Experiential grounding of abstract concepts: Processing of abstract mental state concepts engages brain regions involved in mentalizing, automatic speech, and lip movements

concepts like mental state concepts lack a physical referent, which can be directly perceived. Classical theories therefore claim that abstract concepts require amodal representations detached from experiential brain systems. However, grounded cognition approaches suggest an involvement of modal experiential brain regions in the processing of abstract concepts. In the present functional magnetic resonance imaging study, we investigated the relation of the processing of abstract mental state concepts to modal experiential brain systems in a fine-grained fashion. Participants performed lexical decisions on abstract mental state as well as on verbal association concepts as control category. Experiential brain systems related to the processing of mental states, generating verbal associations, automatic speech as well as hand and lip movements were determined by corresponding localizer tasks. Processing of abstract mental state concepts neuroanatomically overlapped with activity patterns associated with processing of mental states, generating verbal associations, automatic speech and lip movements. Hence, mental state concepts activate the mentalizing brain network, complemented by perceptual-motor brain regions involved in simulation of visual or action features associated with social interactions, linguistic brain regions as well as face-motor brain regions recruited for articulation. The present results provide compelling evidence for the rich grounding of abstract mental state concepts in experiential brain systems related to mentalizing, verbal communication and mouth action.


Introduction
Concepts held in semantic long-term memory (Tulving, 1972) code the meaning of objects, events or ideas and are the cognitive basis of word meaning (Humphreys et al., 1999;Kiefer and Harpaintner, 2020;Levelt et al., 1999).While concrete concepts such as "hammer" have referents, which can be perceived and acted upon, abstract concepts such as "will" or "intention", which refer to mental states, lack such a physical referent per definition.It is a matter of a current debate, which neural circuits support knowledge representations of such abstract mental state concepts.
According to traditional amodal theories, both concrete and abstract concepts are stored in an amodal representation format detached from modal experiential brain circuits of perception, action and introspection (Anderson, 1983;Mahon and Caramazza, 2009).It is assumed that these amodal representations are stored in semantic hubs (Binder, 2016) in heteromodal association cortex located in the anterior temporal (Rogers et al., 2004) or posterior temporal lobe (Hoffman et al., 2012).Activation in modal experiential brain circuits during conceptual processing is thought to be attributed to passive spreading of activation to modal perceptual input or action output systems (Mahon, 2015a(Mahon, , 2015b)), imagery (Machery, 2007) or semantic elaboration (Mahon and Caramazza, 2008), after accessing the amodal conceptual representation.
The importance of mental states for a subgroup of abstract concepts has been demonstrated in the property-listing study of Harpaintner et al. (2018): Participants were asked to generate spontaneous associations for 296 abstract concepts.These associations were classified according to their modality-specific and verbal feature content revealing considerable proportions of sensory-motor, introspective, emotional and social features in addition to verbal associations.Hierarchical cluster analyses identified subcategories of abstract concepts, which were dominated by different feature types.One of these subcategories, for example, was characterized by introspective features related to mental states.Similarly, rating studies of Troche et al. (2014) and Villani et al. (2019) demonstrated that introspective features are highly relevant for specific subcategories of abstract concepts.In line with these findings, some neuroimaging studies revealed an involvement of the mentalizing brain network including orbito-frontal cortex, medial prefrontal cortex, posterior cingulate cortex and superior temporal sulcus during processing of abstract mental state concepts (Baron-Cohen et al., 1994;Conca et al., 2021;Huth et al., 2016;Kiefer et al., 2022;Wilson-Mendenhall et al., 2013).However, abstract mental state concepts are not only associated with mentalizing processes.A study of Dreyer et al. (2018) showed that passive reading of mental non-emotional abstract words elicited activation in face-motor brain regions.According to Borghi et al. (2017Borghi et al. ( , 2016)), such face-motor brain activations can be traced back to the acquisition modality of abstract concepts: Given that abstract concepts are thought to be mainly acquired through linguistic input during the course of social interaction and verbal communication with others, they depend on processing in the language brain areas including mouth motor areas used for articulation.Furthermore, engagement of inner speech during concept acquisition or use, supporting inner monitoring or inner social metacognition, might also contribute to the involvement of language and mouth motor systems during the processing of abstract concepts (Borghi and Fernyhough, 2023;Fernyhough and Borghi, 2023).Thus, processing of abstract mental state concepts might also be associated with activation of motor regions related to mouth movements (Borghi and Zarcone, 2016;Dreyer and Pulvermüller, 2018;Ghio et al., 2013).
In a recent event-related-potential (ERP) study (Kiefer et al., 2022), we investigated the time course of brain activity related to the processing of semantically predefined abstract mental state (MST) compared to abstract verbal association (VAS) concepts while participants performed a lexical decision task (LDT) on visually presented words.MST and VAS concepts were selected based on the results of a previous property listing study (Harpaintner et al., 2018).MST concepts had a high proportion of properties related to introspective or mental state features and referred to aspects of cognition, mental states and introspections (e.g., "thought", "expectation", "surprise", "desire").Abstract VAS concepts had a high proportion of verbal association features (e.g., "ritual", "justice", "difference", "revolution").
Estimated electrical source activity for VAS concepts was found in left occipital brain regions 194 ms to 210 ms after word onset, possibly reflecting activation of the visual word form area involved in processing of verbal associations or retrieval of visual features of associated concepts.Enhanced estimated source activity for MST concepts, however, was found in visuo-motor (left superior parietal cortex and right pre-and postcentral gyrus) and mentalizing brain networks (bilateral lateral and medial prefrontal cortex, right insula, left precuneus and right temporoparietal junction (TPJ)) starting 212 ms and extending to a late time window beyond 500 ms after word presentation.The relatively early onset of enhanced activation for MST compared to VAS concepts reflects the initial access to distinct conceptual features in the corresponding modal brain regions.The somewhat later effects after 300 ms, however, suggest the involvement of a second post-conceptual processing phase of semantic elaboration.The involved superior parietal cortex as part of the visuo-motor brain network has previously been found to be active, when participants observed actions (Sim et al., 2015) and during processing of conceptual action information (Kuhnke et al., 2020b).Activation of primary sensory-motor cortex during retrieval of MST concepts (Dreyer and Pulvermüller, 2018;Muraki et al., 2020), instead, is traced back to articulatory programs, which are recruited in consequence of the acquisition of abstract concepts through verbal communication (see above Borghi and Binkofski, 2014).Thus, activation in this large visuo-motor brain network is supposed to reflect articulatory processes, but also simulations of actions, which occur in situations associated with certain MST concepts (Barsalou et al., 2018).Action simulation within the motor brain system might be one cognitive basis for mentalizing and can be used, for