3.3.1. Left Hemisphere

The most frequently induced errors in object naming occurred in the phonologicalarticulatory category (40%), with highest error occurrences in the mMTG, mSTG, pSMG, and vPrG, followed by lexico-semantic errors (39%) mainly in the anG, pSMG, pMFG, and mPrG. Fewer non-linguistic speech errors (19%) were found mainly in the aSMG and mMTG and with the lowest frequency, grammatical errors (2%) were found in the mMTG, mPrG, and SPL (Table 3).

**Table 3.** Error category rates ± standard deviation and ratios in the left hemisphere per cortical parcellation system (CPS) region. For instance, out of all 31 errors that occurred in the anG during object naming, four were of non-linguistic nature, resulting in a non-linguistic ratio of 12.9%, while the error category rate of 0.008 is based on those four errors out of the entire 480 trials in this region.


115


**Table 3.** *Cont*.

In action naming, a similar pattern of error category frequencies appeared with lexicosemantic errors being most frequent (46%) and found mainly in the dPrG and pSTG, followed by phonological-articulatory errors (34%) being elicited for the vPrG, pMFG, mSTG, and aSTG. Again, fewer non-linguistic speech errors (18%) were found and appeared mostly in the aSTG, mSTG and pSMG, and grammatical errors with the lowest frequency (2%), found in the pSFG, mPoG and vPrG. After applying Mann–Whitney–Wilcoxon tests, action naming overall elicited significantly more errors at the lexico-semantic level than object naming (*p* = 0.013, r = −0.560; Table 4). Due to the small numbers of errors per CPS region, no meaningful comparison in error categories per region was achieved.

**Table 4.** Comparisons of error category rates ± standard deviation in left hemisphere (LH) and right hemisphere (RH). Statistical significance is marked in bold and with an asterisk (\*).


#### 3.3.2. Right Hemisphere

In the RH, for object naming, the most frequent errors were in the lexico-semantic category (41%) mostly in the vPrG, vPoG, and aSMG, followed by phonological-articulatory errors (39%) in the aSTG, mMTG, and mSFG. Fewer errors were elicited in the non-linguistic speech category (18%) in the pSTG, TrIFG, and mMFG and hardly any incidences in the grammatical category (2%) in the mMTG, mPoG, and pMFG (Table 5).

**Table 5.** Error category rates ± standard deviation and ratios in the right hemisphere per cortical parcellation system (CPS) region. For instance, out of all 26 errors that occurred in the anG during object naming, six were of non-linguistic nature, resulting in a non-linguistic ratio of 23.1%, while the error category rate of 0.012 is based on those six errors out of the entire 480 trials in this region.


118


**Table 5.** *Cont*.

Action naming also displayed most errors in the lexico-semantic category, mainly in the vPoG, dPrG, mPrG, and trIFG (47%), and as the second most frequent, errors in the phonological-articulatory category (30%), mainly in the opIFG, pSTG, and vPrG. Fewer errors were found in the non-linguistic speech category (22%), mostly in the aSTG, mMTG, and mPrG, and errors in the grammatical category (1%) in the pSTG, mPrG, and opIFG. Again, action naming elicited more errors in the lexico-semantic category in the RH (*p* = 0.022, r = −0.518) (see Table 4).

#### *3.4. Area-Wise Comparison of Tasks for All Errors*

Action naming had a significantly higher error rate than object naming in both hemispheres. Table 6 shows the error rates according to the hemisphere and CPS region. In the LH, none of the CPS regions showed a significant difference between error rates for object and action naming, but a trend for a higher error rate for action naming was found in the mSTG (*p* = 0.065) and in the vPoG (*p* = 0.051). In the RH, action naming elicited significantly more errors in the aSTG (*p* = 0.036, r = −0.455), dPrG (*p* = 0.026, r = −0.547), mPoG (*p* = 0.020, r = −0.617), mPrG (*p* = 0.012, r = −0.557), mSTG (*p* = 0.034, r = −0.410) and a trend in the opIFG (*p* = 0.050). Figure 4 depicts the cortical distribution of error rates per CPS for object naming (a) and action naming (b) in heat maps for both hemispheres.

