Visual Grouping and Pragmatic Constraints in the Generation of Quantified Descriptions
- Gordon Briggs, U.S. Naval Research Laboratory, Washington, District of Columbia, United States
- Hillary Harner, US Naval Research Laboratory, Washington, District of Columbia, United States
- Sangeet Khemlani, Navy Center for Applied Research in Artificial Intelligence, Naval Research Laboratory, WASHINGTON, District of Columbia, United States
AbstractStudies suggest that people use the least possible effort to generate natural language descriptions of sets of objects. This means that they base descriptions on what is perceptually available to them. For instance, people can subitize, i.e., rapidly assess the exact quantity of small numbers of objects, so when the quantity of objects in the visual scene is beneath this threshold, they give numeric descriptions; when the quantity is above this threshold, they generate non-numeric descriptions. However, no research examines how people describe visual scenes of items in groups. As such, it is unclear how people will form descriptions of scenes that contain a large total number of items in groups. We report on a novel experiment designed to investigate how people produce quantified descriptions of scenes composed of salient visual groups. The results corroborate the least effort hypothesis, and suggest that people's incremental perception of quantity drives their descriptions.