Words from the wizarding world: Fictional words, context, and domain knowledge.

Abstract

The influence of domain knowledge on reading behaviour has received limited investigation compared to the influence of, for example, context and/or word frequency. The current study tested participants with and without domain knowledge of the Harry Potter (HP) universe.Fans and Non-Fans read sentences containing HP, high-frequency (HF), or low-frequency target-words. Targets were presented in contexts which were supportive or unsupportive within a 2 (Group: Fans, Non-Fans) × 3 (Context: HP, HF, LF) × 3 (Word Type: HP, HF, LF) mixed design. Thirty-two Fans and 22 Non-Fans read 72 two-sentence experimental items whilst eye-movement behaviour was recorded: initial sentences established context; second sentences contained target-words. Fans processed HP words faster than Non-Fans. No group difference was observed on HF or LF processing durations suggesting equivalent reading capabilities. In HP contexts, HP and LF targets were processed equivalently. Processing of HF and LF words was facilitated by their supportive context as expected. Non-Fans made more regressions into the target region in HP contexts and regressed more into HP targets than other targets; Fans regressed into target word regions equivalently across all context and word types. Results suggest that domain knowledge influences early but not immediate lexical access, whilst the processing effect of novelty was seen in regressive eye movements.These results are more supportive of modular accounts of linguistic processing and serial models of eye movement control. Words without grounding in reality, or true embodiment, were integrated into Fans' mental lexicons.

Are Emoji Processed Like Words? An Eye‐Tracking Study

“…We also tested the effect of ambiguity introduced in the second dataset through an additional independent variable called “experiment.” Random intercepts for all the variables were included in all the models. Before creating the models, the values of the three dependent measures were first visualized using Q‐Q plots, and then log‐transformed to diminish the positive skew and improve model fit (Baayen & Milin, 2010; Ingram & Hand, 2020). Maximum likelihood was used to fit the models.…”