Oliveri et al. (2004) carried out the first TMS study to measure cortico-spinal excitability while participants processed action- and non-action-related words. The authors asked whether the suggested implication of motor regions in verb processing (Bak et al., 2001) reflected a grammatical class effect, or the semantic distinction between nouns and verbs, frequently denoting objects and actions, respectively.

Participants were instructed to generate aloud morphological transformations of visually presented nouns (singular or plural) and verbs (third person singular or plural). Using paired-pulse TMS over the left M1, where a suprathreshold TMS pulse is delivered immediately (10 ms) after a subthreshold conditioning pulse that has the function to pre-activate the target site (Kujirai et al., 1993), increased MEPs were registered from the right hand for action nouns and verbs, relative to non-action nouns and verbs.

These results provided indication that the motor cortex is sensitive to the action vs. non-action distinction of word meanings. The authors, however, did not rush to the conclusion that a causal involvement of the motor system in word processing had been proven; they rather emphasized how it could not be clarified whether motor activity concurrent with word processing was necessary for action-word processing, or reflected epiphenomenal spreading activation from the retrieved concept.

Pulvermüller (2005) argued that, in order for motor activity to be regarded as a component of word understanding, it should: (1) be somatotopic, respecting the bodily effector involved in the implied-language action; (2) occur as early as lexical-semantic access (i.e., ~200 ms); (3) occur automatically following word presentation, regardless of task demands, and (4) its perturbation should result in a change of language performance.

Pulvermüller et al. (2005) tested these predictions in a study where participants performed a lexical decision task (i.e., deciding whether a letter string is a word) with words related to arm- and leg-actions (grasp vs. kick). Subthreshold single-pulse TMS was applied 150 ms after word-onset, to modulate one of the following sites: arm representation in the left or in the right M1 (arm-M1) and leg representation in the left or in the right M1 (leg-M1). As a consequence of stimulation, participants' decision times were faster to arm-words (vs. leg-words) after TMS to the left arm-M1 and faster to leg-words (vs. arm-words) after TMS to the left leg-M1.

A tricky aspect of these results is that reaction times (RTs) to arm-words did not seem to differ across conditions with TMS to arm-M1, TMS to leg-M1, and sham-TMS (i.e., baseline condition; see Figure 2); that is, RTs to arm-words remained unchanged irrespective of whether TMS was delivered or not to either M1 site. The pattern of results was thus driven by variation in the performance with leg-words, visibly faster during leg-M1 stimulation than during arm-M1 stimulation. Taking advantage of this one data point, the authors concluded that a subthreshold TMS pulse applied to a region responsible for the semantic processing of words facilitated the upcoming word processing, just like a prime stimulus facilitates the processing of a semantically-related target word.

Assuming that the effect for leg-words was reliable and the analogy with semantic priming properly captures the effect of a subthreshold pulse on semantic processing, these results do not yet clarify the role of M1 in word understanding. In fact, it is equally possible that the activation of leg-M1 was directly involved in lexical decision, or that the “subliminal” stimulation of the leg site, encoded in conceptual regions, pre-activated the concept “leg” and thus facilitated the processing of the semantically congruent words.

Other studies used suprathreshold stimulation to elicit MEPs and therefore measure motor activity in language tasks. Buccino et al. (2005) applied single-pulse TMS to the left hand-M1 and leg-M1 and measured MEPs from hand- and foot-muscles, respectively, in correspondence with the acoustic presentation of verbs describing hand-actions (he took the cup), foot-actions (he kicked the ball), or abstract verbs (he loved his wife). Decreased MEPs were recorded from hand muscle after hand-action verbs (vs. foot- and abstract-verbs), and from foot muscle after foot-verbs (vs. hand- and abstract-verbs).

Later, Glenberg et al. (2008) applied suprathreshold TMS to the left hand-M1, while participants performed a semantic-plausibility judgment task on sentences describing physical transfer (you give the papers to Marco), abstract transfer (you delegate the responsibilities to Anna), or no-transfer (you read the papers with Marco). The authors found greater MEPs for both abstract- and concrete-transfer sentences relative to no-transfer items. This facilitation, however, was only found when concrete- and abstract-transfer items were compared, as a single condition, with the no-transfer items; the effect did not reach significance for either transfer-type sentence, when analyzed separately (p-values: 0.08 and 0.09 for concrete and abstract items vs. no-transfer items, respectively).

