Processing of Action and Sound Verbs in Context: An FMRI Study

Abstract Recent theories propose a flexible recruitment of sensory and motor brain regions during conceptual processing depending on context and task. The present functional magnetic resonance imaging study investigated the influence of context and task on conceptual processing of action and sound verbs. Participants first performed an explicit semantic context decision task, in which action and sound verbs were presented together with a context noun. The same verbs were repeatedly presented in a subsequent implicit lexical decision task together with new action and sound verbs. Thereafter, motor and acoustic localizer tasks were administered to identify brain regions involved in perception and action. During the explicit task, we found differential activations to action and sound verbs near corresponding sensorimotor brain regions. During the implicit lexical decision task, differences between action and sound verbs were absent. However, feature-specific repetition effects were observed near corresponding sensorimotor brain regions. The present results suggest flexible conceptual representations depending on context and task. Feature-specific effects were observed only near, but not within corresponding sensorimotor brain regions, as defined by the localizer tasks. Our results therefore only provide limited evidence in favor of grounded cognition theories assuming a close link between the conceptual and the sensorimotor systems.


Introduction
Conceptual representations in human longterm memory constitute the meaning of words and are therefore a major prerequisite for language comprehension and production [1,2]. However, despite the common agreement for conceptual representations being stored in semantic memory [3], it is controversially debated, how concepts are represented within this memory system [2]. According to classical amodal theories, conceptual information is represented in an abstract and symbolic way detached from the sensorimotor brain systems involved in perception and action [4,5]. These amodal representations are thought to be held in heteromodal brain regions such as the anterior temporal cortex including the temporal pole [6,7]. Some variants of amodal theories assume that sensorimotor activation occurs as a consequence of semantic elaboration after access to the amodal concept (secondary embodiment) [8]. In contrast, a functional contribution of the sensory and motor brain systems for conceptual representations is postulated by modality-specific accounts (also called: embodied or grounded cognition theories) [9][10][11][12]. Modality-specific theories [13][14][15][16][17] propose that concepts are represented in the respective sensory and motor brain systems depending on the relevance of sensorimotor features for a given concept (concept-inherent feature-relevance). It is thought, that access to conceptual knowledge at least partially reactivates modality-specific representations established during former sensory and motor experiences with the referent [2,[18][19][20]. This account implicates modality-specific brain activations during conceptual processing at multiple levels involving primary and secondary sensorimotor brain areas [13,21], similar to brain activity during action and perception [22,23]. There are several variants of modality-specific theories, which vary depending on the precise relationship between the sensorimotor systems and the conceptual system [for a review, see 24]: While some modality-specific accounts assume an identical neural substrate between conceptual and sensorimotor processing [25,26], other variants of modality-specific theories do not assume a complete overlap of their neural substrates [22,[27][28][29]. Instead, depending on task or context, a hierarchy of neural circuits involving modality-specific as well as adjacent multimodal higher-level cortices is assumed to be involved in the processing of conceptual information [28,30,31].
Evidence for modality-specific processing of concepts comes from behavioral [32], neuroimaging [13,17,33,34] and electrophysiological studies [35][36][37]. For example, a functional magnetic resonance imaging (fMRI) study by Hauk and colleagues showed that action words like to lick, to pick or to kick somatotopically activated motor and premotor brain regions similar to those activated during movements of the tongue, fingers and feet [13]. Moreover, lesion or transcranial magnetic stimulation (TMS) studies underline the functional role of modalityspecific brain areas for conceptual processing [17,[38][39][40]. Kemmerer et al. [38] reported that brain lesions of the arm-or hand-related part of the left precentral gyrus were specifically associated with impaired knowledge about arm-or hand-related concepts. In a similar way, Trumpp et al. [17]  Originally, conceptual representations were assumed to be stable and thus unaffected by different tasks and contexts [41,42]. Other, more recent approaches, however, propose flexible conceptual representations [43]: Each word can have different shadings of meaning depending on the respective context. For example, a piano is highly related with sound in the context of a musical event but has a strong weight-related meaning in the context of a move [44]. Some amodal approaches like the distributed semantic network model [5][6][7] and the controlled semantic cognition (CSC) framework [45] assume flexible amodal conceptual representations held in anterior temporal cortex, a heteromodal brain region serving as semantic hub. Modality-specific approaches, in contrast, assume a situational recruitment of sensorimotor brain regions, which constitute the concept [23,33,46].
A recent neurobiologically inspired model explains conceptual flexibility within modalityspecific brain systems at a mechanistic level [31]: Flexibility based on context-related semantic processing is assumed to depend on pre-activation of specific modality-specific circuits by semantic priming mechanisms [31]. Task-related flexibility, on the other hand, is explained by cortical gain control mechanisms. It is assumed that, depending on task requirements, attention is directed to a certain sensorimotor meaning aspect. This is mechanistically realized by a change of the gain in the corresponding neural circuits, thereby regulating their activity depending on the task at hand [31].
