Memory retrieval and implicit prosody during reading: anaphora resolution by L1 and L2 speakers of English

Abstract

This thesis focuses on L1 (native English speakers) and L2 (Vietnamese speakers of English) processing and resolution of reflexive-antecedent dependencies. The experimental design crossed the three factors: implicit prosody in the form of text presentation formats (word-by-word, phrase-segment, and whole-sentence); syntactic structure (simple, complex); and grammaticality (target-match, target-mismatch). An example of the stimuli is shown in (I). (I) The actress that {Mary/John} interviewed at the awards ceremony {about two years ago/held outside the theater} described {herself/himself} as an extreme workaholic. Participants were asked to read the experimental sentences presented on the computer screen in one of the three reading formats: word-by-word (rapid serial visual presentation), phrase-segment (self-paced), and whole-sentence (self-paced), then complete the two follow-up tasks: grammaticality judgment and comprehension probes. Participants’ English reading proficiency was measured via a reading test of which the scores significantly and positively correlated with the results of the self-rated proficiency questionnaire. During sentence processing, participants are expected to employ a retrieval probe that is created by the combination of retrieval cues in order to integrate previously accessed input with current linguistic representations (Martin, 2018). Retrieval of cues during processing might be obstructed by interfering items that share linguistic features with those of the target (e.g., +FEM : actress, Mary, herself as in (I)). Cue overload occurs when multiple items have features that match retrieval cues, resulting in longer retrieval latency and higher probability of misretrieval (Engelmann, Lewis, & Vasishth, 2019; Lewis & Vasishth, 2005; Parker, 2019; Parker, Shvartsman, & Van Dyke, 2017). When a comprehender fails to deploy retrieval cues due to heavy cognitive load, the comprehender is predicted to resort to a processing strategy that is “good enough” to maintain general comprehension (Ferreira, Bailey, & Ferraro, 2002; Ferreira, 2003; Ferreira & Patson, 2007). Sacrifice of detailed syntactic analysis in exchange for heuristic computation of input is suggested to result from the language system’s preference for a cognitive equilibrium state that prioritizes early arrival at comprehension (Karimi & Ferreira, 2016). Though prosodic parsing that is in line with the syntactic structure of a sentence is expected to reduce cognitive load, thus increasing probability of accurate retrieval (Bader, 1998; Fodor, 1998, 2002), no prosodic information seems to provide any further help with syntactic processing once the comprehender opts for a simple, good enough processing approach. Language proficiency and reading fluency are predicted to mediate between utility of prosodic information and deployment of cues during retrieval. Results of the study showed that there was evidence of trade-off effects between general comprehension and grammaticality judgment. Interference effects across experimental items were experienced by both groups of participants, particularly among L2. The effects were found to be most robust when there was an interaction between grammaticality and syntactic complexity. Though both L1 and L2 participants benefited from the sentence format, the two groups differed in the effects of the other two reading paradigms. L1 speakers were most disrupted by the word-by-word format, and L2 speakers by the phrase-segment format. I suggest that the differences between the two groups be attributed to differences between L1 and L2 processing strategies and the development of reading fluency. Overall, the findings of the study contribute to the processing model based on the framework of the Good-enough and Online Cognitive Equilibrium hypotheses. This model takes into account the combined effects of cognitive load, memory access, and implicit prosody (Pratt & Fernandez, 2016).