example, to decode mental states like goals, beliefs, desires or perspective taking (Arioli and Canessa, 2019;Canessa et al., 2012;Wang et al., 2016).Lateral and medial prefrontal cortex, insula, precuneus, and TPJ are involved in mentalizing (Frith and Frith, 2006) including Theory of Mind (ToM), social interaction and empathy (Abu-Akel and Shamay-Tsoory, 2011;Arioli and Canessa, 2019;Canessa et al., 2012;Geiger et al., 2019;Singer et al., 2009).In detail, lateral and medial prefrontal cortex are supposed to be involved in analyzing thoughts, intentions, desires, beliefs, and in the assignment of emotional values (Abu-Akel and Shamay-Tsoory, 2011).The insula is activated during emotional tasks (Singer et al., 2009) including emotional word processing (Ponz et al., 2014;Ziegler et al., 2018), associating bodily states with subjective feeling (Damasio, 1994) and emotional empathy (Bird et al., 2010;Singer et al., 2009Singer et al., , 2004)).The precuneus has been associated with self-referential thought (Abu-Akel and Shamay-Tsoory, 2011; Cabeza and St Jacques, 2007), but also with many other forms of mentalizing (Abu-Akel and Shamay-Tsoory, 2011;Schneider-Hassloff et al., 2015).The TPJ is seen as an interface of mentalizing and visuo-motor brain networks.It is involved in decoding intentions of actions by analyzing visuo-spatial information (Arioli and Canessa, 2019).Furthermore, it plays a crucial role in mental perspective taking and social cognition (Martin et al., 2020;Seymour et al., 2018;Wang et al., 2016).Precuneus and TPJ both provide input for the prefrontal brain regions of the mentalizing network (Abu-Akel and Shamay-Tsoory, 2011).Thus, in accordance with the grounded cognition approach, our recent ERP study (Kiefer et al., 2022) demonstrated early and late processing of abstract mental state concepts in specific modal brain circuits distinct from abstract verbal association concepts.It confirmed the heterogeneity and diversity of abstract concepts depending on their semantic content.
N.M.Trumpp et al.However, as scalp potentials have a poor spatial resolution, in particular localizations of sources in structures underneath the cortical surface (e.g., insula or medial prefrontal cortex) have to be viewed with caution.Given this limited spatial resolution, source estimations of scalp ERPs are also not particularly suited to reveal body-part specificity of activity in motor brain regions (hand vs. mouth actions) during conceptual processing of abstract mental states.In recognizing this limitation, the previous ERP study did not include localizer tasks, in order to investigate the functional overlap of abstract MST and VAS concepts with experiential modal brain circuits involved in body-part-specific actions, mentalizing, and processing of verbal associations.Furthermore, as the meaning of abstract MST concepts might be partly determined by verbal communication with other individuals or inner speech, it is important to test whether activity in motor areas occurs in parts of the motor strip specifically involved in mouth-related movements (Borghi and Zarcone, 2016;Borghi, 2022;Dreyer and Pulvermüller, 2018;Fini et al., 2022;Ghio et al., 2013).
In the present functional magnetic resonance imaging (fMRI) study, we investigated the functional neuroanatomy underlying processing of abstract MST and VAS concepts in a fine-grained manner.Most importantly, we assessed the neuroanatomical overlap of the activity pattern associated with the processing of MST and VAS concepts with the neural structures involved in mental states, generating verbal associations, automatic speech as well as hand and lip movements as determined by corresponding localizer tasks.This allows us to identify the contribution of modal experiential brain circuits related to mentalizing and body part-specific action as well as those related to linguistic processing during processing of MST and VAS concepts.
During fMRI scanning, participants performed a LDT on words referring to 30 abstract MST and 30 abstract VAS concepts among 60 pseudowords drawn from the earlier ERP study (Kiefer et al., 2022).In the LDT, conceptual representations are accessed implicitly (Dilkina et al., 2010;Kiefer, 2002).MST and VAS concepts were equated concerning various conceptual and psycholinguistic variables (see methods section).In addition to the LDT, we administered three localizer tasks mapping processing of mental states, automatic vs. generative linguistic processing, and real hand vs. lip movements in the brain.In the mental state localizer task, participants were presented with false belief and false photograph stories, which they should read attentively and subsequently answer a question related to the story.This task probes a form of mentalizing, specifically ToM (see Dodell-Feder et al., 2011;Saxe and Kanwisher, 2003;Saxe and Powell, 2006).In the verbal association localizer task, participants were presented with single words requesting either automatic speech (e.g. the word "months" indicated participants to name calendar months in chronological order) or generation of spontaneous associations for a specific category (e.g."fruits").The motor localizer task tested participants' body-part-specific brain activation associated with movements of their left/right hands by pressing a hand training ball and their lips by quietly forming the phonemes related to the letters "o" and "i" in German language.Furthermore, in order to assess the relevance of mouth vs. hand movements for the present set of abstract MST concepts at a behavioral level, we also asked participants in a separate rating study to indicate how much the mouth or the hand is involved in a possible action with the named MST and VAS concepts, respectively.Additionally, we asked participants to rate if they think they might have learned the meaning of these concepts by experience or by language.Thus, we inquired mode of acquisition (MoA) as an additional psycholinguistic variable as recommended by Borghi and colleagues (Borghi and Zarcone, 2016;Borghi, 2022).
In line with grounded cognition theories, we hypothesized that MST and VAS concepts would activate different modal brain regions spatially overlapping with activation during specific localizer tasks.MST concepts should activate mentalizing brain systems including lateral and medial prefrontal cortex, insula, precuneus and TPJ (Conca et al., 2021;Kiefer et al., 2022) as well as face-motor brain regions (Borghi, 2022;Dreyer and Pulvermüller, 2018;Ghio et al., 2013) overlapping with activation during ToM and lip movements in the respective localizer tasks.VAS concepts, in contrast, should activate left occipital cortex (Kiefer et al., 2022) spatially overlapping with activation during generation of verbal associations (verbal association localizer task).Differential activity of MST and VAS abstract concepts in classical perisylvian language areas in frontal and temporal brain regions were not expected, as general linguistic processing might be equally important for both abstract concept types.This assumption is based on the results of the earlier ERP study (Kiefer et al., 2022), in which perisylvian language areas did not show differential activity for MST and VAS concepts.VAS concepts are characterized by verbal associations as identified in the property listings (Harpaintner et al., 2018).In the coding scheme of the property listings (Harpaintner et al., 2018), verbal associations were defined as "a feature or a situation that does not describe the abstract concept, but which is thematically or symbolically related to it" (p.5).Given this operational definition of verbal associations for the purpose of the present work, this feature category is not specifically related to social interactions and verbal communication.However, in line with the idea that acquisition of concepts related to mental states involves verbal communication with others (Borghi and Zarcone, 2016;Borghi, 2022) and inner speech (Borghi and Fernyhough, 2023;Fernyhough and Borghi, 2023), we expected that MST concepts would receive a higher relevance of mouth actions than VAS concepts in the ratings.Such a pattern of behavioral and neuroimaging findings would provide fine-grained information with regard to the grounding of abstract MST concepts in experiential brain circuits related to mentalizing, verbal communication, and mouth actions.