In line with the assumption of flexible modality-specific conceptual representations, behavioral [47], neuroimaging [33,46] and event-related potential (ERP) studies [46] suggest that conceptual features stored in sensory and motor brain areas are flexibly activated depending on context. For example, the level of activation in corresponding sensory and motor brain areas by visual and action nouns changed depending on which conceptual features were highlighted by a semantic context [46]. Furthermore, several studies show task-related modulations of brain activity during conceptual processing supporting the flexibility assumption. For example, modality-specific effects for verbs were found in semantic decision tasks, where feature-specific information is explicitly retrieved, but not in lexical decision tasks, which probe conceptual retrieval implicitly [48,49]. Conceptual flexibility may explain why modality-specific effects were inconsistently observed across task and context conditions [20,46].
Previous studies on feature-specific conceptual representations [50][51][52] focused on the comparison between nouns and verbs by assuming that the meaning of nouns is primarily characterized by visual objectrelated conceptual features [53], while verbs are primarily associated with action-related features [54]. However, the meaning of nouns does depend not only on visual, but also on acoustic [22] and action [55] features. Likewise, the meaning of verbs is not only constituted by conceptual action features, but also by visual or acoustic features [56]. For this reason, the processing of conceptual features should not be investigated between word classes, but within the respective word class.
Many studies investigated modalityspecific representations of conceptual information within the word class of nouns. Action nouns (e.g. hammer) were found to activate fronto-parietal sensorimotor brain areas that were associated with the use of tools [57]. In a similar way, sound nouns (e.g. radio) were found to specifically activate temporal brain regions that are associated with the perception of real sounds [22,58].
Furthermore, electrophysiological experiments found differential ERPs for action-and soundrelated nouns [36] indicating that different modality-specific neural circuits are involved in conceptual processing of nouns depending on conceptual feature relevance [22].
Several investigations also focused on conceptual processing of concept-inherent feature-specific information within the word class of verbs [13,[59][60][61][62]. Popp et al. [56] showed that action-and sound-related verbs (e.g. to throw or to crackle) elicited differential ERPeffects, which were largely similar to previous ERP findings on action-related and soundrelated nouns with respect to the polarity and time course of ERPs [36,63]. Differential processing of action-and sound-related verbs in corresponding modal brain regions during a lexical decision task has also been demonstrated in a previous fMRI study (Popp et al., submitted for publication). These findings suggest that different activity in motor and auditory brain regions underlie the differential ERP-effects to action-and sound-related verbs.
They indicate that words, whether nouns or verbs, are differentially processed depending on their respective concept-inherent feature relevance as predicted by modality-specific accounts. Popp et al. [56] additionally showed that this feature-specific ERP effects to actionand sound-related verbs depended on the context: The polarity of feature-specific ERP effects for action-and sound-related verbs during a lexical decision task was reversed when these verbs were previously presented together with a context noun during a semantic context decision task.
In several repetition priming studies, deactivation of brain areas, an effect, which has been coined "repetition suppression" [64] has been frequently observed [65][66][67]. Repetition was expected to lead to reduced neural activity in the respective modality-specific systems depending on verb category (action vs. sound verbs), but only if the same processes are involved for repeated and unrepeated stimuli. However, if repeated stimuli induce new processes such as additional exemplar recognition or explicit memory retrieval, repetition enhancement occurs in those brain regions that are involved in additional operations [64] including middle frontal, inferior parietal and middle temporal gyrus [68,69]. However, due to overlap of brain potentials on the scalp surface, the ERP methodology does not allow to determine unequivocally whether these polarity-reversed ERP effects for repeated action-and sound-related verbs were due to repetition suppression or repetition enhancement.
The present fMRI study was conducted to investigate effects of task and context on conceptual processing within the modalityspecific framework and addressed three issues.
First, we investigated effects of context via additional contextual cues (context nouns) by identifying brain areas that were activated during the processing of contextually related vs. unrelated noun-verb pairs during an explicit semantic decision task. Second, to investigate task-dependent effects on concept-inherent feature-specific conceptual processing, we compared action-and sound-related verb processing during a semantic context decision task, in which semantic information is explicitly retrieved, with verb processing during a lexical decision task, in which access to conceptual information is implicit.
Third, to investigate context via the previous presentation of verbs in an explicit context decision task, we investigated how repetition of action-and sound-related verbs from the previous context decision task modulated conceptual processing of action-and soundrelated verbs in a subsequent lexical decision task. To answer these research questions, we have, to the best of our knowledge, for the first time implemented a study design that allows the investigation of conceptual flexibility depending on context and the task within a modality-specific framework.
The present fMRI study comprised four tasks.
In the first task, subjects had to perform a semantic context decision task, in which actionrelated or sound-related verbs were provided either with a semantically related or with a semantically unrelated context noun.  [75,76]. In addition to these general repetition suppression effects, we hypothesized specific repetition suppression effects for action-and sound-related verbs in modality-specific brain systems, which process the task-dependent relevant conceptual feature. Alternatively, not only repetition suppression, but also repetition enhancement could be observed in the present paradigm: It is possible that previous presentation of verbs within the context decision task induces elaborative semantic processing of the words in the subsequent lexical decision task [77]. These additional semantic processes for repeated words could lead to repetition enhancement effects compared to new words, i.e. an increase of activity in corresponding brain systems [64].
In addition to general repetition enhancement effects in parts of the language system, elaborative processing of repeated words could also result in an increase of activity within motor and auditory brain areas depending on verb category (i.e. action vs. sound verbs).

