Functional differentiation in the language network revealed by lesion-symptom mapping
This article has been Reviewed by the following groups
Discuss this preprint
Start a discussion What are Sciety discussions?Listed in
- Evaluated articles (eLife)
Abstract
Theories of language organization in the brain commonly posit that different regions underlie distinct linguistic mechanisms. However, such theories have been criticized on the grounds that many neuroimaging studies of language processing find similar effects across regions. Moreover, condition by region interaction effects, which provide the strongest evidence of functional differentiation between regions, have rarely been offered in support of these theories. Here we address this by using lesion-symptom mapping in three large, partially-overlapping groups of aphasia patients with left hemisphere brain damage due to stroke (N=121, N=92, N= 218). We identified multiple measure by region interaction effects, associating damage to the posterior middle temporal gyrus with syntactic comprehension deficits, damage to posterior inferior frontal gyrus with expressive agrammatism, and damage to inferior angular gyrus with semantic category word fluency deficits. Our results are inconsistent with recent hypotheses that regions of the language network are undifferentiated with respect to high-level linguistic processing.
Article activity feed
-
###Reviewer #3:
Summary
This study used the method of lesion-symptom mapping to disassociate the neural correlates underlying syntactic and semantic functions. The results suggest that different brain regions of the language network do not share similar functions; instead, they should perform different high-level functions that contribute to linguistic processing. Specifically, the pMTG and the aSTS were found associated with syntactic comprehension; the pIFG and the aIFG were found associated with expressive agrammatism; and the iAG was found associated with semantic category word fluency. Overall, I find the research question interesting. However, I have some doubt on the methodology, and the interpretation of experimental results, though not implausible, was somehow hasty. I'll elaborate below.
Detailed comments:
- The fundamental …
###Reviewer #3:
Summary
This study used the method of lesion-symptom mapping to disassociate the neural correlates underlying syntactic and semantic functions. The results suggest that different brain regions of the language network do not share similar functions; instead, they should perform different high-level functions that contribute to linguistic processing. Specifically, the pMTG and the aSTS were found associated with syntactic comprehension; the pIFG and the aIFG were found associated with expressive agrammatism; and the iAG was found associated with semantic category word fluency. Overall, I find the research question interesting. However, I have some doubt on the methodology, and the interpretation of experimental results, though not implausible, was somehow hasty. I'll elaborate below.
Detailed comments:
- The fundamental reasoning underlying the method of lesion-symptom mapping.
I agree with the paper that high-level linguistics functions are intertwined in language performance (in language comprehension and production), and any manipulation of syntax is likely to affect semantic interpretation as well. However, it seems problematic to claim that this conundrum can be solved with the help of lesion-symptom mapping, and that lesion-symptom mapping can identify brain regions "causally" involved in linguistic functions.
Suppose that the execution of function X crucially depends on two other functions Y and Z, while function Z also causally depends on function Y. I doubt we can discover this kind of causal network from lesion-symptom mapping. In other words, simply detecting the correlation between a lesion area and the performance of a certain linguistic task is still far from detecting the actual causal dependence between a certain brain region and a certain linguistic function. Therefore, I think the paper should avoid overclaims and include more details on how the specific procedures of the current study led to contributions "towards" revealing the general or language-specific function of a brain region.
Y → Z
↓ ↙
X
- Methodological details of this paper.
This issue is also related to the previous one. It seems that the assignment of the two groups of participants was based on some other studies. The specific lesion-mapping procedures adopted in this paper also followed some other studies. Though I understand that there might be some word limits for the submission, I still hope that (i) the paper includes more methodological details on these, so that the paper can be better self-contained, and (ii) some explanations are given on how these procedures led to contributions "towards" revealing the general or language-specific function of a brain region.
- The interpretation of results.
The behavioral tasks used in this study, namely the comprehension of sentences with non-canonical word order, the description of pictures, and the naming of animal names, are associated with three kinds of linguistic functions: syntactic comprehension, expressive agrammatism, and semantic category word fluency. There might be alternatives to interpret these three linguistic tasks: e.g., (i) sentence-level processing vs. discourse-level processing vs. word-level processing; (ii) syntax vs. pragmatics vs. lexical ability; etc. The interpretation of results can include a discussion on these.
