Grounded Language Learning Fast and Slow

Felix Hill, Olivier Tieleman, Tamara von Glehn +3

Recent work has shown that large text-based neural language models, trained with conventional supervised learning objectives, acquire a surprising propensity for few- and one-shot…