Embodied cognition theories predict a functional involvement of sensorimotor processes in language understanding. In a preregistered experiment, we tested this idea by investigating whether interfering with primary motor cortex (M1) activation can change how people construe meaning from action language. Participants were presented with sentences describing actions (e.g., "turning off the light") and asked to choose between two interpretations of their meaning, one more concrete (e.g., "flipping a switch") and another more abstract (e.g., "going to sleep"). Prior to this task, participants' M1 was disrupted using repetitive transcranial magnetic stimulation (rTMS). The results yielded strong evidence against the idea that M1-rTMS affects meaning construction (BF01 > 30). Additional analyses and control experiments suggest that the absence of effect cannot be accounted for by failure to inhibit M1, lack of construct validity of the task, or lack of power to detect a small effect. In sum, these results do not support a causal role for primary motor cortex in building meaning from action language.

Dozens of TMS and tDCS studies suggest a functional involvement of motor cortex in action language comprehension, supporting the embodied cognition view. In a recent study (Solana & Santiago, 2022, Neurosci. Biobehav. Rev., 141, 104834), we evaluated the soundness of this literature by means of p-curve analyses and tests for excess significance. The analysis estimated a low average power (≈ 30%) and showed signs of publication bias, which led us to conclude that this body of findings does not stand on solid ground. Yet, the employed techniques seem to perform poorly when high heterogeneity is present (as it is the case) and cannot quantify the amount of publication bias. For these reasons, in the present commentary, we reanalyzed the same set of studies (N = 43) with a method that does not have these limitations: z-curve analysis. Z-curve not only replicated our prior conclusions but showed an even more heartbreaking situation, with a lower power estimation (≈ 20%) and clear signs of a strong publication bias (around 7-10 unpublished contrasts expected to exist for each published one). Researchers on this topic should consider these shortcomings before drawing conclusions from this body of findings, as well as start implementing robust and transparent research practices to improve its reliability.