**Figure 4.** Error rates in percentage according to Table 6 comprised of all errors per cortical parcellation system (CPS) regions in the left hemisphere and right hemisphere during object naming (**upper row**) and action naming (**lower row**).

−

− −

−

−

**Table 6.** Error rates ± standard deviation per region during object naming (ON) and action naming (AN) in the left hemisphere (LH) and right hemisphere (RH) and their differences. For instance, 31 errors in the anG out of 480 trials (20 participants per four stimulation targets per six stimulations) resulted in an error rate of 0.065 ± 0.087. Statistical significance is marked in bold and with an asterisk (\*).


#### *3.5. Hesitation Error Exclusion*

The number of errors and resulting error rates in each task differed significantly, when hesitation errors were excluded. Table 7 summarizes these comparisons for each task and hemisphere.

**Table 7.** Comparisons of error rates ± standard deviation including all errors vs. excluding hesitations. Statistical significance is marked in bold and with an asterisk (\*).


For a second analysis, hesitation errors (both hesitation on the whole phrase and hesitation on the target) were excluded for a separate comparison. Error rates without hesitations differed significantly between tasks: action naming demonstrated a higher error rate than object naming in both hemispheres (LH: mean error rate action naming = 0.040, mean error rate object naming = 0.029 (*p* = 0.042, r = −0.472); RH: mean error rate action naming = 0.043, mean error rate object naming = 0.031 (*p* = 0.035, r = −0.472)). Again, error rates on baseline naming did not correlate significantly with the error rates without hesitation in either of the tasks or hemispheres (LH: rho = 0.080, *p* = 0.737 for object naming, rho = 0.180, *p* = 0.447 for action naming; RH: rho = 0.148, *p* = 0.551 for object naming, rho = 0.224, *p* = 0.342 for action naming).

Multivariate regression models revealed that neither baseline errors (LH: t = 1.231, *p* = 0.237; RH: t = 0.880, *p* = 0.393), rMT (LH: t = −0.680, *p* = 0.507; RH: t = −0.281, *p* = 0.782), handedness (LH: t = 1.065.23, *p* = 0.304; RH: t = 0.783, *p* = 0.446) nor age (LH: t = −0.811, *p* = 0.430; RH: t = −0.986, *p* = 0.340) were significant predictors for the error rates in either of the hemispheres.

#### *3.6. Area-Wise Comparison of Error Rates without Hesitations*

When analyzed separately per hemisphere, a significantly higher error rate was observed for action naming over object naming in both hemispheres (LH: *p* = 0.042, r = −0.472; RH: *p* = 0.035, r = −0.472). Table 8 depicts the error rates per hemisphere and for the two tasks and their comparisons. In the LH, a significantly higher error rate was found for action naming in the pMFG (*p* = 0.037, r = −0.561); in the RH, in the opIFG (*p* = 0.020, r = −0.536) and a trend in mSTG (*p* = 0.067). Figure 5 depicts the cortical distribution of error rates per CPS region for object naming (a) and action naming (b) in heat maps for both hemispheres.

**Table 8.** Error rates ± standard deviation excluding hesitations per region during object naming (ON) and action naming (AN) in the left hemisphere (LH) and right hemisphere (RH) and their differences. For instance, 15 errors in the anG out of 480 trials (20 participants per four stimulation targets per six stimulations) resulted in an error rate of 0.031 ± 0.052. Statistical significance is marked in bold and with an asterisk (\*).


**Figure 5.** Error rates in percentage according to Table 8 comprised of errors excluding hesitations per cortical Parcellation System (CPS) region in left and right hemisphere during object naming (**upper row**) and action naming (**lower row**).

### **4. Discussion**

The objective of this study was to evaluate the potential of tasks in sentence context, specifically a verb-targeting language task, for error elicitation in different cortical surfaces under nTMS. By extensively mapping 20 healthy participants in both hemispheres using the novel task action naming in sentence context, together with object naming in sentence context, task sensitivities were compared. Quantitative differences in error elicitation of the two tasks were investigated overall and per small cortical area. Moreover, we aimed to understand the breakdown in language production caused by nTMS through a detailed qualitative error analysis. Lastly, the effect of hesitation errors on mapping results was examined through a separate analysis to define its significance further.

