The cortical representation of transitivity: Insights from tractography-based inhibitory nTMS

Navigated Transcranial Magnetic Stimulation (nTMS) is commonly used to causally identify cortical regions involved in language processing. Combining tractography with nTMS has been shown to increase induced error rates by targeting stimulation of cortical terminations of white matter fibers. According to functional Magnetic Resonance Imaging (fMRI) data, bilateral cortical areas connected by the arcuate fasciculus (AF) have been implicated in the processing of transitive compared to unergative verbs. To test this connection between transitivity and bilateral perisylvian regions, we administered a tractography-based inhibitory nTMS protocol during action naming of finite transitive ( The man reads ) and unergative ( The man sails ) verbs. After tracking the left and right AF, we stimulated the cortical terminations of the tract in frontal, parietal and temporal regions in 20 neurologically healthy native speakers of German. Results revealed that nTMS induced more errors during transitive compared to unergative verb naming when stimulating the left (vs right) AF terminations. This effect was specific to the left temporal terminations of the AF, whereas no differences between the two verb types were identified when stimulating inferior parietal and frontal AF terminations. Induced errors for transitive verbs over left temporal terminations mostly manifested as access errors (i.e., hesitations). Given the inhibitory nature of our nTMS protocol, these results suggest that temporal regions of the left hemisphere play a crucial role in argument structure processing. Our findings align with previous data on the role of left posterior temporal regions in language processing and by providing further evidence from a language production experiment using tractography-based inhibitory nTMS.


Introduction
Navigated Transcranial Magnetic Stimulation (nTMS) is increasingly used as a non-invasive tool to locate language-relevant cortical regions in the brain (Picht et al., 2013;Sollmann et al., 2015).Language mapping with nTMS can be used in neurologically healthy individuals as well as in people with brain tumors within the preoperative planning of an awake surgery (Hauck et al., 2015;Krieg et al., 2017;Ohlerth et al., 2021).Even though action naming is increasingly being used as part of the preoperative protocol, it remains unclear which linguistic variables of verbs affect the results of language mapping with nTMS.Regarding theoretical purposes, although functional Magnetic Resonance Imaging has been largely used to investigate verb processing, nTMS has only recently started to be used for the same purpose (for a review see Ntemou et al., 2023).As a result, from a clinical perspective, the objective of the present study is to investigate whether the number of arguments a verb carries (i.e., transitivity) affects the outcome of language mapping with nTMS.From a theoretical perspective, we aim to provide insights regarding the anatomical representation of transitive and unergative verbs in the left (language-dominant) hemisphere by virtue of tractography-based repetitive nTMS.

Transitivity: theoretical background
Models of sentence production consist of components such as message conceptualization, lexical selection, sentence building, and phonological encoding (Bock and Levelt, 1994;Caramazza, 1997;Levelt et al., 1999).Verb argument structure plays a crucial role in sentence formation as the verb requires a certain number of arguments to be formulated and assigns grammatical functions to these arguments.For example, the verbs to cough and to feed differ in the number of arguments they require.
b.She is feeding the horse.c. *She is coughing the horse.d. *She is feeding.e.She is reading or She is reading the newspaper.
The verb to cough in 1a requires one argument, namely the agent to which it assigns nominative case (i.e.,She instead of her).However, the verb to feed in 1b requires two arguments, namely an agent and a theme.The agent is assigned nominative case and serves as the subject, whereas the theme receives accusative case and serves as the object.Languages such as English denote the function of the object via the post-verbal position, but languages such as German require overt inflection of the noun in the object position (i.e., Sie pflanzt den Baum: She plants theaccusative tree).Differences between verbs with variable argument structures become apparent when considering that to cough forms nongrammatical sentences when produced with an object (see 1c).The absence of an object, however, does not render all utterances ungrammatical.Unlike obligatory transitive verbs, pseudotransitive verbs such as in 1e, do not always require the overt production of an object.Regardless of the overt expression of an object, linguists argue that verbs such as "to read" require the specification of a second argument in the lemma as it is part of their meaning (Levin, 2006;Rappaport Hovav and Levin, 2002).This distinction is relevant as the transitive stimuli in the present work were essentially pseudotransitive (henceforth referred to as transitive verbs).