Strikingly, despite the similarity of procedures in Buccino et al. and in Glenberg et al., the two studies reported language-related motor interference and a trend toward facilitation, respectively. Motor facilitation, as reported by Glenberg et al., appears the most reliable result in the current literature (Oliveri et al., 2004; Tomasino et al., 2008; Papeo et al., 2009, 2011a). However, the lack of difference between concrete and abstract language in that study is hard to reconcile with the other TMS (and neuroimaging) studies, where abstract items have been used as the control condition to highlight action word-related motor activity. The authors did not explain how abstract relations can be delegated to motor information to an extent that is not distinguishable from concrete interactions.

Overall, the TMS results discussed so far show a certain variability at least in terms of direction of language-related motor effect (decreased vs. increased activity), and verbal materials associated with the effect (concrete-action vs. concrete + abstract language). This variability could extend to the types of language tasks that elicit motor activity. For instance, Papeo et al. (2009) found that MEPs increased for action-words when participants were instructed to think about the motor components of word-stimuli (semantic task), but not when the access to meaning was only incidental (i.e., in syllable counting). We note that not all studies reporting language-related motor facilitation involved explicit instructions to attend the motor components of words' meanings. Hypothetically, however, stimulation of M1 with its tangible consequence (i.e., the twitch) might act as a cue that activates the motor components associated with a word meaning.

So far, evidence has been provided that M1 activity changes when words with motor components are processed. However, to argue for a causal role of this activity, one should be able to show some sort of quantitative relation between changes in motor activity and semantic performance. While such result has not yet been reported, the behavioral consequences of TMS perturbation could help to approach this question. It is therefore surprising that many studies in the field restricted the data report to the physiological effect of TMS (increased/decreased MEPs), leaving aside its on-line behavioral effect (Oliveri et al., 2004; Buccino et al., 2005; Glenberg et al., 2008).

Studies comparing participants' performance (RTs and accuracy) with and without TMS to M1 gave conflicting results. While the study by Pulvermüller et al. (2005) found partial and unclear behavioral facilitation (see our discussion above), two studies by Papeo et al. (2009, 2011a) found increased M1 activity associated with action-related words, with no indication of action category-specific effect at behavioral level. Another study by Lo Gerfo et al. (2008) reported that, relative to the baseline condition (no TMS), participants were slower in morphological transformation of action words after prolonged exposure to repetitive TMS over the left M1(offline protocol). While this protocol may have greater interference strength relative to single-pulse TMS, it increasingly runs the risk of inducing widespread changes in neural activity at long-distant sites connected to the stimulated one (Chouinard et al., 2003; Kobayashi and Pascual-Leone, 2003; Huang et al., 2005). In the absence of any other evidence, the cautious interpretation of Lo Gerfo et al.'s unique observation is, as the authors themselves pointed out, that M1 enjoys connections with semantically-relevant regions. We will later return to this discussion.

Advocates of the “weaker” embodied hypothesis might argue that perturbation of M1 yields no behavioral change, because semantically-relevant information is contained at the level of premotor cortex, particularly in the ventral aspect of the precentral gyrus (e.g., Gallese et al., 1996; Damasio et al., 2004; Rizzolatti and Craighero, 2004; Martin, 2007). On this view, M1 activity would simply reflect downstream effects of premotor activation.

Cattaneo et al. (2010) used state-dependent TMS over the premotor cortex and assessed its effect on processing tool-nouns, a category of words whose meaning is associated with a manipulation program. State-dependent TMS rests upon the principle that physiological effects of TMS result from the interaction between the input-stimulus applied and the initial state of the target region (i.e., its level of activity). The initial state of a brain region, defined as the susceptibility of that region to be activated, can be influenced by any external or internal input, including task demand, experimental setting, individuals' expectations, and psychological state (Silvanto and Pascual-Leone, 2008).

In state-dependent TMS, as used in Cattaneo et al. (2010), the initial state of the target site was modulated behaviorally, through priming. The priming effect (i.e., the facilitation of processing a target-stimulus appearing after a perceptually or conceptually related prime-stimulus; Neely, 1977), is thought to reflect pre-activation or change in tuning of the neural population responsive to the “primed” features (Desimone, 1996; Wiggs and Martin, 1998; Grill-Spector et al., 2006). In a region that contains neurons responsive to a given target-category, a TMS pulse delivered immediately after the prime-stimulus facilitates responses to the target-stimulus when this is unprimed (i.e., preceded by an unrelated prime) relative to when it is primed. One interpretation of this phenomenon is that the firing rate of neurons in the stimulated region increases more, before reaching the ceiling, when the neurons are not pre-activated than when they have been pre-activated by the prime (Silvanto and Pascual-Leone, 2008).

