Corroborating behavioral evidence for the interplay of representational richness and semantic control in semantic word processing

This study aimed to replicate and validate concreteness and context effects on semantic word processing. In Experiment 1, we replicated the behavioral findings of Hoffman et al. (Cortex 63,250–266, https://doi.org/10.1016/j.cortex.2014.09.001, 2015) by applying their cueing paradigm with their original stimuli translated into German. We found concreteness and contextual cues to facilitate word processing in a semantic judgment task with 55 healthy adults. The two factors interacted in their effect on reaction times: abstract word processing profited more strongly from a contextual cue, while the concrete words’ processing advantage was reduced but still present. For accuracy, the descriptive pattern of results suggested an interaction, which was, however, not significant. In Experiment 2, we reformulated the contextual cues to avoid repetition of the to-be-processed word. In 83 healthy adults, the same pattern of results emerged, further validating the findings. Our corroborating evidence supports theories integrating representational richness and semantic control mechanisms as complementary mechanisms in semantic word processing.

accuracy and reaction times, all p < .001.We also found a significant Concreteness × Cue interaction for both the percentage accuracy, p < .001, as well as the reaction time, p = .020.
The resolution of the interaction with dependent samples t-tests mirrored the findings of Experiment 1 as well. Abstract and concrete words showed a significant processing advantage after a contextual cue compared to an irrelevant cue, all p ≤ .002. The concreteness effect was again significant within the contextual and irrelevant cue condition for both measures, all p < .001. The processing advantage after a contextual vs. irrelevant cue was significantly larger for abstract than concrete words for accuracy (mean difference = 2.5%, SE = 0.6%), p < .001, as well as reaction time (mean difference = 29 ms, SE = 12 ms), p = .020, which is again in line with our hypotheses.

Analysis
We performed additional analyses to investigate the influence of potentially confounding psycholinguistic variables on the results reported in the main article. We did not include all the psycholinguistic variables as predictors into one model to avoid the problem of multicollinearity with correlated predictors (Pearson correlation coefficients are displayed in Table B1). Instead, we tested a series of models, each including one psycholinguistic variable as an additional continuous fixed-effect covariate into the (G)LMs specified in the main article. Specifically, for Experiments 1 and 2 we tested five models, each including Concreteness and Cue as categorical fixed-effects factors as well as their interaction, and Participants (with a linear model formula of Concreteness and Cue) and Items as randomeffect factors. In addition, model(1) included the probe length (number of letters), model (2) the written word frequency, model(3) the spoken word frequency, model(4) the absolute valence (irrespective of polarity) and model(5) the arousal ratings as covariate. For Experiment 2 we additionally specified model(6) including the probes' association with emotional experience, model(7) including the probe-target similarity and model(8) including the probe-target association strength as covariate. For all the additional continuous fixedeffects covariates the values were mean centered.

Results
Crucially, the inferential pattern of the Cue and Concreteness main and interaction effects on reaction times and accuracy described in the main article did not change in any of the eight models including additional covariates. Specifically, for all models the (G)LME analyses revealed that Cue and Concreteness factors had significant main effects on the reaction time and accuracy data, all p < .001, while the Cue × Concreteness interaction was significant only for reaction time, all p < .001 (Experiment 1), all p ≤ .008 (Experiment 2), but not for accuracy, all p ≥ .574. The consistent pattern of results across all the tested models, each including a psycholinguistic variable as covariate, validated our a-priori matching of psycholinguistic variables for abstract and concrete words.
We additionally report here the β estimates of the covariate' effect for each tested model and effect-specific χ²/t-tests (see Table B2), and we provide an interpretation of such additionally findings.
Model (1). Probe length had a significant effect on reaction times but not accuracy. This might be due to a reduced reading speed, which previous research linked to word length in healthy adults, when words were longer than 5 letters 1 , which applies to 166 of our 200 probe words and might thus have influenced reaction times.

Model(2 and 3).
Neither written nor spoken word frequency had a significant effect as a covariate. Effects of word frequency have previously been shown for lexical decision times 2 . However, the processes involved in synonym judgments might rely on semantic rather than lexical variables to a greater extent 3 , which might explain why written and spoken word frequency did not affect our results.

Model(4,5,6).
The covariates valence and arousal had significant effects on reaction times as well as accuracy. Both emotional variables led to lower reaction times and higher accuracy. Processing facilitation by emotional information is in line with mechanisms of semantic enrichment assumed by the representational substrates hypothesis 4 as well as the affective embodiment account 5 , with the latter approach assuming that such mechanisms take place only for abstract words. For Experiment 2, just like valence and arousal, also emotional experience benefitted semantic processing performance and showed a significant main effect on reaction times and accuracy. (7 and 8). In Experiment 1, probe-target similarity and association strength as covariates had significant effects on reaction times as well as accuracy. Both variables led to lower reaction times and higher accuracy. Higher values for both variables seem to have facilitated the correct identification of the synonym.  Note. * p < .05, ** p < .01, *** p < .001.

Analysis
We entered reaction times in an LME analysis as specified in the main article with one modification: We included the mean centered imageability ratings as a continuous predictor instead of the factorial dichotomous factor concreteness. The model thus included Cue (relevant/irrelevant) and Imageability (continuous) as fixed effects as well as Participant (with a linear model formula for Cue and Imageability) and Item as random effects.

Results
In Experiments 1 and 2, both main effects and the interaction were highly significant (for descriptive statistics, see Figure C1 and C2, respectively; for β estimates and effectspecific χ²/t-tests, see Table C1). Please note that Imageability was not evenly distributed, due to the (methodologically introduced) dichotomy of concrete and abstract stimuli. Note. SE = standard error, df = degrees of freedom. Simple slope analyses with Imageability as predictor investigated the effect of Imageability within the contextual/irrelevant cue condition. Simple slope analyses with Cue as predictor investigated the effect of Cue within the high (+1 SD) and low (-1 SD) Imageability condition.