Neurocognitive evidence & models
The processing of transitive and unergative verbs has been previously investigated within the neuropsychological literature.Studies of individuals with post-stroke aphasia have reported the existence of fewer transitive verbs during narrative production, but also poor performance with verbs with more than one arguments (i.e., transitive, ditransitive) during overt action naming (De Bleser and Kauschke, 2003;Jonkers, 2000;Luzzatti et al., 2002;Malyutina et al., 2016;Thompson, 2003).The difficulties of individuals with post-stroke aphasia motivated later work that employed neuroimaging methods and examined the same topic.
Neuroimaging studies that focused on sentence comprehension contrasted sentences with verbs having more arguments compared to verbs that carry fewer arguments as well as verbs with varying number of subcategorization options (i.e., the type of phrases that can be complements of a verb; Ben-Shachar et al., 2003;Shetreet et al., 2007).Early studies connected the number of arguments in a sentence with activation in the posterior superior temporal gyrus (STG; Ben-Shachar et al., 2003).However, later studies using sentential contexts revealed that verbs with increased subcategorization options (e.g., transitives) mainly activate the posterior temporal lobe (Shetreet et al., 2007;Shetreet and Friedmann, 2012).Studies that investigated single word processing by using tasks such as lexical decision or action naming, however, reported slightly different cortical regions for verbs with more arguments.fMRI studies of healthy individuals performing lexical decision of verbs with one, two, and three arguments showed that transitive infinitives generate more activation in the left and right supramarginal and angular gyri (SMG and AnG respectively; Thompson et al., 2007Thompson et al., , 2010)).Production of transitive and unergative infinitives also revealed left and right inferior parietal correlates, as well as activation of the left inferior frontal gyrus during transitive versus unergative naming (IFG; den Ouden et al., 2009).
fMRI findings in combination with data from people with post-stroke aphasia led to the formulation of a neurocognitive model of argument structure (Thompson and Meltzer-Asscher, 2014).During language production, argument structure information is accessed in bilateral inferior parietal regions (i.e., SMG & AnG) that in turn kickstarts processes of phrase structure building in inferior frontal regions (Thompson and Meltzer-Asscher, 2014).Integration of semantic and syntactic information of the verb and its arguments then takes place in posterior temporal regions (i.e., posterior STG/MTG), meaning that sentence constituents are parsed, relationships between them are identified, and meaning is assigned (Thompson and Meltzer-Asscher, 2014).
The representation of argument structure information in bilateral inferior parietal regions, however, has been questioned by more recent neuroimaging data.By contrasting verb phrases (VPs) with noun phrases (NPs) with the same lexical material (frightened the boy vs the frightened boy), Matchin et al. (2019) argued that this comparison isolates argument structure processing by keeping semantic load constant.Increased activation for VPs compared to NPs was recorded only in the left posterior temporal lobe as well as the IFG (Matchin et al., 2019).Additionally, right hemisphere activity was absent for VPs over NPs.Similarly to previous work (Ben-Shachar et al., 2003;Malyutina & den Ouden, 2017;Shetreet et al., 2007), Matchin et al. (2019) concluded that posterior temporal effects represent processing of structural information of argument structure.Such findings led to the formulation of another neurocognitive model of syntax that includes the computation of hierarchical information as part of the posterior middle temporal gyrus/temporal sulcus (pMTG; Matchin and Hickok, 2020).According to Matchin and Hickok (2020), the AnG is associated with conceptual processing of events.This suggests that the AnG is engaged for the processing of events and their attributes (e.g., participating entities) in space and time but not for the grammatical properties of lemma level information such as argument structure (for more on event semantics, see Binder and Desai, 2011).As Matchin et al. (2019) also indicate, it is challenging to determine which exact entities constitute the attributes of an event.For example, examining the event of "someone chopping vegetables on a kitchen counter" involves "an entity chopping" and "something being chopped".However, it is unclear whether "a knife", "the counter" or even "the space of a kitchen" would be essential entities of the event as well.
Neurostimulation studies have also provided evidence that sentences with transitive verbs engage left cortical regions more so than sentences with unergatives verbs (Finocchiaro et al., 2015(Finocchiaro et al., , 2021;;Ntemou et al., 2021;Vercesi et al., 2020;Ward et al., 2022).The superior parietal lobe as well as the inferior parietal sulcus have been shown to be causally involved in comprehension of reversible transitive sentences as well as the production of transitive infinitive verbs (Finocchiaro et al., 2015;Ntemou et al., 2021;Vercesi et al., 2020;Ward et al., 2022).

Subcortical connectivity of perisylvian regions
White matter pathways connect regions of the cortex with each other, subcortical structures (e.g., basal ganglia) and cortical regions of the opposite hemisphere (Forkel and Catani, 2019).Bilateral perisylvian regions that have been implicated in argument structure research are mainly connected by the arcuate fasciculus (AF).The AF can be divided into three segments with cortical terminations in inferior frontal, inferior parietal and posterior temporal regions (Bernard et al., 2019;Catani and Mesulam, 2008;Forkel and Catani, 2019).Clinical work with preoperative language mapping has suggested that when tractography of the AF is used to target specific cortical terminations of the tract using nTMS, stimulation induced error rates increase (Reisch et al., 2022).Hence, considering that the cortical regions that are more active during transitive compared to unergative verb processing are connected by the AF and tractography-based nTMS appears to be more effective in terms of error induction, we used the terminations of the AF as targets for our nTMS protocol.

