Referring under Uncertainty
Nicolás Marı́n, Gustavo Rivas-Gervilla, Daniel Sánchez · 2019
In this paper, we study the referring expression generation problem (REG) when the available information about the properties of objects is uncertain, in the sense that we are not sure about the actual properties an object has. We formalize the problem by extending the conventional REG framework through the use of possibility distributions, represented by fuzzy sets. We show the potential benefits of this proposal in the assessment of the referential success for referring expressions. This approach opens a new line of research in the application of fuzzy sets to the REG problem, complementary to those approaches that use fuzzy sets as a suitable bridge between language and raw data.