Naturalness of an Utterance Based on the Automatically Retrieved Commonsense

Rafal Rzepka, Yali Ge, Kenji Araki

Abstract: In this research we investigated user's behavior while facing a system coping with common knowledge about keywords and compared it with not only classic word-spotting method but also with random text-mining. We show how even a simple implementation of our idea can enrich the conversation and increase the naturalness of computer's utterances. Our results show that even very commonsensical utterances are more natural than classic approaches and also methods we developed to make a conversation more interesting. For arousing opinion exchange during the session, we will also briefly introduce our idea of combining latest NLP achievements into one holistic system where the main engine we want to base on commonsense processing and affective computing.