The present study
The present study set out to examine potential differences in the cortical representation of transitive compared to unergative verbs using nTMS.We asked whether nTMS will induce more errors during transitive verb naming compared to unergative verb naming over the terminations of the AF.We predict that since the strongest fMRI effects have been shown in the left hemisphere, nTMS would disrupt production of transitive verbs compared to unergative verbs more over stimulation of the left AF terminations.However, right hemisphere effects have also been reported (den Ouden et al., 2009;Thompson et al., 2007Thompson et al., , 2010)).Additionally, we expect the effect for transitive verbs to emerge mainly over posterior terminations of the AF (i.e., temporal, parietal) as they have more consistently been shown to be activated during neuroimaging experiments.Regarding the quality of errors, previous nTMS studies that employed the language mapping protocol reported mainly access errors (i.e., anomias and hesitations) to be associated with stimulation of posterior regions during action naming (Ntemou et al., 2021;Ohlerth et al., 2021).We would like to examine whether the probability to induce access errors is higher during the naming of transitive compared to unergative verbs.

Participants
Twenty neurologically healthy native speakers of German participated in the study (age range: 23-58, mean = 29.7,SD = 8.6; gender = 10 female, 10 male).Handedness was assessed using the German adaptation of the Edinburgh Handedness Inventory (Oldfield, 1971).All participants were right-handed besides two, who were left-handed.Every participant provided informed consent prior to stimulation and reported no contra-indications to TMS or MRI (Krieg et al., 2017).Demographic data of the participants are provided in the Supplementary Material, Table 1.
The MrTrix3 software was used for data preprocessing as well as for fiber tracking (Tournier et al., 2019).Diffusion weighted data were corrected for noise, eddy currents, motion distortion, and bias field (Smith et al., 2004;Tournier et al., 2010;Tustison et al., 2010).The iFOD2 algorithm was employed to perform probabilistic tractography of the arcuate fasciculus using a two ROI approach (Fekonja et al., 2019;Tournier et al., 2010).Parameters for tracking were set to 0.1 amplitude cutoff value, 50 mm streamline minimum length, and 5000 selected streamlines.ROI placement followed a user-friendly manual for clinical tractography that has been developed for individuals with and without brain tumors (Fekonja et al., 2019).For fiber tracking of the left and right AF, a seed ROI was placed underneath the deepest point of the central sulcus of each hemisphere and another target ROI was placed slightly more laterally, at the level of the parieto-occipital sulcus (Fekonja et al., 2019).
After tracking of the AF for each participant, Karawun (Beare et al., 2022) was employed to transform the.tck generated files to Dicom object format so that they are compatible with the Brainlab Elements software (Brainlab AG, Munich, Germany).To identify the endpoints of the AF, Dicom objects were burned into the T1-weighted MPRAGE image for each participant, using the "Burned-in Dicom" export option.The burned-in T1 was then co-registered with the original T1 MPRAGE using the default co-registration on Brainlab and was uploaded to the navigated brain stimulation (NBS) system to construct the 3D model of each participant's brain (Nexstim eXimia NBS system version 4.3).The co-registered burned-in T1 was then exported to FSL and was co-registered to the Montreal Neurological Institute (MNI) 152 subject T1 normalized template provided by FSL (Jenkinson et al., 2012).Stimulation points were selected according to previous literature and coordinate transformation using FSL's img2imgcoord.The target coordinates that fell within regions reported to be engaged for the processing of verbs with more arguments (see also 1.2) according to the normalized coordinates and the Harvard-Oxford Cortical atlas (Makris et al., 2006) were selected to be stimulated.Stimulation points of frontal terminations were found in the ventral precentral gyrus, inferior frontal gyrus, and middle frontal gyrus.Temporal stimulation points were found in posterior superior and middle temporal regions, whereas parietal stimulations in the supramarginal gyrus, angular gyrus, and the superior part of the lateral occipital cortex (only 4 stimulation points).MNI stimulation coordinates for each participant are listed in Supplementary Material (Table 2) alongside the most probable regions of the Harvard-Oxford Cortical atlas.

Materials
The action naming task of choice was the German version of the Verb and Noun test for PeriOperative testing (Ohlerth et al., 2020).This standardized task consists of black and white images of 52 (pseudo) transitive and 22 unergative verbs (see Fig. 1).On top of each picture, a lead-in phrase triggers verb inflection for person, number, and tense (e. g., Der Mann … liest 3rd person singular, present tense ; The man … reads).Transitive and unergative verbs were controlled for variables known to affect naming performance such as frequency, age of acquisition, length in syllables, and naming agreement (Bastiaanse et al., 2016;Martin et al., 1989).

Set-up
The burned-in T1-weighted images were uploaded to the NBS system (Nexstim eXimia NBS system version 4.3).Two stimulation points were placed on the frontal, temporal, and parietal cortical terminations of the AF (see also right panel of Fig. 3).Hence, a total of six stimulation points were placed on each hemisphere of every participant.
Following the language mapping protocol by Krieg et al. (2017), the resting Motor Threshold (rMT) for each hemisphere was recorded via electrodes over the contralateral abductor pollicis brevis and abductor digiti minimi muscles.A figure-of-eight coil was used to deliver single pulse stimulations over the hand area of the motor cortex to identify the most excitable spot.The ideal threshold was then automatically calculated by the NBS system.