Rating study
Twenty-one healthy, native German-speaking volunteers (M age = 29.43 years, range from 21 to 43 years, 17 females) participated in an online rating study created with UNIPARK (Software: EFS survey by Tivian; https://www.unipark.com).They were presented with a list of abstract MST and VAS concepts used in our previous ERP (Kiefer et al., 2022) and the present fMRI study (see below) and rated mode of acquisition (MoA) as well as relevance of mouth and hand action on a seven-point Likert scale.With regard to MoA, participants were asked to rate how they think they have learned the meaning of the respective concept.The score of "1″ corresponded to "totally acquired through experience" and the score of "7" to "totally acquired through language".Scores in-between corresponded to a combination of experience and language with different weighting.Participants were also asked to rate how much the mouth and the hand, respectively, is involved in a possible action with the named concept.Scores ranged from "1" corresponding to "not involved" to "7" corresponding to "highly involved".Instructions were based on the study of Villani et al. (2019).Differences between MST and VAS abstract concepts regarding MoA and the involvement of the mouth and hand, respectively, were assessed with two-tailed paired t-tests on mean scores of the ratings across the sets of MST vs. VAS concepts.

fMRI study 2.2.1. Participants
Fifty right-handed (according to Oldfield, 1971), healthy, native German-speaking participants voluntarily took part in the study.Three participants were excluded from the fMRI data analysis because of intense motion artefacts (n = 2) or fronto-lateral signal loss (n = 1, wearing normal instead of MR-compatible glasses in order to better recognize the presented stimuli).Forty-seven participants (n = 35 females) with a mean age of 23.09 years, ranging from 19 to 44 years, were included in the final analysis.According to self-report, participants had no history of psychiatric or neurological disorders.Participants N.M.Trumpp et al. signed a written declaration of consent.The expense allowance for participation was 30 Euros or three course credits, respectively.The Ethical Committee of Ulm University approved all procedures and confirmed that they adhere to the tenets of the Declaration of Helsinki.

Stimuli for the lexical decision task (LDT)
Stimuli in the LDT consisted of 60 critical abstract words and 60 pseudowords, which served as distractors.Critical abstract words were subdivided into 30 words with a high proportion of introspective and internal mental state features (mental state (MST) concepts; Table S1; e. g., "thought", "expectation", "surprise", "desire") and 30 words with a high proportion of verbal association features (verbal association (VAS) concepts; Table S2; e.g., "ritual", "justice", "difference", "revolution").All these stimuli were already used in a previous ERP study (Kiefer et al., 2022).Classification of the critical abstract words was based on an earlier property listing study (Harpaintner et al., 2018), in which participants generated associations for 296 abstract concepts.These associations were subsequently categorized according to their verbal and modality-specific content (for further details see Harpaintner et al., 2018).For creating the 60 pseudowords, one vowel as well as one consonant was replaced in abstract words not used in the present experiment, resulting in meaningless yet pronounceable letter sequences (Table S3, e.g., "Reuchsum").Notice, pseudowords and critical abstract words were matched for word length.Matching was also performed between abstract MST and VAS concepts with regard to other relevant, but potentially confounding conceptual features (visual, acoustic, motor, tactile, gustatory, valence, arousal, concreteness/abstractness, familiarity) as well as with regard to psycholinguistic variables (age of acquisition, word length, lemma frequency, character bi-and trigram frequency).Conceptual feature norms, familiarity ratings, as well as valence, arousal and concreteness/abstractness ratings were drawn from the already mentioned property listing study (Harpaintner et al., 2018).Age of acquisition ratings were obtained from Kuperman et al. (2012).Lemma frequency, character bi-and trigram frequency were taken from the dlexDB database (Heister et al., 2011).Following the recommendations of Sassenhagen and Alday (2016), mean values, standard deviations and effect sizes of the matched conceptual and psycholinguistic variables of abstract MST and VAS concepts are shown in Table 1.Complementing these descriptive statistics, p-values of two-tailed unpaired t-tests are also listed.

Procedure
Participants performed a lexical decision task (LDT) in go/nogo response mode.The selected 60 abstract and 60 pseudowords were presented in white font on a black background in the center of a 32″ LCD screen (NordicNeuroLab AS, Bergen, Norway; 75 Hz refresh rate; 1280×720 pixels resolution), which was located behind the magnetic resonance imaging (MRI) scanner bore.Participants were able to see the presented stimuli via a mirror attached to the MRI head coil.Each trial of the LDT had a duration of 2400 ms starting with the presentation of a fixation cross for 500 ms.After the fixation cross, the target (word/ pseudoword) was displayed for 400 ms followed by a blank screen of 1500 ms.Before a new trial started, the screen remained blank for a variable intertrial interval lasting from 0 to 15600 ms with a mean of 3530 ms.At the time of target presentation, participants' task was to specify as quickly and accurately as possible, if the target was a meaningful word or a pseudoword.In case of a pseudoword (go trial) participants should press a button with their right index finger.In case of a meaningful word (no-go trial), participants should not respond.We employed a go/nogo response mode in order to avoid hand motor activity induced by the key press to interfere with potential motor activity elicited by conceptual processing of abstract words.The task was explained in detail during an instruction period outside the MR scanner and practiced (10 words, 10 pseudowords not used in the MR study) directly before the LDT experiment within the MR scanner.
The software program "Optseq2" (http://surfer.nmr.mgh.harvard.edu/optseq/ (see also Dale, 1999)) was used to create an initial sequence of trial order and onsets.This sequence was further modified to avoid a direct succession of the same condition more than four times and to obtain jittered onsets by randomly adding fractions of the fMRI repetition time.After all, stimuli were randomized within each condition and allocated to the trial sequence.The software program "Presentation 18.1″ (Neurobehavioral Systems Inc., San Francisco, USA) was used for stimulus presentation and recording of behavioral data (reaction time and error rate).

