Logical necessity and transitivity of causal relations in stories

Abstract

! central concern of this study is the problem of identifying inferences that connect units in narrative discourse. A recursive transition network model is described that analyzes and represents a story as a causal network of categorized clauses and labeled relations. Procedural criteria used to identify inferences and the assumptions that inferences operate over distances in the text and are transitive, resulting in a network representation, are then tested. In each of two experiments, naive judges rated the strength of relations between pairs of clauses sampled from four different stories. In one experiment, the causal distance between clauses was varied and compared to temporal and reference factors; in the other, the judgment criteria and kinds of relations were varied. The data show that the causal relations are operative and transitive, in that the strength of the relations declined linearly as a function of causal distance in the network representation, independent of temporal and reference distance. Further, the counterfactual criterion of necessity, in the circumstances of the story, distinguished causal from noncausal relations and yielded high strength ratings for physical, motivational, psychological, and enabling relations, in that order. In contrast, a causal criterion differentiated the first three causal relations from enabling relations. The concept of causality employed by naive judges relies upon both necessity and sufficiency criteria. These criteria govern the inferences that bridge clauses containing agent actions and patient state changes in space and time.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 392 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.