Baseline testing
Participants were seated ~60 cm in front of a computer screen and were instructed to name all items of the action naming task as fast and accurately as possible in the absence of stimulation.Naming consisted of overtly producing the lead-in phrase and the verb with the required inflection for person, number, and tense (e.g., Der Mann … liest; The man … reads).Participants were instructed to simply name the verb after producing the lead-in phrase and were not required to overtly produce the object.As all transitive items were pseudotransitive, participants naturally refrained from producing the object.Presentation parameters were set according to Krieg et al. (2017) with picture presentation time at 1000 ms and inter-picture interval at 2500-3000 ms.Each participant completed two rounds of baseline testing.After each round, misnamed or inconsistently named items were excluded from naming with nTMS.As a result, participants named individualized sets of images, a common practice during preoperative language mapping (Krieg et al., 2017;Picht et al., 2013;Tarapore et al., 2013).

nTMS language mapping
Instructions as well as picture presentation parameters remained the same as during baseline.Following suggestions by Krieg et al. (2017), the interval between stimulation and picture onset was set at 0 ms.Stimulation frequency was set at 5 Hz/5 pulses and intensity at 110% of the ipsilateral rMT.
The first 10 participants first received nTMS over the left hemisphere AF terminations and then over the right hemisphere terminations, whereas for the last 10 participant the opposite order was followed.The order of termination stimulation (i.e., frontaltemporalparietal) per hemisphere was randomized for every participant.Each cortical termination of the AF was stimulated 10 consecutive times with each verb type.In other words, each of the six cortical terminations of the AF was stimulated 20 times in total, 10 times during naming of transitive verbs and 10 times during the naming of unergative verbs.The total of delivered stimulations were approximately 120 per hemisphere (6 stimulation points x 10 stimulations for transitive verbs x 10 stimulations for unergative verbs).The order of picture presentation was randomized for each verb type.This means that transitive and unergative verbs were named separately and their presentation was randomized.To allow for post-hoc evaluation of nTMS-induced errors mapping sessions were video and audio recorded.

Error classification
As a post-hoc analysis for the AF terminations that showed an effect, we examined the types of errors that were induced.A clinical linguist or a medical doctor analyzed the video recordings of nTMS stimulation and compared them with video recordings of baseline naming, while being blinded to the exact location of stimulation (E.N., F.B.).Following previous literature on preoperative and intraoperative mapping protocols, the error classification is as follows (Corina et al., 2010;Hauck et al., 2015;Picht et al., 2013;Rofes et al., 2017): No response: lack of intelligible response, including the lead-in phrase.
Anomia: the lead-in phrase is produced, but the target is missing (e.g., Der Mann …; The man …).
Hesitation (on the target): the lead-in phrase is produced, but the target is delayed (e.g., Die Frau … uhm … photographiert; The woman … uhm … photographs/takes a photo).Semantic paraphasia: the target verb is replaced by another existing word (e.g., The man … runs for the target The man … walks).
Performance error: production of the lead-in phrase or target verb is stuttered or slurred (e.g., Der Mann … s-s-schäft, The man … s-s-sleeps).
Grammatical error: missing or incorrect inflection of the lead-in phrase or the target (e.g., *Die Frau … laufe, *The woman … run 1st person singular ).
Mixed errors that fell within more than one of the above categories, were inserted and counted in both categories.For example, the error *Der Mann … streiche (The man … paint 1st person singular ) for the target Der Mann … schert (The man … shears) would count both as a semantic paraphasia due to the incorrect verb used, as well as a grammatical error due to the incorrect inflection.

Statistical analysis
To assess whether verb type affected the number of errors per hemisphere, we conducted a logistic regression with Verb type (transitive unergative) and Hemisphere (leftright) as independent variables with interaction.The model also included random intercepts for Participant and Item (i.e., Error_binary ~ Verb type * Hemisphere + (1|Participant) + (1|Item)) as model comparisons with ANOVA revealed that this was the best fitting model.To determine the influence of the stimulated cortical terminations in relation to verb type, we conducted additional logistic models with Cortical termination (frontaltemporalparietal) * Verb type (transitiveunergative).The factor of Verb type was coded with unergative verbs as the baseline condition, while sum contrasts were employed for the Cortical Termination factor.This choice was made as Cortical Termination lacked a specific control condition.Due to the low incidence of nTMS induced errors (see 3.1; Hauck et al., 2015;Picht et al., 2013;Reisch et al., 2022), we opted for a penalized logistic regression model.The reason for this decision is the issue of "rare events problem of maximum likelihood estimations".According to King and Zeng (2001), when the incidence of events (i.e., 1) in small samples is under-represented (n < 200), common logistic regression models underestimate the true probability.Unlike common logistic models, penalized maximum likelihood estimations (PMLEs) seem unbiased even in cases of small samples and few positive events (Firth, 1993;King and Zeng, 2001).PMLEs were implemented in R (R Core Team, 2020) using the logistf package (Heinze and Schemper, 2002).Regarding error types, we used Barnard's tests via the Barnard package to determine whether distributions of error types differed according to verb type (Erguler, 2016).The R script used for the statistical analysis as well as anonymized raw data of the present paper can be found on OSF (https:// doi.org/10.17605/OSF.IO/4PJEY).