Localizer tasks
Following the LDT, we conducted three different localizer tasks to measure mental state, generative and automatic linguistic, and motor processing in the brain.Stimuli used in the localizer tasks were depicted on the LCD screen analogous to the LDT.
The mental state localizer task was developed based on the study by Dodell-Feder et al. (2011).Stimuli consisted of 20 stories taken from Dodell-Feder et al. (2011) and translated into German by one of the authors.Ten stories described false beliefs, the other 10 control stories described outdated photographs (for a complete list of belief and photograph stories, see Tables S4 and S5).The mental state localizer task c Item-scales: six-point Likert scale for concreteness/abstractness and familiarity ratings from 1 = "abstract"/"low familiarity" to 6 = "concrete"/"high familiarity"; six-point Likert scale for valence from − 3 = "negative" to +3 = "positive"; self-assessment manikins (Bradley and Lang, 1994) for arousal from 1 = "weak" to 5 = "strong".d according to Kuperman et al. (2012).e according to Heister et al. (2011).
N.M.Trumpp et al. consisted of 21 fixation blocks lasting 12 s each, and 20 experimental blocks, each lasting 14 s.Within an experimental block, one of the belief or photograph stories was presented on the screen for 10 s followed by a statement (4 s) which referred to the story.Participant's task was to read the story attentively and to specify whether the statement is true or false by pressing the appropriate button of a small keypad with the right hand (true → left button with index finger; false → right button with middle finger).Half of the statements in each category was true or false, respectively.Belief and photograph stories were randomized and alternately presented with fixation blocks.A fixation block consisted of 5 fixation crosses presented for 300 ms each with a clear screen of an average of 2100 ms in between (ranging from 1500 ms to 2700 ms).
Participants should attentively look at the fixation crosses.Fixation blocks were placed at the beginning and ending of the task.
The generative and automatic speech localizer task was designed following the study of Birn et al. (2010).It consisted of nine fixation blocks and eight experimental blocks lasting 24 s each.Experimental blocks were subdivided into four blocks requiring automatic speech and 4 blocks requiring generation of verbal associations.During automatic speech blocks, the words "months", "counting" and "weekdays" successively appeared on the screen in randomized order for 8 s each.Participant's task was to name the calendar months in chronological order starting with January, to count starting with one, and to name the weekdays starting with Monday, respectively, as long as the corresponding word was presented on the screen.During verbal association blocks, likewise, three words successively appeared on the screen for 8 s each.These words denoted 12 (3 words per verbal association block, assigned randomly) different categories: "animals", "plants", "colors", "tools", "fruits", "vegetables", "beverages", "vehicles", "clothes", "furniture", "games" and "instruments".Participants were asked to spontaneously name things that come to their mind when thinking of this category (as long as the corresponding word was presented on the screen).Participants were instructed to speak loudly and clearly.Blocks with automatic speech and generation of verbal associations were presented alternately and interleaved with fixation blocks.Fixation blocks in this task consisted of ten fixation crosses, which were presented for 300 ms each, interleaved with clear screens of 1753 ms to 2307 ms (M = 2100 ms).Again, participants should just attentively look at the fixation crosses.Analogous to the mental localizer task, fixation blocks were set at the beginning and ending of the task.
The motor localizer task was adopted from an earlier study (Harpaintner et al., 2020a) testing movements of the right and left hands and was extended with regard to four additional experimental blocks requiring lip movements.The motor localizer task consisted of 25 blocks, each lasting 24 s.Thirteen fixation blocks (identical to those in the verbal association localizer task) and 12 experimental blocks alternated, starting and ending with a fixation block.Experimental and fixation blocks were identical, except that fixation crosses were replaced by double arrows or letters, respectively.There were 4 blocks with arrows pointing to the left (<<), 4 blocks with arrows pointing to the right (>>), and 4 blocks, in which the letters "i" and "o" were presented alternately (5 times each).Participants task was to either press a hand training ball in their left (arrows pointing to the left) or right hand (arrows pointing to the right) ten times firmly and evenly in the rhythm of the flashing arrows, respectively, or to form the letters "i" and "o" quietly, but clearly with their lips.Movements should be limited to participants hands and mouth in order to avoid motion artifacts.Fixation and experimental blocks alternated in a fixed sequence (fixationright handfixationleft handfixationlipsfixationright handfixationleft handfixationlips -… -fixation).
All tasks were explained in detail before participants entered the MR Scanner.The mental state and generative and automatic speech localizer task were additionally practiced (one belief, one photograph story/one verbal association, one automatic speech block, not used in the MR experiment) directly before the corresponding experiment within the scanner.

Post-scan ratings
In order to confirm the distinct relevance of conceptual content related to mental states and verbal associations for the abstract MST and VAS concepts used in the LDT during fMRI, participants performed a corresponding rating study immediately after scanning.These post-scan ratings were performed using a paper-and-pencil-questionnaire containing the 30 MST and 30 VAS concepts used in the LDT as well as abstract filler concepts.The filler concepts, which were also drawn from the above mentioned property listing study (Harpaintner et al., 2018), had a medium relevance of internal state and verbal association properties.After written and verbal instruction (see supplementary material), participants were asked to rate each concept regarding the extent to which this concept reflects internal states and the extent of verbal associations related to this concept on a six-point Likert scale (1 = "no/very few internal states/verbal associations", 6 = "many internal states/verbal associations"), respectively.Internal states thereby include internal cognitive processes like motivation, volition and introspection, as well as thoughts, emotions and evaluations.Verbal associations were defined as other thematically or symbolically related words which come to one´s mind spontaneously when thinking of the given concept.Order effects were avoided by preparing three different versions of the questionnaire with a randomized sequence of the abstract concepts.Two-tailed paired t-tests on mean scores of the internal state and verbal association ratings for MST and VAS concepts were used to assess statistical differences.

MRI data acquisition
A 3 Tesla MAGNETOM Prisma and a 64 channel head/neck coil from Siemens AG in Erlangen, Germany, was used for MRI data acquisition.The T2*-weighted blood oxygenation level-dependent (BOLD) signal was measured to obtain functional images based on the specific task.Echo-planar (EPI) pulse sequences were used with the following parameters: 2000 ms repetition time (TR), 34 ms echo time (TE), flip angle of 90 • , bandwidth of 2268 Hz/Px, PAT factor 2 (GRAPPA mode), mm field of view (FOV), matrix size: 76 x 76, 33 transversal slices in ascending order, 3.5 mm slice thickness, interslice gap of 0.88 mm and 2.53 mm x 2.53 mm x 4.38 mm voxel size.For each task, there was a different number of EPI volumes (and scan time): LDT: 368 volumes (12.3 min), mental localizer: 276 volumes (9.2 min), verbal association localizer: 213 volumes (7.1 min), motor localizer: 305 volumes (10.1 min).In a final step, high resolution T1*-weighted structural images were acquired using a magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence with a TR of 2300 ms, a TE of 2.32 ms, 900 ms inversion time, a flip angle of 8 • , 200 Hz/Px bandwidth, PAT factor 2 (GRAPPA mode), 240 mm FOV, matrix size: 256 x 256, 0.90 mm x 0.94 mm x 0.94 mm voxel size, sagittal slice orientation and a scan time of 5.4 min.

