The importance of narrative and other lessons from an evaluation of an NLG system that summarises clinical data

Abstract

The BABYTALK BT-45 system generates textual summaries of clinical data about babies in a neonatal intensive care unit. A recent task-based evaluation of the system suggested that these summaries are useful, but not as effective as they could be. In this paper we present a qualitative analysis of problems that the evaluation highlighted in BT-45 texts. Many of these problems are due to the fact that BT-45 does not generate good narrative texts; this is a topic which has not previously received much attention from the NLG research community, but seems to be quite important for creating good data-to-text systems.