Inducing Commonsense Knowledge Using Vector Space Embeddings
Induction de connaissances de sens commun à partir de plongements vectoriels
Résumé
My habilitation provides a high-level overview of my contributions on inducing commonsense knowledge using vector space representations, with a focus on :
(1)Learning conceptual space representations (learning entity embeddings and region-based representations of concepts, learning interpretable dimensions)
(2) Modelling relational knowledge (relation induction in word embedding and pre-trained language models, learning of distributional relation vectors)
(3) Deriving high quality vectors from contextualised LMs and applications to few-shot learning.
(4 )Plausible reasoning about ontologies (automated rule base completion, inconsistency handling and belief merging)
Origine : Fichiers produits par l'(les) auteur(s)