Image preprocessing and statistical analysis
Statistical Parametric Mapping 12 (SPM12, Wellcome Department of Cognitive Neurology, London, UK) running on MATLAB 2019b (The MathWorks Inc., Natick, MA, USA) was used to preprocess and statistically analyze the fMRI data.Normalization of fMRI data and T1 images was performed with the Computational Anatomy Toolbox (CAT12.8.1, Version 2043, Gaser et al., 2022).In the LDT run, the first four EPI volumes were discarded subject-wise.In the runs of the mental state, generative and automatic speech, and motor localizer tasks, the first 5, 6, and 6 EPI volumes, respectively, were discarded subject-wise.In a first preprocessing step, the individual T1 images were segmented using the CAT12 toolbox (default settings, except for 1 mm voxel size) resulting in bias, noise and global intensity corrected T1 images and tissue class images (gray matter, white matter and cerebrospinal fluid) in native and normalized (Montreal Neurological Institute, MNI) space.Additionally, forward deformation fields (native -> normalized) were produced.Functional MRI data of the lexical decision task were first slice time corrected (reference slice: 17).Then, EPI images of each run N.M.Trumpp et al. (LDT, mental state, generative and automatic speech, and motor localizer tasks) were spatially realigned to their run-specific mean EPI image.Resulting mean functional images of the localizer runs (mental state, generative and automatic speech and motor localizer tasks) were co-registered to the mean EPI image of the LDT.Thereafter, the mean EPI image of the LDT was co-registered to the individual T1 image in native space produced by the CAT12 toolbox.Afterwards, the co-registration matrix was applied to all other functional MR images, and the forward deformation fields, also produced by CAT12, were used to MNI-normalize the functional images.Preprocessing ended with resampling (2 mm x 2 mm x 2 mm) and smoothing (Gaussian kernel with 8 mm full width at half maximum) of the functional MR images.
For each task, a separate first-level General Linear Model (GLM) was set up using the preprocessed functional MR images.Main effects of the LDT were modeled by coding the onsets of correct pseudoword, MST and VAS word trials in three regressors with a duration of 0 s, each.All incorrect trials were modeled in an additional fourth regressor.The design matrix also included the spatial realignment parameters.Resulting stick-functions were convolved with the canonical hemodynamic response function (HRF) and its time derivative.A high-pass filter with a cutoff at 128 s was used to remove low-frequency drifts of the scanner.Time-correlated residual errors were accounted for using a first-order autoregressive model (AR1).Main effects of each task condition (pseudowords, MST and VAS versus implicit baseline, respectively) were tested with first-level t-contrasts.Resulting contrast images from each participant were further analyzed in a second-level randomeffects model with flexible factorial design (subjects as random factor).Contrasts of interest were MST > VAS and VAS > MST.A voxel-height threshold of p < .001(uncorrected) was applied, family-wise error rate (FWE)-corrected at cluster level with a threshold of p < .05,requiring at least 160 contiguous voxels per cluster.
With regard to the mental state and generative and automatic speech localizer tasks two regressors of interest coded the onsets and durations (14 s/24 s) of the experimental blocks (false belief and false photograph stories/automatic speech and generating verbal associations).For the motor localizer task, three regressors of interest coded the onsets and durations (24 s) of the right hand, left hand, and lip movement blocks.Resulting boxcar functions were convolved with the canonical HRF.The design matrix again included the realignment parameters, and a highpass filter and an AR1 model was applied.First-level t-contrasts testing for the main effect of each condition (versus implicit baseline) in each task were set up.Resulting contrast images from all participants were subjected to separate, task specific second-level random-effects analyses with a flexible factorial design (subjects as random factor).Contrasts of interest were "belief > photograph stories" for the mental state localizer task, "automatic speech > generating verbal associations" and "generating verbal associations > automatic speech" for the generative and automatic speech localizer task and "hands > lips" and "lips > hands" for the motor localizer task.A threshold of p < .05,FWEcorrected was applied to each statistical parametric map.
The potential overlap between significant brain activation during the LDT and activation during the three localizer tasks was of theoretical interest.Therefore, the statistical parametric maps (p < .05,FWEcorrected) from the localizer tasks were saved as binary images to use them as specific masks.Contrasting brain activation during VAS versus MST abstract concept processing in the LDT did not show any significant effects at cluster-level.Thus, only the contrast of MST versus VAS abstract concepts was inclusively masked by the belief > photograph stories mask of the mental localizer task, the automatic speech > generating verbal associations and generating verbal associations > automatic speech masks of the verbal localizer task and the hands > lips and lips > hands masks of the motor localizer task.

Behavioral data of the LDT
Since the LDT was designed as a go/nogo task, reaction time (RT) data was only available for pseudowords (mean RT = 784 ms, SD = 137 ms).Regarding error rates (ER), participants failed to press the button in 0.82 % (SD = 1.77 %) of pseudoword trials, false alarms for MST abstract concepts occurred in 2.48 % (SD = 3.07 %) and for VAS abstract concepts in 3.69 % (SD = 3.63 %) of real word trials.In a repeated measures ANOVA on mean ER, a significant main effect of stimulus category (F(2,92) = 15.80,p < .001) was found, which was, however, based on significant differences between pseudoword and MST/VAS word trials (Bonferroni post hoc-tests: p < .01).Mean ER to MST and VAS word trials did not differ significantly (Bonferroni post hoc-tests: p = .063).

LDT.
Contrasting brain activation associated with processing of VAS versus MST abstract concepts did not reveal any significant effects at the applied statistical threshold.Greater activation for MST compared to VAS abstract concepts (see Fig. 1 and Table 2, as well as contrast estimates depicted in Fig. S1) was observed in occipito-parietal, temporal, frontal and subcortical regions.Activity increases in occipitoparietal brain regions included the bilateral calcarine gyrus, left lingual gyrus, bilateral precuneus, bilateral inferior parietal gyrus, right cuneus and right superior occipital gyrus, as well as left superior/middle occipital gyrus and right supramarginal gyrus.Increased activity to MST concepts in (fronto-) temporal areas comprised the left superior/middle temporal gyrus, left superior temporal gyrus reaching into superior temporal pole and insula and right superior/middle/inferior temporal gyrus as well as the TPJ.Activity increases in frontal brain regions encompassed bilateral superior frontal gyrus, precentral and left postcentral gyrus, right middle frontal gyrus as well as right middle/posterior cingulate cortex.Activity increases in subcortical regions were obtained in the right basal ganglia (putamen and nucleus caudatus) as well as in the bilateral cerebellum.(For an overview of significant brain activation to MST concepts relative to the implicit baseline, see Fig. S2a and Table S6, and to VAS concepts against the implicit baseline, see N.M.Trumpp et al.Fig. S2b and Table S7.)

