Sentence Production in Naturalistic Scenes with Referential Ambiguity

Language production often happens in a visual context, for example when a speaker describes a picture. This raises the question whether visual factors interact with conceptual factors during linguistic encoding. To address this question, we present an eye-tracking experiment that manipulates visual clutter (density of objects in the scene) and animacy in a sentence production task using naturalistic, referentially ambiguous scenes. We found that clutter leads to more fixations on target objects before they are mentioned, contrary to results for visual search, and that this effect is modulated by animacy. We also tested the eye-voice span hypothesis (objects are fixated before they are mentioned), and found that a significantly more complex pattern obtains in naturalistic, referentially ambiguous scenes.