ABSTRACT
Numerous studies have found facilitation for lexical processing in highly constraining contexts. However, less is known about cases in which immediately preceding (local) and broader (global) contextual constraint conflict. In two eye-tracking while reading experiments, local and global context were manipulated independently, creating a critical condition where local context biases towards a word that is incongruent with global context. Global context consisted of a clause introduced by a concessive marker generating broad expectations about upcoming material. Experiment 1 compared high- and low-predictability critical words, whereas Experiment 2 held the critical word constant and manipulated the preceding verb to impose different levels of local constraint. Facilitation from local context was reduced when it was incongruent with global context, supporting models in which information from global and local context is rapidly integrated during early lexical processing over models that would initially prioritise only local or only global context.
Acknowledgements
Author note: Data collection, data processing, and analysis were conducted at the University of California, Los Angeles. These findings were presented as a poster at the 31st Annual CUNY Conference on Human Sentence Processing in Davis, CA, March 2018.
Disclosure statement
No potential conflict of interest was reported by the author(s).
Notes
1 The particular set of expectations depends on the individual or the particular context; for example, the mentioned expectations for December are based on higher latitudes in the Northern hemisphere, but expectations for December weather would differ drastically depending on geographical location.
2 A reviewer suggested presenting a power analysis for the studies, given the failure to fit models with complex random effect structures. Doing so requires estimating the degrees of freedom, which is not trivial in linear mixed effects models as the parameter estimates are calculated from the residual maximum likelihood, and not from observed and expected mean squares used in power analyses for ANOVAs (Pinheiro & Bates, Citation2006). We addressed power in two ways. We first conducted post-hoc power analyses with G*Power 3.1 using estimates based on the F-distribution of first fixation times, assuming the standard significance level (α = 0.05) and a small-to-medium effect size (Cohen’s F = 0.25). We conducted the test for within-factors repeated measures 2 x 2 MANOVA. For Experiment 1 (N = 48), the power was calculated to be 0.80. For Experiment 2 (N = 52), the power was 0.84. As repeated measures designs complicate power calculations (Brysbaert & Stevens, Citation2018; Judd et al., Citation2017), we also estimated power using the method described in Judd et al. (Citation2017). The power for Experiment 1 was calculated at 0.96, whereas the power for Experiment 2 was 0.98. However, observed (post-hoc) power calculations are often perceived as flawed or misleading when reasoning about non-significant results (e.g. Hoenig & Heisey, Citation2001). Finally, the two experiments essentially replicated each other, suggesting that the findings are robust.