Mental state localizer task.
As illustrated in Fig. 2a and Table S8, processing false belief stories, compared to false photograph stories, was associated with grater activation in a network of occipitoparietal, temporal and frontal brain regions.Increased occipitoparietal activity to false beliefs involved the bilateral precuneus, the bilateral calcarine gyrus, the right lingual gyrus, bilateral fusiform gyrus and the left middle occipital gyrus.Activity in temporal areas comprised bilateral superior/middle temporal gyrus and the right inferior temporal gyrus.Increased activity in frontal regions included bilateral inferior frontal gyrus pars orbitalis, right inferior frontal gyrus pars triangularis, bilateral superior/middle frontal gyrus, bilateral SMA and left precentral gyrus.Further activation was found in the hippocampus, the parahippocampal gyrus, the bilateral cerebellum and the thalamus.

Generative and automatic speech localizer task. Brain regions
showing greater activation to generation of verbal associations compared to automatic speech (see Fig. 2b and Table S9) were found in a large cluster in the frontal lobe including left SMA, left middle frontal gyrus, left inferior frontal pars opercularis/triangularis, bilateral anterior/middle cingulate cortex, left insula, left nucleus caudatus and right hippocampus.Furthermore, greater activation to the generation of verbal associations was found in left calcarine gyrus, left medial superior frontal gyrus, left inferior temporal gyrus and left superior occipital gyrus as well as in the right cerebellum and in the left thalamus.Greater activation for automatic speech compared to generation of verbal associations (see Fig. 2c and Table S10) was found in a parieto-frontotemporal cluster in both hemispheres including supramarginal gyrus, inferior parietal gyrus, precuneus, superior/middle temporal gyrus, middle cingulate cortex, left rolandic operculum, left postcentral gyrus, left insula and left angular gyrus.Additional significant activation associated with automatic speech processing was found in right inferior frontal pars triangularis/orbitalis, left precentral gyrus, bilateral olfactory gyrus, left middle/inferior temporal gyrus, right superior frontal pars orbitalis, left cerebellum and right middle frontal gyrus.

Motor localizer task.
Comparing hand versus lip movements revealed greater activation in bilateral pre-and postcentral gyrus, bilateral cerebellum, bilateral olfactory gyrus and right gyrus rectus and in right superior/middle/inferior occipital gyrus (see Fig. 2d and Table S11).Greater activation for hand movements was also found in bilateral rolandic operculum, left middle cingulate cortex, left SMA and left hippocampus, as well as in right superior frontal gyrus, right calcarine gyrus and right posterior cingulate cortex.In contrast, comparing lip versus hand movements (see Fig. 2e and Table S12) revealed greater activation in a large fronto-temporal cluster in each hemisphere including pre-and postcentral gyrus, middle/superior temporal gyrus, middle frontal gyrus, inferior frontal pars orbitalis, rolandic operculum, middle cingulate cortex and thalamus, as well as right SMA, right putamen, left insula and left inferior frontal pars opercularis.Other clusters with greater activation to lip movements were found in parieto-temporal cortex (inferior parietal gyrus, supramarginal gyrus and superior temporal gyrus), right superior/middle frontal gyrus and bilateral cerebellum.Comparing Fig. 2d and 2e, activation of lip movements in bilateral pre-and postcentral gyrus was more anterior and inferior than activation of hand movements (cf. the motor and sensory homunculus, respectively).

Spatial overlap between activity to MST abstract concepts in the
LDT and the localizer tasks.Brain activity elicited during processing of MST versus VAS abstract concepts in the LDT significantly overlapped with brain activity in the mental localizer task (false belief vs. false photograph stories) in bilateral occipito-parietal cortex including calcarine gyrus, lingual gyrus, cuneus and precuneus as well as in the right superior/middle temporal gyrus including TPJ (see Figs. 3a, S3a and Table S13).
Overlap between activity to MST abstract concepts in the LDT and generation of verbal associations was restricted to right cerebellum (see Figs. 3b, S3b and Table S14).MST abstract concept processing in the LDT and the processing of automatic speech elicited overlapping activation in the bilateral precuneus, right cuneus, right superior/middle occipital gyrus, right superior/inferior parietal gyrus, right middle/ posterior cingulate cortex and bilateral (fronto-)temporal cortex (left superior temporal gyrus including superior temporal pole and insula, right superior/middle/inferior temporal gyrus and right supramarginal gyrus) (see Figs. 3c, S3c and Supplementary Table S15).
Significantly overlapping activation to MST abstract concepts and lip movements was found in bilateral superior and right middle frontal gyrus, left inferior parietal gyrus and bilateral precentral gyrus (see Figs. 3d, S3d and Supplementary Table S16).Activity to MST abstract concepts in the LDT and hand movements in the motor localizer task did not significantly overlap.