#### *4.1. Overall Task Comparison*

As for the primary comparison, action naming delivered a higher error rate than object naming in both hemispheres. This leads to the conclusion that retrieving a verb in sentence context is more easily disrupted under nTMS than retrieving a noun in sentence context. No significant correlation between the number of baseline errors and errors under stimulation was found. Moreover, all error-prone verb stimuli are removed during the baseline process. As a result, errors under stimulation should be considered true positives. The higher error rate of action naming, therefore, cannot be explained by this task being more error-prone, per se, but points towards a higher vulnerability of the more complex process of verb versus noun retrieval in sentence context under stimulation. This finding seems contrary to previous studies, which have argued that verb tasks under nTMS did not reach the same sensitivity as object naming and were, hence, not worth including for nTMS language mapping [43,44].

The discrepancy with the present findings may be attributed to the designs of the tasks across studies. Firstly, the picture-naming paradigm differed. Former studies employed well-established databases for the object naming variant [51], while using homemade drawings or photos for the verb task. This hampers the direct comparison. Our design entails stimuli for action naming of similar complexity and style as object naming. Moreover, our stimuli had been previously tested for a high naming agreement. Secondly, whereas the former studies targeted single word retrieval for both tasks ('ball' for object naming, 'throwing' for action naming), the current study made use of object naming and action naming in sentence context. Next to target word retrieval, this required inflection and embedding of the target in a short lead-in phrase. The higher cognitive effort needed for our action naming task is likely to be differently affected by nTMS and may have resulted in the higher error rate under stimulation in our sample. This finding is entirely in line with data from DES mapping, arguing that verb tasks are more sensitive under stimulation [23,28,52]. The following sections will provide a closer look at the root of this sensitivity by looking at error types and specific cortical locations.

#### *4.2. Error Category Comparison*

In most protocols for nTMS language mapping, different error types are used for a more detailed mapping depiction [6,18,44]. However, these classifications are hardly ever used to further unravel the origin of the errors. In the present study, we employed the common error types found in word production under stimulation, projected them on errors in sentence context, and assigned them to the level of production disruption they indicate (see Figure 1). This classification was used to better understand the difference in error rates on the two tasks under investigation. Additionally, testing in sentence context allows for screening of more subtle errors than testing a single word and allows categorization of the errors into different stages of breakdown in the sentence production process.

When taking together all tasks and hemispheres, a similar pattern of error category frequency appeared: errors at the phonological-articulatory and lexico-semantic level were most prevalent, whereas fewer non-linguistic errors and even fewer grammatical errors were induced (Tables 3–5). This pattern is consistent with previous reports [6,34–37]. While intraoperative DES, possibly due to its higher frequency and direct application on the exposed cortex, easily elicits full disruptions of speech [38], nTMS is known to hinder mainly the phonological-articulatory processes and lexico-semantic retrieval [6,34–37]. The effect of nTMS versus intraoperative DES is not yet well understood. It has been established, however, that the timing of the nTMS pulses in relation to the stimulus to name can alter the error pattern [53]. The common protocol of a delay of 0 ms was found to produce the best mapping results when compared to the intraoperative gold standard [16,18]. This protocol evokes about 40% of errors on the sound level, as confirmed by our study with 257/729 errors in the LH and 269/815 error in the RH on the sound level.

Regardless of the task, nTMS seems to disrupt the later levels of word and sentence production with varying location in both hemispheres. Analysis of errors at this level, thus, did not reveal differences between object and action naming. Due to the additional inflectional effort needed to embed a target verb in a sentence compared to embedding a target noun, one expects errors at the grammatical level to be more pronounced for action naming than for object naming. However, no significant difference was detected. Instead, the error rates on the two tasks differed at the lexico-semantic level (Table 4). These errors occurred more frequently for the verbs than the nouns. Therefore, nTMS seems to affect verb retrieval more than noun retrieval. A possible reason for this is the verb's more complex conceptual and lexico-semantic information. As head of the sentence, lexical entries of verbs carry information about argument structure of the sentence. Moreover, higher abstractness of actions compared to nouns and objects adds to the semantic complexity and may result in a higher vulnerability for verbs. This known distinction has been reported after brain damage [54,55] and seems to hold for nTMS mapping as well.