Baseline exclusions
Baseline exclusions were computed for the 2 baseline naming rounds for each verb type.For transitive verbs the median for baseline exclusions was 18% and for unergative verbs it was 10%.Baseline exclusions did not differ for each of the participants included in the study (results for each participant can be found in Supp.Material, Table 1).Additionally, baseline exclusion rates did not correlate with the stimulation error rates for either verb type according to Kendall's correlation (transitives: z = − 1.57, p = 0.11; unergatives: z = 0.29, p-value = .76).Figures that visualize the correlation results can be found in Supp.Material, Fig. 1.

Hemispheric analysis
To check whether the nTMS-induced error rate for transitive compared to unergative verbs differed in the two hemispheres, we conducted a logistic regression with random intercepts for participant and item (i.e., Error binary ~ Verb type * Hemisphere + (1|Participant) + (1|Item)).The model with interaction between the two main effects was compared to a model without interaction (i.e., Error binary ~ Verb type + Hemisphere) using an ANOVA, to determine whether the interaction between Verb type and Hemisphere was necessary.The ANOVA confirmed that the model with interaction was a better fit for the data (χ 2 (df) = 9.83 (1); p < 0.001).The model with interaction and random effects was also compared to the model without random effects and was also identified as a better fit for the data χ 2 (df) = 11.8 (1); p < 0.001).
The results of the logistic model revealed that there was no main effect of Verb type (β = − 1.99; p = 0.11) or Hemisphere (β = − 2.31; p = 0.76), but there was a significant interaction between Verb type and Hemisphere (β = − 1.6; p < 0.001).Fig. 2 shows the error rates according to verb type and hemisphere.

Left hemisphere: errors between AF cortical terminations
Over the frontal cortical terminations of the left AF, nTMS induced on average 11.3% of errors during unergative and 18.8% during transitive verb naming.Over the temporal cortical terminations mean unergative error rate was 7.5%, whereas transitive error rate was 21.2%.During stimulation of parietal terminations, nTMS induced 10.6% of errors during unergative verb naming and 13.1% with transitive verbs naming.Regarding whether nTMS induced more errors during transitive rather than unergative verb naming over the left AF terminations, the PMLE using Verb type * Cortical terminations as explanatory factors revealed a main effect of transitive verbs (β = − 1.8; p = 0.006) and a significant interaction between transitive verbs and stimulation of the temporal terminations of the AF (β = − 1.67; p = 0.03).Table 3 in Supplementary Material lists the results of the PMLE and Fig. 3 shows the nTMS-induced error rates of the left AF cortical terminations according to verb type.

Left hemisphere: errors within AF cortical terminations
Overall, the left posterior STG was stimulated 220 times and the posterior MTG a total of 589 times.Barnard's tests revealed that in the STG (W = − 2.6; p < 0.01) as well as MTG (W = − 4.9; p < 0.001) nTMS induced more errors during transitive rather than unergative action naming.The alpha level has been adjusted for 2 comparisons using the Bonferroni correction and p-values are considered significant for p < .025.

Discussion
The present study investigated the effect of transitivity during action naming in neurologically healthy individuals using tractography-based inhibitory nTMS.We showed that during stimulation of the cortical terminations of the left AF, nTMS induced more errors during naming of finite transitive compared to unergative verbs.Moreover, among the cortical terminations of the left AF, nTMS affected more the production of transitive compared to unergative verbs when the temporal terminations of the AF were stimulated.Specifically, these terminations were E. Ntemou et al. located in the left posterior STG and MTG in most participants.Looking at the qualitative analysis of error types, we demonstrated that nTMS disproportionately affected access to transitive verbs compared to unergative verbs when stimulating left AF temporal terminations by inducing more hesitations.This effect was specific to stimulations that targeted the posterior MTG.