- How the findings were consistent with the theory proposed in Matchin & Hickok (2020)
I read the paper of Matchin & Hickok (2020) ("The cortical organization of syntax", Cerebral Cortex), and found some discrepancies between the theory proposed in that paper and the finding from the current experiment. In that paper, the pMTG is associated with the lexical-syntactic function, underlying both language production and comprehension, while the pIFG is associated with linearization, underlying specifically language production. In the current study, the association between the pMTG and syntactic comprehension seems to suggest that the pMTG is specifically related to the processing of sentences with non-canonical order. Isn't the processing of this kind of sentences an issue related to linearization, not issues related to argument structure or other lexical-syntactic issues?
-
###Reviewer #2:
This paper attempts to disentangle the neural instantiation of syntax and semantics using VLSM correlations between regions of brain-damaged tissue and language performance across three tasks in relatively large groups of stroke patients. Although the work addresses an important, and currently debated, issue in cognitive neuroscience, the paper is significantly methodologically flawed and the results are untenable.
Major problems:
- Independent measures. Three tasks were used to index (1) syntactic comprehension, (2) expressive agrammatism, and (3) semantic processing. All are problematic and reliability of measurement was not addressed for any of the tasks. This is particularly problematic for expressive agrammatism, but is of concern for all measures.
For syntactic comprehension, a combined score reflecting comprehension …
###Reviewer #2:
This paper attempts to disentangle the neural instantiation of syntax and semantics using VLSM correlations between regions of brain-damaged tissue and language performance across three tasks in relatively large groups of stroke patients. Although the work addresses an important, and currently debated, issue in cognitive neuroscience, the paper is significantly methodologically flawed and the results are untenable.
Major problems:
- Independent measures. Three tasks were used to index (1) syntactic comprehension, (2) expressive agrammatism, and (3) semantic processing. All are problematic and reliability of measurement was not addressed for any of the tasks. This is particularly problematic for expressive agrammatism, but is of concern for all measures.
For syntactic comprehension, a combined score reflecting comprehension of three complex sentence types with long-distance dependencies (wh-movement constructions) were contrasted with scores for active sentences. This contrast is linguistically unfounded: it is not possible to isolate syntactic process using this contrast, since there are critical differences between the experimental and control sentences on several variables, beyond syntactic processing, including the number of propositions, lexical-semantics, sentence length, etc. as well as domain-general processes, etc. For any studies seeking to determine the cognitive and/or neural resources engaged for syntactic processing, a fundamental requirement is that experimental conditions consist of pairs of stimuli that differ along a single dimension - the dimension of interest - with all else kept constant across conditions, lest the comparison be confounded by additional dimension(s) (cf. Grodzinsky, 2010, for discussion). To do so in the present study the non-canonical forms would need to be contrasted with their canonical counterparts, e.g., subject-relatives for object-relatives, subject questions for object questions, etc.
Expressive agrammatism was determined based on samples of connected speech elicited by picture description or story retelling and the "presence of expressive agrammatism was . . . rated by speech and language experts . . ." This is problematic. Subjective judgement is insufficient for a study of the scope reported. Objective analysis of the speech samples is needed to quantify salient dimensions of agrammatism or, better, inclusion of a constrained task, like that used to quantify sentence comprehension is recommended.
A very gross measure of "semantic" processing was used - a word fluency task. This is arguably not a semantic task and no rationale for using it is provided. Given this, the title of the paper is inappropriate and misleading: ". . . dissociations of syntax and semantics . . .". It also is stated that assessment occurred at "a variable number of timepoints". Why? When were the time points? Were there any intervening variables between time points? Why was performance "averaged" over samples? In what way does this make the data more "reliable"? Were all participants beyond the period of spontaneous recovery (this is not evident based on data presented in Table 1)?