We cannot rule out that lexico-semantic errors indicate the difficulty to retrieve even the concept of the target or to inflect the target. Inflection in sentence context may require more cognitive effort, even though the error is not grammatical in nature, but rather leads to hesitations of the inflected target or to anomia. While the present setup cannot disentangle this further, elicitation of the targets in sentence context allows distinguishing between a full speech breakdown and disturbance in target retrieval. Narrowing down the origin of the errors revealed a higher vulnerability for action naming at this level under nTMS. The inclusion of a small lead-in phrase is, hence, not only informative, but also crucial for sophisticated error classification and creating a more effortful task, that is evidently easier to disturb with nTMS.

### *4.3. Area-Wise Comparison in the Left Hemisphere*

Smaller scaled comparisons per predefined CPS region were performed to reveal whether one of the tasks elicited a higher error rate in a more localized cortical surface area. In the LH, none of the comparisons between object and action naming reached significance for any CPS region (Table 6). Whereas on the entire hemisphere, action naming seems to be more easily disrupted; this could not be localized to a specific area.

Taken together with the fact that, during both tasks, at least a few errors appeared in every CPS region, this is in line with the body of navigated stimulation mapping studies that do not support a classical double dissociation of noun and verb production in the LH [25,54,56] (for reviews see [57,58]). Literature on stroke-induced aphasia suggests a left-sided temporal lobe hub for comprehension and production of object names and a left-sided frontal lobe hub for tasks related to action/verb production [26,27,59] (for a recent review see [60]). This claim was questioned by data from many methodologies [57,58], including mapping studies under nTMS and intraoperative DES.

In nTMS mappings using single-word targets, no selective areas for either task were reported [43,44], but a widespread region in the perisylvian area, covering all three lobes for both object naming and action naming. The conclusion is similar in studies using intraoperative DES with single-word targets [49,52]. A double dissociation could be delineated in single cases; however, this distinction did not surface at the group level. The same conclusion holds for intraoperative mappings with object and action naming in sentence context. Single cases of an exclusive involvement of the opIFG in action naming have been described [23] and a more prefrontal/premotor network for action words [28], but no clear-cut group pattern of a double dissociation arose either.

The grand conclusion emerging from group analysis in intraoperative DES mapping in patients and nTMS mapping in healthy participants, that our data add to as well, is a mainly to entirely shared perisylvian network for verb and noun production. While this conclusion cannot help to resolve the decade-old debate about a neural, clear-cut segregation of nouns versus verbs in the brain, it stresses the usefulness of a two-task design: a few double dissociations were obtained in single cases [28,49,52]; moreover, cases have been described in which action naming was the only task to elicit errors [23]. These observations let the authors to conclude that action naming is a necessary addition to the standard object naming task when trying to avoid postoperative deficits. That this is not confirmed at the group level is likely due to high inter-individual variability [28,38,61–63], but becomes evident in the reported single cases as well as in the present data.

The group data in this experiment still gives insight into error patterns per task and hemisphere. In the LH, object naming errors were spread over all CPS regions (Table 6), but the highest error rates were found in the vPrG, mPrG, pSMG, and mMTG. The middle and ventral parts of the PrG are known as components in articulatory planning [29], which is reflected in our data in a high ratio of phonological-articulatory errors (Table 3) as well as in other nTMS [36,53] and intraoperative DES studies [64]). The function of the pSMG ranges from access to semantic representations as part of Gschwind's region [36,65] to phonological decision making [28,29,36,38], and showed no clear error association in our data either. In the mMTG as a presumed semantic hub [29], a high ratio for phonological-

articulatory errors was elicited in our data and, therefore, cannot confirm this common function relation. With that said, it is important to point out that none of the areas were correlated with a specific error type, frequently occurring in that area. The areas rather seem to be frequent network hubs during language production. The above-named functional associations are, therefore, to be taken with caution regarding their presumed underlying processes.