Stimuli
The stimulus set consisted of 40 action verbs and 40 sound verbs that were used in a previous study [56]. For the semantic context decision task, each verb was combined with a semantically related context noun (e.g. ball -to throw/bell -to ring), which was also used in a previous study [56].
Context nouns, which were combined with

Acoustic localizer task
The third task was an acoustic localizer task adopted from previous studies [22,80] The duration of the acoustic localizer was approximately 7 minutes (214 brain volume images acquired in total).

Motor localizer task
As a fourth task, a motor localizer task was administered. The motor localizer was similarly constructed as the acoustic localizer task:

FMRI data acquisition and analysis
Structural and functional MR images were acquired using a 3-tesla scanning system   [13,22,82]. Anatomical labels for activated brain areas were determined using the SPM Anatomy toolbox [83].

Semantic context decision task (explicit task)
In the semantic context decision task, we Further activation was found in the right calcarine sulcus, the right hippocampus, the left precuneus and the thalamus. No significant activations could be observed for the reversed interaction between semantic relatedness and feature category (Table 4, Figure 1E).
To explore these interactions, we further    Figure 2).

Lexical decision task (implicit task)
In a first step, we assessed the main effect of

Brain region MNI coordinates (mm) Peak T Cluster size (voxels) Cluster p (uncorrected)
Inferior parietal L -56 - 48         to action verbs compared to sound verbs in the explicit task, whereas this feature-specific effect was absent in the implicit task. The reversed interaction did not reach significance (Table 8).  (Table 9).

