Common Sense Knowledge graph generation for information-gap requests in dialogue systems

In this study, we propose a methodology to represent Common Sense Knowledge and its application for conversational purposes. Common Sense is defined as “[. . . ] a tremendous amount and variety of knowledge of default assumptions about the world, which is shared by (possibly a group of) people and seems so fundamental and obvious that it usually does not explicitly appear in people’s communications” [Zang et al., 2013, p. 690]. Three aspects of this definition appear to be important for our purposes: i) sharedness, ii) implicitness, iii) communication. These aspects are important in this work as they will guide our conversational agent capable to inferring implicit shared information to process and generate appropriate clarification requests in an incomplete dialogue state, i.e., when relevant arguments of an utterance (mandatory slots) are not filled. This way, the system will be able to pose an appropriate question, such as “Who directed this movie?”, by conversely avoiding obvious questions like “Did someone directed this movie?”