Strong semantic biases make demonstrative pronouns act like personal pronouns

Abstract

Anaphoric demonstrative pronouns (d-pronouns), which are found in some languages in addition to the more common personal pronouns (p-pronouns), provide an interesting test case for theories of pronoun resolution. With regard to structural factors (e.g., syntactic function, linear position, topichood), d-pronouns have been found to stand in complementary relation to p-pronouns. Whereas the latter prefer structurally prominent antecedents, the former prefer structurally non-prominent antecedents. In contrast to structural factors, semantic factors (e.g., implicit causality and consequentiality) have been intensively explored for p-pronouns but not for d-pronouns. In particular, it is an open question whether semantic biases can override structural biases in the case of d-pronouns in the same way as it has been shown for p-pronouns. We have addressed this question in three sentence completion experiments and one acceptability experiment that investigated German p- and d-pronouns. Semantic bias was manipulated by including causal and consequential discourse markers in the interpretation experiments. The production experiments used questions to elicit continuations expressing causes and consequences. The results show that the semantic factor of coherence influences p- and d-pronouns in similar ways. Based on our experimental results, we show that the Bayesian Theory of Pronoun Resolution (Kehler et al. 2008), which captures the relationship between pronoun interpretation and pronoun production in a quantitative way, provides a good fit of observed and predicted values for both p- and d-pronouns.