作者: Kees van Deemter , Margaret Mitchell , Ehud Reiter
DOI:
关键词:
摘要: This paper discusses the basic structures necessary for generation of reference to objects in a visual scene. We construct study designed elicit naturalistic referring expressions relatively complex objects, and find aspects that have not been accounted work on Referring Expression Generation (REG). includes object parts, size comparisons without crisp measurements, use analogies. By drawing research cognitive science, neurophysiology, psycholinguistics, we begin developing input structure background knowledge an algorithm capable generating kinds observe.