Using conceptual spaces to model domain knowledge in data-to-text systems

Abstract

This position paper introduces the utility of the conceptual spaces theory to conceptualise the acquired knowledge in data-totext systems. A use case of the proposed method is presented for text generation systems dealing with sensor data. Modelling information in a conceptual space exploits a spatial representation of domain knowledge in order to perceive unexpected observations. This ongoing work aims to apply conceptual spaces in NLG for grounding numeric information into the symbolic representation and confronting the important step of acquiring adequate knowledge in data-to-text systems.