Grounding, Justification, Adaptation: Towards Machines That Mean What They Say

Meaningful language use rests on the grounding of the language in the nonlinguistic world and in the practices of language users. This grounding is built up and maintained in interaction, through Conversational Grounding, which is the interactive process with which interlocutors build mutual understanding; Justification, the ability to explain and provide reasons for one’s language use; and Adaptation, the ability to accept corrections and adapt future language use accordingly. We outline a model of grounded semantics that combines perceptual knowledge (how to visually identify potential referents of terms; realised as classifiers taking visual information as input) and taxonomic knowledge (covering lexical relations such as hyponymy and hypernymy), and we sketch a proof-of-concept implementation of a dialogue system that realises the interactional skills that ground this knowledge.