Discussion
In the present fMRI study, we investigated the functional neuroanatomy underlying processing of abstract MST and VAS concepts during a lexical decision task to elucidate the grounding of abstract concepts in different modal experiential brain circuits.More specifically, we tested, whether processing of abstract MST concepts would involve brain circuits spatially overlapping with activity patterns associated with the processing of mental states, generating verbal associations, automatic speech, hand movements and lip movements as defined by the corresponding localizer tasks.MST concepts are highly related to introspective and internal mental state features.As their meaning might also be partly acquired during the course of social interaction and verbal communication with other individuals requiring articulation, we expected neural activation in mentalizing (Conca et al., 2021;Kiefer et al., 2022) and face-motor brain regions (Borghi and Zarcone, 2016;Borghi, 2022;Dreyer and Pulvermüller, 2018;Fini et al., 2022;Ghio et al., 2013) according to the mental state and motor localizer tasks.VAS concepts, instead, are strongly linked to verbal association properties and should thus activate the visual word form area involved in verbal association processing (Kiefer et al., 2022) spatially overlapping with activity during generation of verbal associations.
At a behavioral level, analysis of the rating study revealed that abstract MST concepts were more likely learned through experience, whereas VAS concepts were more likely learned through language.Importantly, although the relevance of conceptual motor features for abstract MST and VAS concepts was equated in general, we found significant differences regarding the relevance of mouth vs. hand actions associated with MST and VAS concepts: abstract MST compared to VAS concepts showed a higher involvement of mouth and a lower involvement of hand movements.In line with previous findings (Borghi and Zarcone, 2016;Borghi, 2022;Dreyer and Pulvermüller, 2018;Fini et al., 2022;Ghio et al., 2013) this result indicates a particular relevance of mouth actions for establishing the meaning of MST concepts.
At a neural level, processing of MST and VAS concepts evoked different activation patterns: Greater activation for MST compared to VAS concepts was found in occipito-parietal (calcarine gyrus, left lingual gyrus, precuneus, inferior parietal gyrus, right cuneus and right superior occipital gyrus, as well as left superior/middle occipital gyrus and right supramarginal gyrus), (fronto-) temporal (left superior/middle temporal gyrus, left superior temporal gyrus reaching into superior temporal pole and insula and right superior/middle/inferior temporal gyrus including TPJ), and frontal brain regions (superior frontal gyrus including precentral and left postcentral gyrus and right middle frontal gyrus), as well as in right middle/posterior cingulate cortex, right basal ganglia, and cerebellum.This result largely confirms findings of our previous ERP study (Kiefer et al., 2022), which found estimated brain electrical sources for MST concepts in brain areas related to visuo-motor processing (precentral and postcentral cortex, superior parietal cortex) and mentalizing (TPJ, precuneus, insula, lateral/medial prefrontal cortex).
Activation of occipital visual brain regions during processing of MST concepts may reflect simulations of visual features belonging to specific situations associated with these MST concepts (Ulrich et al., 2022).Such activation in visual brain regions has already been found in previous studies associated with mentalizing (Fehlbaum et al., 2022;Ulrich et al., 2022).Furthermore, according to Fehlbaum et al. (2022), inferior parietal lobule, as part of the mirror system, is involved in imitations of actions needed to adapt to social situations.In addition, Fehlbaum et al. specified medial prefrontal cortex (mPFC), precuneus, TPJ, middle/inferior frontal gyrus, insula, occipital pole, middle temporal gyrus (posterior superior temporal sulcus/gyrus (pSTS/pSTG), temporal pole) and anterior/posterior cingulate cortex to be involved in mentalizing.In particular the pSTS, precuneus, TPJ, the temporal poles and medial prefrontal cortex have consistently been found to be activated in mentalizing tasks (Arioli and Canessa, 2019;Fehlbaum et al., 2022;Frith andFrith, 2006, 2003), specifically ToM (Dodell-Feder et al., 2011;Saxe and Kanwisher, 2003;Saxe and Powell, 2006).The precuneus has been associated with self-referential thought (Abu-Akel and Shamay-Tsoory, 2011;Cabeza and St Jacques, 2007).Posterior STS and especially TPJ support social cognition and perspective taking (Martin et al., 2020;Seymour et al., 2018;Wang et al., 2016), whereas the temporal poles provide access to social knowledge and general knowledge about the world (Frith andFrith, 2006, 2003).These processes are important for the ability to mentalize (Frith and Frith, 2006).Furthermore, TPJ as well as precuneus are assumed to provide input for mentalizing brain regions in prefrontal cortex (Abu-Akel and Shamay-Tsoory, 2011).Posterior Table 2 Lexical decision task.Brain regions showing greater activation to MST in comparison to VAS abstract concepts at a voxel-height threshold of p < .001,family-wise error rate (FWE)-corrected at the cluster level (p < .05),corresponding to k = 160 contiguous voxels.Shown are peak voxels with highest tvalues for significant clusters and their local maxima more than 8 mm apart.For clusters comprising more than 5000 voxels, embedded local maxima more than 4 mm apart are listed in order to provide a more detailed characterization of the activated brain regions.Abbreviations: L: left; R: right; MNI: Montreal Neurological Institute.cingulate cortex, right premotor cortex and left cerebellum (as well as dorsal/ventral medial prefrontal cortex (d/vmPFC), posterior middle temporal cortex (pMTC), pSTS, TPJ and amygdala) subserve neural processing of social interactions as a joint involvement of the action observation and mentalizing brain networks (Arioli and Canessa, 2019).Activation of areas associated with action meaning (pMTC, pSTS and right premotor cortex) might thereby reflect visuo-motor decoding of shared motor goals in social interactions (Arioli and Canessa, 2019).Spatial overlap of MST concept activation with activation during processing of false belief compared to false photograph stories in the mental localizer task was found in bilateral occipito-parietal (calcarine gyrus, lingual gyrus, cuneus and precuneus) and right temporal brain regions including TPJ.Thus, in line with our expectations, processing of MST concepts activate modal brain regions, which are highly involved in mental processing (Conca et al., 2021;Kiefer et al., 2022) including ToM, social interaction and perspective taking, but also in the sensory-motor brain network again reflecting simulation of perceptual features and actions associated with social interactions.
Spatial overlap of activation to MST concepts with activation during generation of verbal associations was confined to the right cerebellum.The right cerebellum has previously been found to be active during linguistic tasks like word generation (Drager et al., 2004) and has been related to implicit motor movements associated with articulation.Spatial overlap of increased activity to MST concepts with activity during automatic speech was found in bilateral occipito-parietal (bilateral precuneus, right cuneus, right superior/middle occipital gyrus, right superior/inferior parietal gyrus) and temporal (left superior temporal gyrus including insula, right superior/middle/inferior temporal gyrus and right supramarginal gyrus) brain regions as well as in right posterior/middle cingulate cortex.Overlapping brain activity was more extensive in the right hemisphere consistent with Birn's (2010) findings of a more right hemispheric activation to automatic speech.The left anterior insula is known to be responsible for planning of articulation and left STG (specifically posterior planum temporale/ventral supramarginal gyrus) is involved in subvocal articulation (Price, 2010).In line with this finding, Borghi and Fernyhough (2023) recently pointed out that (articulated) inner speech, like rehearsing and inwardly repeating of word meaning, is used to consolidate abstract concepts in memory.In parts, especially within the precuneus, overlapping activation for automatic speech and MST concepts is typical for mentalizing processes (see above).The involvement of these parts of the mentalizing brain network active during automatic speech was not expected.We suggest that automatic speech, in our case naming of overlearned words (months, days of the week and numbers), might be such an easy task, that participants were engaged in mind wandering, which involves mentalizing.This assumption is supported by a study of Mason et al. (2007), who showed that a reduction of processing demands was accompanied with increased activity in the default mode network (DMN) reflecting mind wandering.This overlap with the automatic speech localizer indicates that processing of abstract MST concepts not only activates the mentalizing brain network, but also recruits brain regions related to linguistic processing (Borghi et al., 2017;Borghi and Zarcone, 2016).
In line with the results of the present rating study showing higher relevance of mouth actions for MST concepts, overlapping activation for MST concepts and the motor localizer task was only found for lip, but not for hand movements in bilateral superior and right middle frontal gyrus, left inferior parietal gyrus and bilateral precentral gyrus.The present ratings also revealed that abstract MST compared to VAS concepts to be more likely learned through experience possibly resulting in stronger activations of modal brain regions in general.Although not specified in the ratings, these experiences could include interactions with others to clarify the meaning of abstract MST concepts, in accordance with the high relevance of mouth actions for this abstract word category (Borghi et al., 2017;Borghi and Zarcone, 2016).Our results are in agreement with the study of Dreyer at al. (2018), who showed that abstract mental words elicit activation preferences in face over hand motor areas during passive reading.Furthermore, Ghio et al. (2013) demonstrated in a rating study that mental states and emotions are more associated with mouth actions, in contrast to number concepts which had stronger associations with hand actions.According to Borghi and colleagues (Borghi et al., 2017;Borghi and Zarcone, 2016), abstract concepts including mental state concepts are mainly acquired through social interaction and verbal communication with others.In addition, as already discussed above, acquisition and use of abstract concepts might be accompanied by inner speech to clarify their meaning (Borghi and Fernyhough, 2023).Both include speech and articulatory processes involving movements of the mouth.In a behavioral study, they demonstrated that an advantage of a response given by the hand over a response given by the mouth is less pronounced with abstract than with concrete target words (Borghi and Zarcone, 2016).Furthermore, Fini et al. (2022) showed that articulatory suppression slows processing of abstract words more than of concrete ones, suggesting a functional role of the mouth motor system in accessing word meaning of abstract concepts.Silent articulation of the letters "i" and "o" in the lip motor localizer task may also correspond to emotional mimics (laughing/astonishment) which are typically represented in the right hemisphere.Thus, emotional processing might contribute to activation of MST concepts particularly in the right sensorimotor cortex overlapping with lip movements.
VAS compared to MST concepts, however, did not show any significant activity increases.Thus, contrary to our expectation, we did not find activation differences in visual occipital brain regions involved in processing of verbal associations (Kiefer et al., 2022) for VAS compared to MST concepts.This suggests that verbal associations, as defined in the property listings (Harpaintner et al., 2018) used for stimulus selection as well as shown in the post-scan ratings, play an equal role for both VAS and MST concepts.Furthermore, our post-scan ratings revealed that the relevance of verbal associations differed only little between MST and VAS concepts (M MST = 3.41, M VAS = 3.57, difference 0.16, d Diff = 0.33).Rating differences concerning internal states, in contrast, were much larger (M MST = 3.92, M VAS = 3.44, difference 0.48, d Diff = 0.82).The absence of significant activity increases for VAS compared to MST concepts might also result from a more heterogeneous content of VAS concepts.VAS concepts are characterized by verbal associations as identified in the property listings (Harpaintner et al., 2018).Verbal associations, thus, are rather accessory, with an indirect syntagmatic relation to the content of the concept (e.g., the verbal association "cloverleaf" for the abstract concept "luck").Furthermore, the semantic content of the associated words was not further specified and is probably highly heterogeneous.This might have resulted in variable activity patterns across the words of the VAS category, which prevented to detect systematic activity increases to VAS concepts relative to MST concepts.Please also note that our focus rested on abstract MST concepts, whereas abstract VAS concepts mainly served as a conceptual comparison category.
Although the different localizer tasks in the present study allowed us to functionally characterize the experiential brain areas contributing to processing abstract mental state concepts, fMRI measurements only provide correlative evidence.A causal relevance of identified neural brain circuits for processing MST concepts thus needs to be verified in studies using transcranial magnetic stimulation (Kuhnke et al., 2020a) or investigation of brain-lesioned patients (Dreyer et al., 2015;Trumpp et al., 2013).
In conclusion, the present fMRI study revealed that processing of abstract MST compared to VAS concepts elicits greater activation in occipito-parietal, temporal and frontal brain regions as well as in cingulate cortex, basal ganglia, and cerebellum, spatially overlapping with activation during mentalizing, generation of verbal associations, automatic speech, and lip movements.MST concepts activate the mentalizing brain network, complemented by perceptual-motor brain regions involved in simulation of visual or action features associated with social interactions, linguistic brain regions as well as face-motor brain regions recruited for articulation.In accordance with the grounded cognition theories, the present results demonstrate that abstract mental state concepts are grounded in experiential brain systems related to mentalizing, perceptual-motor processing, including mouth actions, and automatic speech.

