LIHLITH: Learning to Interact with Humans by Lifelong Interaction with Humans

Eneko Agirre , Arantxa Otegi , Camille Pradel , Sophie Rosset (1) , Anselmo Peñas , Mark Cieliebak

1 LIMSI - Laboratoire d'Informatique pour la Mécanique et les Sciences de l'Ingénieur

The LIHLITH project will research, innovate and validate a new lifelong learning framework for the interaction of humans and machines on specific domains with the aim of improving the quality of existing dialogue systems and lowering the cost of deployment in new domains. LILITH will develop dialogue systems that learn autonomously from their interactions with the users, and retain this new knowledge in order to answer more accurately in future interactions. The key insight is that the dialogue systems will be designed to get feedback from the user. Based on this feedback, the system will keep improving after deployment all modules down in the pipeline. LIHLITH project will also develop and deliver evaluation protocols and benchmarks to allow public comparison and reproducibility.

Mots clés: Dialogue systems; lifelong learning; question answering; knowledge induction

Domaines: Informatique [cs]; Informatique et langage [cs.CL]

Eneko Agirre, Arantxa Otegi, Camille Pradel, Sophie Rosset, Anselmo Peñas, et al.. LIHLITH: Learning to Interact with Humans by Lifelong Interaction with Humans. Spanish Conference for Natural Language Processing (Sociedad Española para el Procesamiento del Lenguaje Natural), Sep 2019, Bilbao, Spain. ⟨hal-02358023⟩