Localizer tasks
To test whether activation during the We observed differential main effects for action-and sound-related verbs only during the explicit semantic context decision task that were only partially in accordance with our hypotheses. These differential category-specific effects for action-and sound-related verbs in the explicit task were more pronounced, when these verbs were presented together with a semantically related context noun.
With semantically unrelated context nouns, differences between action-and sound-related  Table 8: Activation peaks for the main effect type of task and main effect feature category across both tasks as well as the interactions between feature type and task.
Reported clusters are significant at p < .05 (FWE-corrected) at the cluster-level. Listed are peak voxels with highest t-values for significant clusters and their local maxima more than 8 mm apart. Asterisks indicate clusters that survive cluster FWE-correction (p < .05). MNI: Montréal Neurological Institute, FWE: Family-wise error, R: right, L: left.

Brain region MNI coordinates (mm) Peak T Cluster size (voxels) Cluster p (uncorrected)
Explicit versus implicit task

Context effects in the semantic context decision task
During the semantic context decision task, we first investigated the neural correlates of processing the verbs in a related versus unrelated context. Stronger activation to semantically related compared to unrelated noun-verb-pairs was found in left inferior parietal, middle temporal, middle frontal and inferior frontal brain areas. Enhanced activation in a fronto-parieto-temporal network subserving semantic retrieval [84] is consistent with findings obtained during semantic priming in previous studies [73,85,86]. The same network was also active when the explicit and the implicit tasks were compared in the present study. Hence, as expected, the explicit semantic context decision task more heavily recruited the semantic retrieval network than the lexical decision task. Stronger activation to unrelated compared to related noun-verb-pairs was found in left motor and somatosensory cortex. A comparable activation pattern in primary and premotor areas was found for unrelated compared to related word pairs in a previous study [87]. Greater activation in this brain region to unrelated word pairs could indicate that stimulus inherent action features of the context nouns were more strongly recruited, in order to decide, whether the noun is related to the verb or not. Alternatively, it might be referred to the longer reaction times for unrelated compared with related word pairs. Compared to unrelated word pairs, semantic decisions on related word pairs were faster as it is typically observed in such a task [88]. Although faster reactions were frequently associated with deactivation at the neural level [64], a growing number of studies [73,85,86] as well as our results indicate that facilitation at the behavioral level is not necessarily related to decreased neural activity. Instead, behavioral facilitation can also be accompanied by previously been identified to be involved in the processing of action-related concepts in several neuroimaging studies [12,58]. These activations in posterior middle temporal gyrus were close to motion-sensitive areas [90,91] which were also recruited by the performance of movements as shown by the motor localizer (smallest distance: 4 mm) in the present study.
Interestingly, recent research indicated that the posterior middle temporal gyrus is functionally connected to the pre-and post-central gyrus [92]   This could explain why activations to action words were found in the motor system only during passive tasks [13,97] which do not require a motor response, but not during active tasks [98]. Therefore, a task that does not require a motor response could possibly yield stronger activations in primary and secondary motor brain areas [96]. A significant interaction between feature category and semantic relatedness was found in a network of inferior parietal/angular, middle temporal and superior frontal brain regions, which were found to be involved in semantic processing [84]. To explore this interaction, effects of feature category were calculated separately within semantically related and semantically unrelated word pairs. Greater activation to action verbs was found in the middle frontal/precentral gyrus near activations obtained during the motor localizer and located in ventral premotor cortex, a brain region that is associated with the representation of hand actions [100]. 107] and has been related to action-related motion [90,91]. Its homologue in the macaque brain involves mirror neurons that respond to action observation and action performance [108]. This suggests that action information is specifically retrieved during the decision,

Comparisons between the explicit and the implicit task
The comparison of activations obtained during the implicit lexical decision task with activations during the explicit semantic context decision task revealed greater activations during the explicit task in a large fronto-parieto-temporal semantic retrieval network [84], which is largely comparable to those activations obtained in the main effect semantic relatedness during the semantic context decision task. Greater activations to the implicit compared with the explicit task were found in a parietal brain region comprising the precuneus, the cingulate cortex and the superior temporal cortex. These brain regions were associated with implicit word processing in previous studies [115].
These results indicate that the explicit task recruited a semantic retrieval network more strongly than the implicit task. Significant interaction of feature category and task were found in left posterior middle temporal cortex.
This indicates that action features of verbs were more strongly recruited in the explicit than in the implicit task (see also discussion above).

