Visual and linguistic predictors for the definiteness of referring expressions

This study explores the impact of visual context on the conceptual salience of a discourse entity, using descriptions of how to find specific targets in cartoon scenes. Significant positive correlation is observed between larger and more salient objects and definite expressions, whereas more cluttered images are positively related to indefinite expressions. Incorporating these findings with other linguistic factors, we build a mixed-effects logistic regression model for predicting referring forms. The model reaches 62% accuracy. This study helps us to understand better how physical context, like an image, determines the linguistic properties of a discourse