Towards Empirical Evaluation of Affective Tactical NLG

One major aim of research in affective natural language generation is to be able to use language intelligently to induce effects on the emotions of the reader/ hearer. Although varying the content of generated language ("strategic" choices) might be expected to change the effect on emotions, it is not obvious that varying the form of the language ("tactical" choices) can do this. This chapter discusses two experiments carried out to show emotional effects of tactical variations. With the first experiment we were not able to show clear, statistically signicant differences between the effects of the different texts in readers. We discuss a number of possible reasons and, building on our discoveries, we present a second experiment which does demonstrate such effects. This represents an important step towards the empirical evaluation of affective NLG
systems.