Cue-based retrieval models have largely been adopted as a description of how linguistic content is retrieved from memory. Under this framework, a retrieval cue is projected at the site of a dependency and matched with its target using a parallel matching procedure (e.g., Van Dyke and Lewis, 2003). Although this is a highly efficient mechanism, retrieval difficulties occur when there are multiple items stored in memory that serve as potential matches for the retrieval cue(s), which is known as similarity-based interference (SBI). Several studies have demonstrated that a wide variety of linguistic information can generate SBI effects, but the theory of what serves as a retrieval cue is still relatively unknown (Van Dyke and Johns, 2012). Moreover, recent empirical evidence has proposed the similarity-based interference can arise from another source: the encoding mechanism (e.g., Villata et al., 2018).
Three hypotheses are addressed regarding three potential retrieval mechanisms: (1) a retrieval mechanism that only relies on cues relevant to the dependency being resolved, (2) one that is sensitive to all of the features overlapping between a target and distractor(s), or (3) a mechanism that is primarily sensitive to relevant features but produces additive interference effects for irrelevant features. Moreover, a fourth hypothesis investigates if similarity-based interference also arises from the encoding mechanism.
In an attempt to disentangle whether sentence processing disruptions occur as a result of retrieval mechanism (1) + encoding interference or due to one of the other mechanisms, 7 self-paced reading experiments were conducted on Brazilian Portuguese. In all of the studies, number was a relevant feature for the resolution of the grammatical dependency (subject-verb dependency in relative clauses or wh-remnant-correlate pairing in sluices) and gender features varied in their relevance. The rationale behind using these dependencies and features was to test whether syntactically relevant features produced stronger interference effects than irrelevant features and to propose why these results differed. Any findings that showed that irrelevant feature (gender) matches caused reading time slowdowns or decreased comprehension question accuracy before the retrieval site were interpreted as encoding interference.
Although results vary across studies, the findings in this thesis provide the most support for a combination of retrieval (mechanism 1) and encoding interference. Although the other two retrieval mechanisms cannot be completely ruled out at this time, the evidence that gender produces earlier and weaker effects reminiscent of encoding interference and that number produced interference reflective of retrieval interference are novel.