Left hemisphere and transitivity
Comparisons between verbs with different number of arguments have been previously examined in groups of neurotypical adults by employing fMRI.Activation related to comprehension and production of verbs with more arguments (i.e., transitive and ditransitive) is located in left posterior but also right homologous cortical regions (Den Ouden et al., 2009;Thompson et al., 2007Thompson et al., , 2010)).It is not clear, however, whether right hemisphere activation reflects linguistic operations specific to argument structure or cognitive processes that cannot be directly disentangled.For example, den Ouden et al. (2009) noted that even though action naming of transitive actions shows bilateral activation in inferior parietal regions, activation clusters are especially right lateralized only when action stimuli consist of videos rather than pictures.During lexical decision of verbs with different number of arguments, Thompson et al. (2007) also highlighted that right hemisphere regions are activated only when both ditransitive and transitive verbs are compared to unergative verbs, but not when transitives are compared to unergatives as such.Considering that right hemisphere regions have only been reported during processing of ditransitive verbs and video stimuli, it seems reasonable that we did not observe such an effect.Our stimuli consisted exclusively of pictorial material and we only included finite transitive and unergative verbs.Besides neuroimaging studies, however, previous TMS evidence does not support the involvement of right hemisphere cortical regions in the processing of argument structure (Finocchiaro et al., 2015;Ntemou et al., 2021;Ward et al., 2022).On the contrary, Ward et al. (2022) reported that stimulation of the right superior parietal cortex interfered equally with both transitive and unergative action naming.
According to theoretical accounts, the number of arguments a verb carries is part of its argument structure and is specified in the lemma level (Grimshaw, 1990).Information that is retrieved as part of the lemma guides the process of phrase structure building that subsequently takes place in the stage of morphological encoding (Levelt et al., 1999).
Hence, argument structure information can be considered the essence of lexical syntax that underlies sentence structure (Matchin and Hickok, 2020).This process has not been reported to interface with other cognitive or executive functions and has been consistently regarded as part of the lexicon (Bock and Levelt, 1994;Caramazza, 1997;Levelt et al., 1999).Considering that the left hemisphere has long been considered the primary hemisphere for linguistic processes, it seems reasonable that applying nTMS to the left hemisphere would disproportionately affect the production of verbs with more arguments.Similarly, studies of post-stroke aphasia after left hemisphere lesions have also demonstrated impairments with arguments structure (De Bleser and Kauschke, 2003;Luzzatti et al., 2002;Malyutina et al., 2016;Thompson, 2003).Even though lesion studies of the right hemisphere in relation to argument structure are scarce, work on verb processing has suggested that posterior temporal lesions of the right hemisphere do not lead to difficulties with verb comprehension (Neininger and Pulvermüller, 2003).More importantly, neuroimaging and neurostimulation studies concur in that grammatical aspects of verb processing are localized in left hemisphere regions, unlike pragmatic and prosodic processes, which may be represented bilaterally (Hartwigsen and Siebner, 2012;Jung-Beeman, 2005).

Posterior temporal terminations of the left AF
nTMS induced more errors during transitive than unergative verb production in the left hemisphere.However, this effect was particularly prominent when stimulating the posterior temporal terminations of the tract.Importantly, nTMS did not induce more errors during naming transitive verbs when inferior parietal or inferior frontal terminations of the AF were stimulated.These findings can inform theories about the processing of argument structure and lemma-level information in the brain.
The neurocognitive model of argument structure by Thompson and Meltzer-Asscher (2014) proposed that inferior parietal areas are engaged for retrieval of argument structure information, whereas posterior temporal regions are engaged for integration of retrieved material (see also 1.2).As a result, according to this model, suppression of posterior temporal regions using nTMS should result in issues with integration of the verb and its arguments into the sentence.When stimulating the temporal terminations of the AF, we induced more hesitations before verb retrieval of transitive than unergative verbs.In Fig. 3. nTMS-induced mean error rates according to the terminations of the left AF (frontalparietaltemporal) and verb type (transitive vs unergative).The right panel presents a 3D representation of the stimulation points placed over the terminations of the left AF for participant 3 in individual space.The 3D figure was created using MRIcroGL (https://www.nitrc.org/projects/mricrogl).
E. Ntemou et al. our paradigm, however, transitive stimuli were pseudotransitive verbs (e.g., to cook, to eat, to drink) and their second argument was optional (see 2. Methods).This implies that transitive verbs did not require an object to be encoded or at least overtly produced (e.g., the boy cooks vs the boy cooks pasta).Theoretical accounts regarding the optionality of arguments in transitive constructions also agree that arguments of predicates do not have to be satisfied in the case of pseudotransitive verbs (Williams, 2015).Hence, integration of the verb and its arguments could be regarded as not necessarily more complex for pseudotransitive than unergative stimuli.From this perspective, it becomes challenging to explain our findings based on the account of sentence integration taking place in the posterior temporal lobe (Thompson and Meltzer-Asscher, 2014).Additionally, the only error type that differed between the two verb categories were hesitations that occurred before the encoding of the verb.According to seminal psycholinguistic work, when transitive sentences are produced, pre-planning does not happen for the object but only for the subject and the verb (Lindsley, 1975(Lindsley, , 1976)).Hence, if a process related to integration was inhibited by nTMS, it would have to be a process related to the integration of the verb rather than the object.However, considering that our paradigm was not designed to assess processes of integration vs lemma retrieval, such speculations should be verified by future appropriate experimental designs.For example, an experiment that provides the required lemmas for sentence construction during repetitive TMS could be contrasted with an experiment that does not provide lemmas in order to target retrieval.
Another interpretation of our results comes from the neurocognitive model of syntax by Matchin and Hickok (2020).According to this model, inferior parietal areas such as the AnG are only involved in conceptual processing, whereas posterior temporal regions (i.e., posterior MTG) are in charge of generating lemma-level information for individual verbs (Matchin and Hickok, 2020).Within this view, suppression of the posterior MTG using nTMS would lead to difficulties with lemma-level information.Given that lemma representations need to be accessed prior to morpho-phonological encoding, delay of access to lemma representations could lead to hesitations or anomias when attempting to produce the verb.Even though anomic errors failed to reach the corrected alpha threshold for multiple comparisons, hesitations on the target verb were induced disproportionately more often when the posterior MTG was stimulated during naming of transitive items.Our findings seem therefore in accordance with fMRI investigations that keep semantic complexity constant but vary the presence or not of argument structure (e.g., frightened the child vs the frightened child; Matchin et al., 2019).Importantly, such findings stress the exclusive role of posterior temporal regions (Matchin et al., 2019; see also Ben-Shachar et al., 2003).
The question that arises is what precisely causes the increased complexity of lemmas of pseudotransitive compared to unergative verbs?Earlier work with fMRI has suggested that it is not merely the higher number of arguments that causes increased activation, but it is primarily argument optionality or the potential of having more than one argument structure options (Shetreet et al., 2007).Given our experimental paradigm, it is challenging to disentangle whether the observed effect stems from pseudotransitive lemmas carrying information for a higher number of arguments or whether lemmas have two distinct specifications for potential syntactic environments the verb can appear in (i.e., an unergative treelet and a transitive treelet).We reason, however, that since both neurocognitive models converge on the processes carried out by frontal regions (i.e., linear sequences built based on lemma-level information; see below), if indeed pseudotransitive verbs contain one transitive lemma, their retrieval would kickstart the building of an SVO structure in the left IFG.SVO structures would contrast with SV structures prompted by unergative verbs.Considering that we did not observe an effect for transitive over unergative verbs when stimulating left frontal terminations of the AF, the argument that the locus of the effect is the retrieval of a second argument (i.e., the theme) becomes less likely.Future work with targeted linguistic contrasts between obligatory 2-argument verbs (i.e., pure transitives) compared to optional 2-argument verbs (i.e., pseudotransitives) can shed light on this question.