Action naming highest error occurrences lay in the vPrG, vPoG, and mSTG, with trends for significantly higher error rates compared to object naming in the vPoG and mSTG (Table 6). Again, no CPS region appeared without errors. The ventral parts of the PrG and PoG play a role in the embodiment hypothesis [66] and, therefore, would be compatible with lexico-semantic errors during action naming. However, our data show several error types prevalent in these regions (Table 3), not predominantly lexico-semantic errors. The medial part of the STG, both found with high error rates in nTMS [43,44] and intraoperative DES studies [49] for action naming, is so far the only constant area throughout several studies that is essentially involved in action naming. Classically thought of as a semantic hub close to Wernicke's area, a mixed error ratio in our data cannot further specify its exact role in production (Table 3). However, regardless of the distinct error category, the persistent appearance of the mSTG throughout different methodologies stresses its role in action naming and may make the verb task a better candidate for mapping in this temporal area.

#### *4.4. Area-Wise Comparison in the Right Hemisphere*

Studies of mapping with a direct comparison of object naming and action naming are rare. The literature is even scarcer for a comparison of tasks in the RH. Since language functions are dominantly hosted by the LH, mapping of the contralateral hemisphere is usually deemed unnecessary [57] However, attention to the RH has been renewed after studies using multiple methodologies have suggested that the involvement of the RH in language has long been underestimated [67,68]. Recruitment of RH areas was found in functional MRI studies, ranging from domain-general processes, such as attention and working memory [69,70] to linguistic processes, such as sound to lexical meaning mapping in the IFG [71], bilateral conceptual knowledge in the anterior temporal lobe [72–74], bilateral phonological decision making in the IFG and SMG [75,76], and explicit impairment in comprehension after RH damage [67]. Studies using intraoperative DES have described a mirrored pattern of LH homologues in the RH with stimulation of frontal areas resulting in articulatory errors and speech arrest, and a temporal hub in the RH for conceptual and semantic knowledge, seen in semantic errors and anomias [61,64,68,77–80].

Only a handful of nTMS studies have mapped both the LH and RH according to segmentation of the brain by the CPS, all employing object naming. Overall, a comparably high error rate was reported for the RH [6,35,81,82], albeit lower than the LH counterpart. The detection of language areas in the RH by nTMS has therefore been described, but has not been compared between tasks. The present study is, hence, the first to systematically compare of object naming and action naming in the RH under nTMS.

In our sample, object naming errors were elicited in all CPS regions. The highest error rates were located rostrally to the Sylvian fissure, but also over all three lobes with the highest occurrences in the mMTG, vPoG, trIFG, mMFG, and vPrG (Table 6). Both the mMTG and vPrG mirror the pattern of higher error rates of the LH and align with findings from nTMS [6,81] and intraoperative DES [68,77,80]. However, error categories in our sample—phonological-articulatory errors in the mMTG and lexico-semantic errors in the vPrG (Table 5)—do not fit the described clear function allocation described by Duffau and colleagues [61,64,68,77,80] where frontal stimulation would result in speech motor errors and temporal stimulation in lexico-semantic errors. The vPoG as well as the trIFG and mMFG displaying mixed errors can be considered the RH homologues of known LH language regions, engaging in speech motor functions [81]. Bilateral activation for articulatory

and speech motor functions is evident and falls in line with reports for a bilateral language recruitment from nTMS, intraoperative DES, and neuroimaging [6,64,68,70,72–76,79–81].