In each participant, Cattaneo et al. (2010) targeted the left ventral premotor cortex (vPMC) as the experimental site and the left dorsal premotor cortex (dPMC) as the control site. Having the word “tool” as prime and nouns of tool-exemplars as targets, they found that categorical decisions (i.e., deciding whether the target was an exemplar of the tool-category) to unprimed targets (tool-names preceded by the unrelated prime “animal”) were faster with state-dependent TMS to vPMC, relative to the conditions with TMS to dPMC and no-TMS. RTs to the primed target (tool-names preceded by the prime “tool”) did not differ across TMS conditions; that is, the priming effect was abolished with TMS to vPMC. The priming effect was never affected by TMS, when prime-target pairs belonged to the “animal” category (control condition).

The category-specific effect in Cattaneo et al. rests upon the assumption that the firing rate of “tool-responsive” neurons in vPMC reached ceiling when the prime “tool” was presented. A parametric variation of the semantic distance of the prime from the tool-concept could prove this assumption true, by showing that the closer the semantic relation of the prime with “tool,” the weaker the facilitation of processing target-tools after state-dependent TMS. Leaving aside this methodological issue, the results by Cattaneo et al. do not clarify what kind of information is represented in vPMC (e.g., biological motion performed by the tool-user or tool motion), but they do provide indication that that brain site contains information, specific to the processing of tool-nouns.

The “virtual lesion” approach could extend those results, revealing selective TMS interference with the semantic processing of tool-nouns. A similar approach has been implemented in Willems et al. (2011) to investigate the role of the dorsal aspect of the premotor cortex (dPMC) in word processing. Participants performed a lexical decision task involving manual-action verbs, non-manual-action verbs (abstract) and legal pseudowords, after exposure to continuous theta burst stimulation (TBS). This protocol affects the excitability of neurons in the motor cortex in the direction of long-term (i.e., up to 1 h) inhibition (Huang et al., 2005).

Based on their own fMRI results (Willems et al., 2010), the authors selected the left dPMC as the target, and the right dPMC as the control site. Decision times to manual-action verbs were faster after TBS over left dPMC, than after TBS over right dPMC; RTs to abstract verbs did not differ with left and right TBS.

Although most theoretical and empirical reports implicate left vPMC in action-related conceptual processing, evidence also exists for semantic category-specific responses in dPMC (Beilock et al., 2008; Postle et al., 2008). Bringing support to the latter position, Willems et al.'s results appear, at first sight, in conflict with those of Cattaneo et al. (2010), where left dPMC was the control site and its stimulation led to no behavioral effect. Recall, however, that in Cattaneo et al. stimuli were nouns, while in Willems et al. they were verbs and, when nouns were used, no effect was found in left dPMC (Table 1; Figure 1).

One possibility is that the left dPMC is recruited when processing verbs and the left vPMC is specific to the processing of nouns. Tool nouns and action verbs carry different types of action-related information: one gross distinction is that, in the case of nouns, action information relates to a specific context (i.e., the specific tool), in the case of verbs, it relates to a specific movement or motor program (e.g., grasping) that applies to several contexts. Conjecturally, the one-to-one vs. one-to-many ratio between the verbal label and the implied motor context could capture the difference between tool-nouns and manual-action verbs and underlie a functional segregation within the precentral gyrus⁴.

One serious issue raised by Willems et al.'s results concerns the direction of the effect: behavioral improvement as opposed to the behavioral impairment that is expected as a consequence of the inhibitory effect of continuous TBS (Huang et al., 2005). If the metaphor of TMS as “virtual lesion” stands, Willems et al.'s pattern is reminiscent of the paradoxical facilitation of performance reported in the brain-lesion literature (Kapur, 1996), and interpreted as evidence for a competing/inhibitory function of the lesioned site relative to the assessed behavior. In this perspective, it is entirely possible that the physiological response of dPMC—even if inhibitory—contributes to some aspect of lexical performance. Certainly, the violation of the expected TBS-induced disruption of behavior solicits caution in interpreting Willems et al.'s findings as conclusive demonstration of brain-behavior causality.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.