (semantic annotations on top of syntactic representations) is ongoing :

• We wrote the annotation guide. In particular Marianne Djemaa focused on how to annotate phenomenon known to exhibit syntax/semantic divergences [42] .

• We designed the annotation workflow and built an automatic pre-annotator, which proposes candidate semantic annotations that must be disambiguated manually.

• We started in july 2014 to manage six annotators, who were hired to perform the manual annotation phase.