Séminaire au DIC: «Meaning modeling for virtual assistants»

Voici l’annonce du 7e séminaire au DIC pour la session HIVER 2018  par Joumana GHOSN

Titre : Meaning modeling for virtual assistants

Résumé

Several virtual assistant systems are currently based on the flat intent/slot semantic meaning representation whose expressive power is highly limited. More powerful alternatives exist, including graph-based meaning representations which can capture complex relations between concepts. We will discuss those two representation paradigms. We will then review sequence to sequence approaches used to model graph-based representations and describe some of the challenges encountered when training those models, including the linearization of the graph, the handling of large output vocabularies, and the presence of limited amounts of labeled data.

Biographie

Joumana Ghosn obtained a PhD in machine learning from University of Montreal. She did a post-doc at the McGill University Health Center focusing on the analysis of the effect of anti-inflammatory drugs on cancer. After a few years working on NLP at Idilia, a Montreal-based startup, she joined Nuance where she currently directs a team of researchers working on clinical language understanding and on natural language understanding for mobile devices.