This paper investigates the structure of situation models as revealed by the process of anaphor resolution. The situation model for a text combines the meanings of its individual sentences with inferences generated from prior knowledge into an overall understanding of the text. We conceptualize situation models as implemented by collections of traces in episodic long-term memory that are retrieved based on their similarity to cues in working memory. Traces encode both contextual (i.e., spatial and temporal) and semantic information, providing structure to situation models. Four predictions were derived regarding the retrieval of antecedent traces during anaphor resolution. Experiment 1 evaluated the contextual structure of situation models, finding evidence for temporal gradients but not spatial gradients. Experiment 2 evaluated their semantic structure, finding an interaction between (a) the semantic similarity of anaphors and antecedents and (b) the interference caused by non-antecedent referents. These findings have implications for the representation of space and time in situation models, the continuity of working memory and long-term memory, and the conceptualization of situation models.
Bibliographical noteFunding Information:
This research was funded by Grant-in-Aid #21251 from the University of Minnesota to Sashank Varma.
- Anaphor resolution
- Exemplar models
- Situation models