Action naming was most frequently disturbed around the central sulcus and Sylvian fissure, with the aSTG, vPoG, and mMTG as the most receptive areas (Table 6). As part of the STG, the aSTG is to some degree mirroring the LH pattern. Being prone to non-linguistic speech errors (Table 5), the aSTG may be crucial for early conceptual processes and thereby in accordance with bilateral activation during conceptual knowledge recruitment found in fMRI and DES [68,72–74,78,80]. The function of the vPoG in bilateral sensorimotor activation during performance of actions may explain its involvement during action naming. The dominant error category in vPoG of lexico-semantic disruption underlines this further and points towards embodiment in the RH. A high ratio of lexico-semantic errors in the mMTG in our sample aligns with reports of semantic errors in this region under intraoperative DES [61,68,77,80] and is a clear indicator of this area's role in bilateral recruitment for naming through lexico-semantic involvement.

Differences between object naming and action naming were significant in the aSTG, dPrG, mPoG, mPrG, and mSTG for action naming being more easily disturbed in these areas (Table 6). Compared to mere trends for significance in the LH, even more areas were prone to errors in action than in object naming in the RH. The middle and dorsal parts of the primary motor and sensory cortex may once more be crucial for action-related word production, as part of the embodiment theory [66]. Accordingly, mostly lexicosemantic errors were elicited here (Table 5). Both the aSTG and mSTG are so far not known for their strong involvement in the RH during verb tasks, but the mSTG mirrors the consistent reports of involvement during verb tasks in the LH [43,44,49]. Overall, the RH's STG as a semantic hub under intraoperative DES [61,64,68,77,78,80] and nTMS [81] could be sensitive once more to action naming's higher lexico-semantic complexity. Mostly conceptual and lexico-semantic errors arising from this region in our data confirms this interpretation.

In conclusion, the even more pronounced recruitment of the RH in action naming could be rooted in a bilateral conceptual and lexico-semantic knowledge processing [61,68,72–74,77,78,80], manifesting itself in many lexico-semantic errors in these areas. The verb task's higher demand on conceptual and meaning retrieval could make it more sensitive for area detection in the RH. This finding may specifically be visible in the current setup, employing action naming and sentence context.

#### *4.5. Overall Involvement of the Right Hemisphere*

An overall mirrored pattern in the RH as compared to the LH of easily disrupted areas was discovered in our data, in other nTMS studies [6,81], and during intraoperative DES as well as fMRI [61,64,68,72–77,79,80]. However, a clear distinction to LH areas is usually made by the authors, claiming the RH regions may have to be considered languageinvolved in contrast to language-eloquent areas in the LH [5,81,83]. This translates to the distinction that resection of or damage to these language-involved areas would not result in the same drastic language impairments as damage to eloquent counterparts. Their involvement may be secondary.

The nature of nTMS to reveal involved areas and, hence, overcall positive areas is a known phenomenon and can, to some extent, explain the very spread effects in the RH in the present data. Moreover, it cannot be excluded that nTMS can activate long distance interhemispheric connectivity, as has been shown with dual-coil stimulation [84]. Stimulation of involved right hemispheric areas to eloquent left hemispheric parts may have contributed to the current findings.

Two other uncertainties deliver possible explanations for the current RH data. Firstly, a training effect could be assumed. After administering two rounds of mapping with about 140 stimulations necessary to cover all CPS regions three times, the participant has to name each of the approximate 75 items about 10 times. It has been shown that for (novel) verb learning both the LH and to an even greater extent, the RH are recruited in the same area

sensitive in our sample [85]. It can therefore not be excluded that repeated exposure to the same stimulus can result in similar effect as a training effect and thereby is related to the high involvement of the RH in those areas. Secondly, the timing of the nTMS onset in relation to the stimulus to name should be considered. The early onset of 0 ms used here as well as in common protocols is likely to disrupt the earliest stages of naming, namely conceptual retrieval [29,53], while delayed onsets have shown to elicit errors at the sound level [53]. While it is a necessary pre-linguistic step, disrupting this process may deliver conceptually involved areas on top of language-eloquent areas. As conceptual knowledge of a word is thought to be more holistically recruited from bihemispheric regions [68,72–74], the high error rate may be explained by this parameter choice of an early onset of stimulation.

None of these explanations for the RH can be ruled out at this point. A study using different paradigms with varying stimulation onset and potentially even more stimuli to avoid learning effects is needed to clear up the uncertainty between language-eloquent and language-involved areas in the RH.

