We describe an applied methodology to build fuzzy models of geographical expressions, which are meant to be used for natural language generation purposes. Our approach encompasses a language grounding task within the development of an actual data-to-text system for the generation of textual descriptions of live weather data. For this, we gathered data from meteorologists through a survey and built consistent fuzzy models that aggregate the interpersonal variations found among the experts. A subset of the models was utilized in an illustrative use case, where we generated linguistic descriptions of weather maps for specific geographical expressions. These were used in a task-based evaluation to determine how well potential readers are able to identify the geographical expressions grounded on the models.
Keywords: natural language generation, linguistic descriptions of data, data-to-text, geo-referenced data, language grounding, fuzzy sets