Frontal and parietal regions in relation to argument structure
Unlike previous work with fMRI using a similar paradigm, we did not observe an effect for transitive compared to unergative verbs when stimulating AF terminations in left inferior frontal areas (den Ouden et al., 2009).The reason for this discrepancy might lie in our experimental stimuli in combination with the role of the IFG in language processing.The posterior IFG seems to participate mainly in linguistic processes that have to do with linearization of linguistic information (e. g., word order; Ben-Shachar et al., 2003;Friederici, 2018;Matchin and Hickok, 2020;Walenski et al., 2019).The outcome of hierarchical processes, however, does not always need to be modified to abide by the rules of linearization.For example, the arguments of both unergative as well as transitive verbs used in the present study, do not need to be moved from their base position to satisfy linear word order.It is also important to note that for our transitive items, participants were not required to overtly produce the second argument (i.e., object) but rather only the inflected verb and as a result, the full syntactic sentence with the object was not necessarily produced.In addition, morphosyntactic relations between subject and verb were minimized to 3rd person singular present tense inflection (e.g., Der Mann telefoniert) that was consistently repeated for every item.It is, hence, questionable whether inflectional processes were truly computed for every item or whether they were somewhat automatic phonological operations.These potential reasons may have led to the IFG to be less engaged for the given task.Lesion symptom mapping studies from both people with brain tumors as well as post-stroke aphasia also fail to identify frontal areas as predictive for tasks that involve argument structure within the modality of production (Ntemou et al., 2023a,b;den Ouden et al., 2019).
If information about event semantics is indeed being processed in bilateral inferior parietal areas, we would have found more nTMSinduced errors for transitive verbs when stimulating parietal terminations of the left and right AF.However, this hypothesis presupposes that transitive stimuli in the present experiment were more complex than unergative stimuli in terms of event structure.As previously suggested (Matchin et al., 2019), it is challenging to determine the conceptual entities that are part of an event.The pictorial stimuli used in the present study included transitive actions for which the object was not necessarily overtly expressed (e.g., to read) as well as unergative actions (e.g., to sail).As can be seen in Fig. 1, both types of pictorial stimuli represent additional entities other than the agent.In the case of the transitive stimulus, the grammatical object is represented (i.e., a book).In the case of the unergative stimulus, contextual entities that characterize the action of sailing are also depicted (e.g., a sailing boat, a lake).Hence, it is possible that our transitive stimuli did not differ from the unergative stimuli in terms of the entities that participated in the depicted events and as a result inferior parietal areas were not more engaged during transitive verb production.
Studies using TMS additionally showed that superior regions of the left parietal lobe are engaged for the processing of transitive sentences (Finocchiaro et al., 2015(Finocchiaro et al., , 2021;;Ntemou et al., 2021;Vercesi et al., 2020;Ward et al., 2022).The present paradigm was guided by tractography of the AF and, as a result, we did not stimulate superior parietal regions or the inferior parietal sulcus.Considering that parietal terminations of the AF were located in the supramarginal and angular gyri in combination with the focality of TMS effects, we cannot exclude potential co-stimulation of the inferior parietal sulcus (Numssen et al., 2023;Pascual-Leone and Walsh, 2003).However, we did not specifically target or fine-tune our protocol to assess potential differential contributions to transitive and unergative verb naming for the inferior parietal compared to the superior parietal lobe or the inferior parietal sulcus.