Fig. 1 .
Fig. 1.Greater brain activation for abstract mental state (MST) compared to abstract verbal association (VAS) concepts in the lexical decision task (LDT) depicted on 2D slices (left) and a 3D whole brain view (right; created with MRIcroGL(Rorden and Brett, 2000)).For the 3D view, statistical parametric maps were rendered to display brain activation from a right-hand side perspective.Clusters near the right hemisphere's cortical surface are shown in vivid colors, while those closer to the left hemisphere's surface appear in fading color, reflecting their decreased visibility due to the viewing angle.Participants' mean T1 image, computed with CAT12(Gaser et al., 2022), served as background image for superimposition of the statistical parametric maps.Voxel-height threshold was set to p < .001(uncorrected) with an extent threshold of k = 160 voxels achieving family-wise error rate (FWE) correction at cluster level with p < .05.The specified coordinates refer to Montreal Neurological Institute space.The color bar visualizes t-values.

Fig. 2 .
Fig. 2. Significant brain activation for the different localizer tasks at a statistical threshold of p < .05,FWE-corrected depicted on 2D slices (left) and 3D whole brain views (right).a) Greater activation to false belief compared to false photograph stories in the mental state localizer task.b) Greater activation to generative compared to automatic speech in the generative and automatic speech localizer task.c) Greater activation to automatic compared to generative speech in the generative and automatic speech localizer task.d) Greater activation to hand compared to lip movements in the motor localizer task.e) Greater activation to lip compared to hand movements in the motor localizer task.Statistical parametric maps were overlaid on participants' mean T1 image computed with CAT12 (Gaser et al., 2022).The designated coordinates correspond to the Montreal Neurological Institute space.The color range bars indicate t-values.

Fig. 3 .
Fig. 3. Greater activation to MST compared to VAS abstract concepts in the LDT (depicted in blue) overlaid with activation during the different localizer tasks (depicted in red, except for hand movements, which are depicted in green).Overlapping activation is depicted in yellow.(a) Greater activation to MST compared to VAS concepts overlaid with activation during processing of false belief compared to false photograph stories.(b) Greater activation to MST compared to VAS concepts overlaid with activation during generative compared to automatic speech.(c) Greater activation to MST compared to VAS concepts overlaid with activation during automatic compared to generative speech.(d) Greater activation to MST compared to VAS concepts overlaid with activation during hand compared to lip movements and lip compared to hand movements.Statistical parametric maps were overlaid on participants' mean T1 image computed with CAT12 (Gaser et al., 2022).Coordinates are in Montreal Neurological Institute space.

Table 1
Conceptual and psycholinguistic variables for MST and VAS abstract concepts.Shown are mean values and standard deviations (in parenthesis) for variables with a proportion larger than 0.01, as well as effect sizes and p-values resulting from the comparison of MST and VAS word sets.MST: mental states; VAS: verbal associations.