Choosing anaphoric expressions: do people take into account likelihood of reference?

Research has shown that following a sentence fragment such as John impressed Mary because…, people are most likely to refer to John, whereas following John admired Mary because…, Mary is the preferred referent. Two written completion experiments investigated whether such semantic biases affect the choice of anaphor (pronouns vs. names). Experiment 1 investigated biases due to verb semantics, and Experiment 2 contrasted biases due to different connectives (because vs. so). Frequency-based accounts such as proposed by Arnold (2001) and functional linguists (e.g., [Givón, 1988] and [Givón, 1989]) suggest that the likelihood of reference to a particular discourse entity should affect the choice of anaphor: more pronouns (relative to names) for the bias-consistent entity than the bias-inconsistent entity. Although the semantics of the verb and connective had strong effects on the choice of referent, neither experiment showed any effect of semantic bias on the choice of anaphoric form. In contrast, structural factors did affect anaphoric choice.