Limitations & future work
The present work employed a clinically-relevant protocol that is being used for preoperative language mapping in people with brain tumors (Krieg et al., 2017;Picht et al., 2013;Reisch et al., 2022).However, preoperative nTMS protocols have been shown to have rather low concordance with the gold standard of causal language mapping that is intraoperative Direct Electrical Stimulation (Picht et al., 2013;Tarapore et al., 2013).Therefore, it is necessary to validate our results using data obtained from intraoperative DES.Also, nTMS protocols that aim to induce language production difficulties are characterized by low numbers of induced errors (e.g., Hauck et al., 2015;Ohlerth et al., 2021;Reisch et al., 2022;Sollmann et al., 2015).Our study is not unique in this regard.Given this intrinsic issue of nTMS studies, it would be useful to include measurements of reaction times (RTs) as a more sensitive measure to track nTMS effects.With these studies, we would expect nTMS effects to manifest as increased RTs for transitive compared to unergative verbs particularly when stimulating the posterior temporal lobe.Additionally, the standardized task used for the present study included different numbers of transitive and unergative verbs.This implies that in order to achieve equal stimulations for transitive and unergative verbs, unergative items had to be repeated more so compared to transitive items.However, to account for this limitation, we ensured that exclusion rates for transitive compared to unergative verbs did not differ during baseline and they did not correlate with nTMS-induced error rates (Supp.Material; Table 1; Fig. 1).If differences in the number of items for each verb type influenced the nTMS-induced error rates, we would have expected to see these differences emerge during baseline as well.
Last, a central constraint of the present study concerns our tractography-based protocol for language mapping.Even though tractography-based protocols have been shown to be more sensitive compared to conventional preoperative mapping protocols, the stimulation points are by definition dependent on the terminations of white matter pathways (Reisch et al., 2022).Considering the variability of white matter structures (Bernard et al., 2019), it was not possible to target the same cortical area for every participant (e.g., see 3.3 for temporal lobe terminations).Tractography methods are also subject to various limitations that range from inconsistencies with manual ROI placement to inherent issues of tractography algorithms (Schilling et al., 2021).To minimize these potential issues, we opted for the use of standardized ROI placement protocols and probabilistic tractography algorithms (Dell'Acqua and Tournier, 2019;Fekonja et al., 2019).Yet, the issue of variability in white matter reconstructions persists and should be addressed in future work and potential further technological advancements in the field of tractography.An additional point for future contributions is the involvement of different white matter fascicles in language processing.For example, terminations of fascicles that have been implicated in language processing during intraoperative DES, such as the inferior longitudinal fasciculus and the inferior fronto-occipital fasciculus, should also be probed with a similar method (Duffau et al., 2014).This could shed light on the specificity of the reported effect for transitive verbs over terminations of the AF.
Concerning our theoretical questions in relation to the processing of argument structure information in the brain, more evidence is needed for the differential contributions of posterior temporal regions.Although the posterior MTG has been mainly connected to the processing of grammatical information, previous work with lesion-symptom mapping has not always clearly located the effect over the posterior MTG (Matchin et al., 2020;Matchin and Hickok, 2020).In the present study, it was indeed the MTG that was primarily affected by nTMS stimulation during transitive verb naming, but the occurrence of errors was rather low (20% for transitive verbs; 5.4% for unergative verbs).Future work could verify these results by specifically targeting the posterior STG and MTG with nTMS.Furthermore, given that the inferior parietal sulcus as well as more superior areas of the parietal lobe have been connected to the processing of transitivity (Finocchiaro et al., 2021;Ward et al., 2022), it seems relevant to clarify the contributions of parietal structures compared to posterior temporal areas.Last, our items were controlled for several psycholinguistic variables (see 2.3).However, semantic variables such as semantic class of the verbs were not controlled for.These verb variables might play a role in the outcomes of nTMS language mapping, but given that research on this topic is scarce (e.g., Finocchiaro and Miceli, 2002), future studies could shed light on these questions.

Conclusions
The present study set out to investigate the differential contributions of perisylvian regions connected by the AF in the production of finite transitive and intransitive verbs.We used tractography-based inhibitory nTMS to induce errors during a standardized action naming task.Our results emphasize the role of left posterior perisylvian regions in the processing of grammatical information of verbs.These results are in line with the previous fMRI and lesion-based literature (e.g., Matchin et al., 2019Matchin et al., , 2022;;Matchin and Hickok, 2020;Shetreet et al., 2007) and provide novel support for the contribution of left posterior temporal areas in argument structure processing using causal evidence via tractography-based nTMS.

Fig. 1 .
Fig. 1.Picture stimuli of a transitive (a) and an unergative (b) verb used in the present study.Stimulus (a) elicited the target response Der Mann liest -The man reads and stimulus (b) the target response Der Mann segelt -The man sails.

E
.Ntemou et al.