Time to Throw in the Towel? No Evidence for Automatic Conceptual Metaphor Access in Idiom Processing

Abstract

The goal of the current research was to determine if conceptual metaphors are activated when people read idioms within a text. Participants read passages that included idioms that were consistent (blow your top) or inconsistent (bite his head off) with an underlying conceptual metaphor (ANGER IS HEATED FLUID IN A CONTAINER) followed by target words that were related (heat) or unrelated (lead) to the conceptual metaphor. Reading time (Experiment 1) or lexical decision time (Experiment 2) for the target words were measured. We found no evidence supporting conceptual metaphor activation. Target word reading times were unaffected by whether they were related or unrelated to underlying conceptual metaphors. Lexical decision times were facilitated for related target words in both the consistent and inconsistent idiom conditions. We suggest that the conceptual (target) domain, not a specific underlying conceptual metaphor, facilitates processing of related target words.

Appendices

Appendix A

Appendix B

We used mixed-effects models to replicate the effects of the F1 and F2 for each Experiment, but controlling for subjects and items in the same analysis (Baayen et al. 2008), using the lme4 version 1.1–15 package in R (Bates et al. 2015). The analyses for Experiments 1 and 2 were conducted through a forward-fitting processes and the random effect for each model crossed subjects and paragraphs (i.e., items: which were first nested in the Idiom Phrase Consistency condition for both studies). The Idiom Phrase Consistency and Target Word Type conditions were interacted and were allowed to vary as a function of the subject and Target Word Type was allowed to vary as a function of the Passage (nested in Idiom Phrase Consistency). The Delay condition in Experiment 2 was not treated a random effect as it was a between-subject condition. The only exception to the random effects was the fixation count variable, which was analyzed with Poisson mixed effects model with a simpler random structure because of convergence failures. Modeling of the fixed-effects tested in Experiments 1 and 2 followed the same procedures and tested for significant model improvement (via a deviance test) when fixed terms were added. The Null model contained only the random effects and fixed intercept. Model 1 added the main effects (Experiment 1: Idiom Phrase Consistency + Target Word Type; Experiment 2: Idiom Phrase Consistency + Target Word Type + Delay). Model 2 and beyond added in interactions.

Table 4 shows the results of Experiment 1 modeling for the four dependent variables. In the case all the dependent variables, there was no significant improvement (via the deviance test) of Model 1 (main effects only) from the Null model (no fixed effects). In addition, there was no improvement from Model 2 (interaction) to the Model 1 (main effect only). This suggests that the Idiom Phrase Consistency and Target Word Type factors were unrelated to the eye-tracking measures.

Table 4 Summary of comparisons between mixed effect models asking whether the additional predictors improved model fit over the previous model for Experiment 1 dependent variables

Table 5 shows the results of Experiment 2 modeling for the dependent variable of lexical decision time. There was a significant improvement (via the deviance test) of Model 1 (main effects only) from the Null model (no fixed effects). This suggested there was a main effect for either Idiom Phrase Consistency, Target Word Type, or Delay factor. However, subsequent models testing for interactions between factors (Models 2–5) showed no significant improvement over Model 1. To determine which main effect was significant, Model 1was examined in Table 6 in detail.

Table 5 Summary of comparisons between mixed effect models asking whether the additional predictors improved model fit over the previous model for Experiment 2

Table 6 shows the individual parameters from Model 1 from Experiment 2. The degree of freedom for each parameter were estimated by Satterthwaite approximations (Singmann et al. 2016) to yield p values. The main effects (Idiom Phrase Consistency, Target Word Type, Delay) were effects-coded (−0.5, 0.5). The intercept thus reflects the grand mean of the experiment. The results show the intercept was significant (meaning the lexical decision time was different from zero) and there was a main effect for Target Word Type and Delay, which basically match the results of the ANOVAs reported.

Table 6 Model 1 from Experiment 2 (best fit model) mixed effect model results