Section: New Results

Development of Verb$\ni$net

Participants : Laurence Danlos, Quentin Pradet, Lucie Barque.

VerbNet is an English lexical resources for verbs, which is internationally known and widely used in numerous NLP applications [74] . Verb$\ni$net is a French adaptation of this resource. It is semi-automatically developed thanks to the use of two French existing resources created in the 70's: LG, Lexique-Grammaire developed at LADL under the supervision of Maurice Gross, and LVF, Lexique des verbs du français by Dubois and Dubois-Charlier. The idea is to map English classes, which gather verbs with a common syntactic and semantic behavior, into classes of LG and LVF, then to manually adapt the syntactic frames according to French grammar while keeping the thematic roles and the semantic information, [84] , [68] [14] . A first version of this work has been achieved in June 2015 in collaboration with Takuya Nakamura (Institut Gaspard Monge) [33] .

The next step was to verify the coherence of the resource. A particular focus has been to check the way alternations have been encoded and to document this encoding. A journal article extracted from this documentation has been submitted to the TAL journal and Verb$\ni$net will be released after getting the feedback of the editorial board.