Statistics and spice and everything nice: what are discourse biases made of?

Pronoun comprehension is guided by contextual biases, for example in Ana admired Liz because she … , people often assume that she refers to the assumed cause (Liz). We test whether these biases could be influenced by exposure statistics, namely calculations about the most frequent type of continuation. Using an adaptation paradigm, we manipulate whether participants are exposed to repeated unambiguous pronouns that always refer to either the subject or the object, always in a context with strong implicit causality biases. The critical question is whether participants adapt to the exposure items, shifting their interpretation of ambiguous pronouns in the test items. We ask both whether this affects pronoun comprehension in similar implicit causality contexts (it does), and whether the adaptation generalises to other contexts (it does not). These findings show that discourse biases are influenced by statistical learning, and that statistics about causal contexts may be categorised separately.