Implications for theories of conceptual representations
During the explicit semantic context decision task, we found differential main effects of feature category for action-and sound-related verbs. However, anatomical localization of these activations were only partially compatible with our expectations given the stronger activation of motor brain regions to sound-compared with action-related verbs. During the lexical decision task, category-specific effects for action-and sound-related verbs were observed only in terms of repetition effects. Contrary to our expectations and incompatible with the basic assumptions of any variant of modality-specific theories [13][14][15][16], newly presented action-and sound-related verbs in the lexical decision task did not elicit differential brain activations.
Similar to the present results, sensorimotor activation during conceptual processing has previously been mainly observed during explicit semantic tasks [48,49]. Generally, such task-specific sensorimotor activation has been typically interpreted within amodal theories, which assume that sensorimotor activation occurs only as a consequence of semantic elaboration after access to the amodal concept (secondary embodiment) [8]. A limitation of the present work are the lenient statistical thresholds applied to assess task and feature-specific effects. This indicates that effect sizes were small, and results were statistically not very robust. It should be noted that more lenient statistical thresholds were also applied in earlier studies [80] to capture subtle differential contrasts with small effect sizes. The missing robustness of the effects and the observation of verb category effects on sensorimotor activity in the explicit semantic context task and of different repetition effects as a function of category in the lexical decision task thus seem to favor amodal theories.
However, we think that it is premature to take the present results as well as comparable earlier findings as falsification of modalityspecific theories. Firstly, it should be noted that earlier studies observed sensorimotor activity during a lexical decision task for both nouns [22,58] and verbs (Popp et al., submitted for publication). Secondly, ERP recordings revealed an early onset of sensorimotor activity [14,36,56,116] indicating that it reflects rapid access to modality-specific concept-inherent features and not post-conceptual elaboration processes [77]. Thirdly, it should be noted that the number of stimuli per condition (20 verbs repetition effects in a within-subject design.
The relatively low signal-to-noise ratio in the present study could have prevented to detect weak sensorimotor activation in the lexical decision task. The number of used stimuli for each category in the present study was the same as in a previous study, which also found no differential feature-specific effects between hand-and foot-related action verbs in a lexical decision task [48]. Fourthly, task and context effects on sensorimotor activation observed in the present study are compatible with the notion of conceptual flexibility within a modality-specific framework [20,31,33,43,46]. According to this view, the featural composition of a concept strongly depends on the context. Therefore, contribution of specific sensory and motor modalities to a concept is not situationally invariant, but is modulated by task demands or context. This explanation is also compatible with the fact that some studies reported differential effects for verbs with a concept-inherent focus on specific features such as action during implicit tasks [62,117]: Interestingly, these previous studies included more difficult implicit tasks (e.g. speeded lexical decision or additional handmovements during the lexical decision). We assume that task orientation (explicit context decision vs. implicit lexical decision task) and the previous presentation of verbs together with a context noun during a semantic context decision task enhances conceptual processing within sensorimotor areas during a subsequent lexical decision task. Future studies with a higher number of stimuli and thus a better signal-to-noise ratio could further test the notion of modality-specific flexible conceptual processing. Finally, the lack of a substantial functional anatomical overlap between conceptual feature activation during the semantic tasks and activation during the acoustic or motor localizer seem to contradict modality-specific theories. Please note that some variants of modality-specific theories [20,22,23,118] do not assume a complete neuroanatomical overlap of activation during the processing of concept-inherent features and neural circuits associated with real action or perception. Re-activation of sensorimotor experiences during conceptual processing may also involve higher-level modalityspecific or multimodal association cortex within a hierarchy of functional brain networks at various levels of abstractions, which were not activated by the localizer tasks [28,30,31]. It is therefore possible that the functionalanatomical overlap between conceptual and sensorimotor processing depends on the stimuli and actions used in the localizer tasks.
To conclude, in the present study we behavioral interference or TMS studies [40,119].