Similarity-based interference in sentence comprehension: Literature review and Bayesian meta-analysis

Abstract

We report a comprehensive review of the published reading studies on retrieval interference in reflexive-/reciprocal-antecedent and subject-verb dependencies. We also provide a quantitative random-effects meta-analysis of eyetracking and self-paced reading studies. We show that the empirical evidence is only partly consistent with cue-based retrieval as implemented in the ACT-R-based model of sentence processing by Lewis and Vasishth (2005) (LV05) and that there are important differences between the reviewed dependency types. In non-agreement subject-verb dependencies, there is evidence for inhibitory interference in configurations where the correct dependent fully matches the retrieval cues. This is consistent with the LV05 cue-based retrieval account. By contrast, in subject-verb agreement as well as in reflexive-/reciprocal-antecedent dependencies, no evidence for inhibitory interference is found in configurations with a fully cue-matching subject/antecedent. In configurations with only a partially cue-matching subject or antecedent, the meta-analysis reveals facilitatory interference in subject-verb agreement and inhibitory interference in reflexives/reciprocals. The former is consistent with the LV05 account, but the latter is not. Moreover, the meta-analysis reveals that (i) interference type (proactive versus retroactive) leads to different effects in the reviewed dependency types and (ii) the prominence of the distractor strongly influences the interference effect. In sum, the meta-analysis suggests that the LV05 needs important modifications to account for the unexplained interference patterns and the differences between the dependency types. More generally, the meta-analysis provides a quantitative empirical basis for comparing the predictions of competing accounts of retrieval processes in sentence comprehension.