#### *4.6. Hesitation Errors*

The majority of stimulation protocols refrain from including hesitation errors as positive error occurrences. As of now, no ready-made program is available to identify a hesitation at a subject-tailored level, but currently requires reprocessing of the video material and analysis through third-party programs [45]. This leaves it to the subjective opinion of the experimenter to draw a line between a normal response and a hesitation, when comparing baseline naming to naming under stimulation [37]. Due to the difficult quantification of these errors, separate analyses were conducted, excluding the categories in question: hesitation on the whole sentence as a weaker pronounced no response and hesitation of the target as a weaker pronounced anomia were excluded from the total error count in each task. Doing so decreased the number of errors significantly in both tasks and hemispheres (Table 7). However, even in the remaining errors, action naming demonstrated significantly more errors than object naming.

Two conclusions are to be drawn from this. Firstly, action naming's higher sensitivity is also apparent in a more conservative error count. This strengthens the claim to include the task in mapping. Secondly, it is still important to screen for subtler errors, such as hesitations, as they after all indicate a disruption in language processing [29], and constitute an essential part of the error rates. These errors may be the only elicitable error category in some individuals and, therefore, the only data on which to base a mapping.

#### *4.7. Area-Wise Comparison Excluding Hesitations*

As another approach, we performed an anatomical analysis in which hesitation errors were excluded. Task comparison per CPS regions revealed higher error rates for action naming in the pMFG in the LH and opIFG in the RH (Table 8). The pMFG as part of the prefrontal action related network and the opIFG as a bilateral production area are no surprising components. Since, however, no double dissociation was present and no specific error type pattern is evident in our data, the relations again remain tentative.

When describing the areas with the highest error rates, the following picture arose in both hemi-spheres: the areas with the highest error rate excluding hesitations are by large distributed in a similar pattern as the areas of maps including all errors (Figures 4 and 5, Tables 6 and 8: LH object naming in the vPrG, mMTG, and mPrG; LH action naming in the vPrG, mSTG, and vPoG; RH object naming in the mMTG and vPoG; RH action naming in the mSTG and opIFG). This study did not aim to quantify comparisons of all errors included versus excluding hesitations per CPS region. However, as seen from a descriptive analysis regarding this matter, similar CPS region patterns of high error rates result from both analyses. Hence, including all errors did not deliver any new error rate pattern in the CPS regions. It rather strengthens the error count per area, which would have been revealed in a more conservative map, excluding hesitations. This leads to the

exploratory conclusion that including hesitations does, on the one hand, influence the error count significantly, on the other hand, it does not dull the mapping result by revealing an unexpected area pattern based on these more subjective errors.

#### *4.8. Clinical Implications*

Using a balanced and pretested set of tasks [48], the setup of the present study showed the value of action naming in sentence context as a more sensitive tool than object naming in sentence context under nTMS. This is relevant for clinical application: Cases of entire "zero maps"—no effect at all of stimulation while performing a language task like object naming—are a well-known phenomenon in clinical practice, at least when it comes to certain regions. In these instances, using a more sensitive task, such as action naming, may lead to a positive mapping result.

More specifically, action naming may be the more suitable task to use for these cases, where a tumor is infiltrating areas prone to action naming, such as the mSTG, as seen in the data here and in previous studies [28,43,44,49]; and in frontal regions, as seen in the analysis excluding hesitations (Table 8). Moreover, action naming may be a more accurate tool for RH nTMS mappings, as it was shown to be more sensitive specifically in this hemisphere.

On that premise, it is important to keep the cooperation of the two tasks in mind. We do not suggest that the action naming variant should replace the standard object naming, but it is considered an addition to the test battery. The aforementioned single cases of distinct functional allocation for object naming justify its importance in mapping as much as action naming [28,49,52]. Furthermore, marked aphasia in patients with brain tumors may not allow for testing a more complex task with verbs. In that case, object naming may be required for informative results. Conversely, some cases may demonstrate an inability to perform object naming and might do well during action naming. The ideal setup would, therefore, entail administration of both tasks under nTMS. If fatigue or other time constraining factors play in, one may choose to administer solely action naming.

