KBGen - Text Generation from Knowledge Bases as a New Shared Task

In this paper we propose a new shared task, KBGen, where the aim is to produce coherent descriptions of concepts and relationships in a frame-based knowledge base (KB). We propose to use the AURA knowledge base for the shared task which contains information about biological entities and processes. We describe how the AURA KB provides an application context for NLG and illustrate how this application context generalizes to other biology KBs. We argue that the easy availability of input data and a research community -- both domain experts and knowledge representation experts -- which actively uses these knowledge bases, along with regular evaluation experiments, creates an ideal scenario for a shared task.