Regarding error evaluation, clinical practice could opt to screen for more subtle errors such as hesitation on the target that become evident through integration of a lead-in phase. For those patients not demonstrating a strong effect of nTMS, as seen in full speech arrests, categories such as hesitations (on the target) may be the only error source to build on.

#### *4.9. Limitations and Resulting Future Steps*

The current setup only allowed comparing the tasks' relative sensitivity. To fully evaluate the sensitivity and reliability of the nTMS mappings, including tasks in sentence context, data from the gold standard of intraoperative mapping with DES are required. Only this direct comparison allows the conclusion that areas predicted to be positive in action naming and/or object naming proved to be critically involved under the gold standard. The highly invasive nature of DES precludes application in samples of healthy volunteers. Therefore, future studies are needed enrolling a group of patients who undergo both preoperative mapping with nTMS and intraoperative mapping with DES to fully validate this study's protocol.

As a second limitation, subjective measurements of a delayed response were used for defining hesitations. Future studies could profit from built-in response time measurements to establish participant-tailored cut-offs for objectively delayed responses. At present, this requires extensive reprocessing through external software, not compatible with the TMS setup and hence not feasible in the workflow. However, to compensate for this shortcoming, the present study excluded hesitation errors for a separate analysis. Third, relations between error categories and function allocation per CPS region can only tentatively be drawn from the descriptive analyses used in this study. While our conclusions did not arise from a statistical comparison per CPS region, using error ratios per category gave nonetheless clear indication of the prevalent errors. Future investigations tailored for function allocation may tackle this in a bigger sample size.

Moreover, unbalanced handedness resulted in a more heterogeneous participant group than in other studies. However, no correlation between handedness and error rates overall and per hemisphere was obtained. This is in accordance with a low prevalence of RH dominance, in both right- and lefthanders [86–89]. Including left-handers should, therefore, not have altered the findings.

Lastly, we did not look into network effects of mutual areas involved in each task. While it would have shed more light on the interconnection of cortical areas and their dependence of each other, it would not directly help improve the mapping results using different tasks; hence, it exceeded the scope of the current study.

#### **5. Conclusions**

During language mapping of the healthy brain under nTMS, action naming in sentence context proved to be more easily disrupted than object naming in sentence context. Through inclusion of a lead-in phrase and error categorization, it was argued that this difference predominantly arises from the disruption at the lexico-semantic level, where verb retrieval seems to be more affected than noun retrieval. This was observed for both the left and the right hemisphere and the pattern persisted in a more conservative error count, excluding hesitation errors. The role of hesitation errors was more clearly defined as contributing significantly to the error rates, while, at the same time, not dulling the mapping results and, thus, worth including.

**Author Contributions:** A.-K.O., R.B., N.S. and S.M.K. conceived and designed the study. A.-K.O., C.N., S.S. and A.S. collected and analyzed the data. A.-K.O. prepared the manuscript, with contributions and editing by all authors. All authors have read and agreed to the published version of the manuscript.

**Funding:** The study was financed by institutional grants from the Department of Neurosurgery at the Klinikum rechts der Isar, Technical University of Munich, Munich, Germany, and received no external funding.

**Institutional Review Board Statement:** The study was approved by the local ethics committee (Institutional Review Board of the Technical University of Munich; registration number: 202/18 S) and was conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all subjects prior to study inclusion.

**Informed Consent Statement:** Written informed consent was obtained from the participant(s) to publish this paper.

**Data Availability Statement:** The datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

**Acknowledgments:** We gratefully acknowledge all participants for their time and efforts partaking in this study. Roelien Bastiaanse is partially supported by the Center for Language and Brain NRU Higher School of Economics, RF Government Grant, ag. No. 14.641.31.0004.

**Conflicts of Interest:** S.M.K. is a consultant for Brainlab AG (Munich, Germany). N.S. received honoraria from Nexstim Plc (Heldinki, Finland). None of the authors report conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

#### **References**


*Article*