Dependent measures. Six ROIs were selected for analysis and the rationale for their selection is based on one model of sentence processing. There are two main issues here: (1) there is no rational for using an ROI rather than a voxel-based approach; of the two approaches, a voxel-based approach is the most rigorous as ROI analyses may lead to spurious results simply based on the ROIs selected, (2) the voxel-wise analyses were uncorrected; tables reporting the coordinates derived from voxel-wise analyses are needed; the corrected voxel-wise analyses (with corresponding data tables) should replace the ROI analysis at least for first-pass analyses, (3) greater motivation/justification for selection of the 6 ROIs is needed; there are well-known and well-conceptualized data-based models of sentence processing that include ROIs other than the six tested, e.g., pSTG/pSTS (Friederici, 2012, 2018; Friederici & Gierhan, 2013; Bornkessel-Schlesewsky & Schlesewsky, 2013; Bornkessel-Schlesewsky et al., 2012). It is questioned why the authors overlook this important body of work? ROI selection could be better motivated based on data derived from well-controlled studies of syntactic and semantic processing (e.g., for syntactic processing: Bahlmann et al., 2007; Bornkessel et al., 2005; Bornkessel-Schlesewsky et al., 2010; Constable et al., 2004; Fieback et al., 2005; Friederici et al., 2006; Meltzer et al., 2010; Sonti & Grodzinsky, 2010; Thompson et al., 2010). In addition, there are several published meta-analyses within these domains that would better elucidate appropriate ROIs.
Discussion/conclusions. Several statements in this section are overstatements, not supported by the study:
a) "Research critically needs to incorporate insights from lesions symptom mapping in order to understand the architecture of language...". Why? Lesioned brains arguably have undergone reorganization (particularly in chronic stroke). This issue is not addressed in the paper.
b) "...results are ...consistent with neuroanatomical models that posit distinct syntactic and semantic functions to different regions...". It is not possible to determine precise functions of brain regions based on lesioned tissue. The only conclusion that can be drawn is that the infarcted region is involved in and may disrupt the function of interest, but it cannot be said that it is responsible for it. Such an assertion fails to recognize the well-known fact that brain regions do not work in isolation, rather a network of regions is required for execution of complex tasks.
c) "The [Matchin & Hickok] model posits that the ...pMTG is critical for processing hierarchical structure for production and comprehension.". The data presented do not address or support this claim.
d) "Damage to the pMTG was significantly associated with semantic comprehension deficits...". Semantic comprehension was not tested.
e) "damage to the pIFG was ...associated with agrammatic speech deficits". This observation, albeit unreliable based on limitations of the method used for quantifying agrammatism, does not support the M&H model; the authors claim that it does in spite of the fact that there was a "marginally" significant interaction between IFG and MTG.
Given the substantial methodological limitations inherent in this study, the results and conclusions are unreliable.
-
###Reviewer #1:
This is a lesion-symptom mapping study of syntactic comprehension, syntactic production, and a semantic measure, namely category word fluency. The authors argue that each of these language functions depends on a different brain region. With some revision this paper could be a worthwhile contribution to the literature, but in my opinion it largely replicates prior work, and the aspects in which it attempts to go beyond prior work are not very strong.
The links between the brain regions and linguistic functions studied here have all been firmly established already. For the IFG and agrammatism, the authors cite two papers from their own work and two from other labs that already make this case (p. 8). For the pMTG and receptive syntax, there are many previous findings, most of which are cited in the present paper and/or the …
###Reviewer #1:
This is a lesion-symptom mapping study of syntactic comprehension, syntactic production, and a semantic measure, namely category word fluency. The authors argue that each of these language functions depends on a different brain region. With some revision this paper could be a worthwhile contribution to the literature, but in my opinion it largely replicates prior work, and the aspects in which it attempts to go beyond prior work are not very strong.
The links between the brain regions and linguistic functions studied here have all been firmly established already. For the IFG and agrammatism, the authors cite two papers from their own work and two from other labs that already make this case (p. 8). For the pMTG and receptive syntax, there are many previous findings, most of which are cited in the present paper and/or the authors' 2020 review paper; Pillay et al. (2017) is a particularly compelling study reporting this association. Semantic fluency has previously been associated with inferior parietal cortex by Baldo et al. (2006), also appropriately cited in the present paper. In sum, none of the major findings of the present study are novel.
The most novel aspect of this study is that the authors carry out some interaction analyses, which indeed are often not carried out when they should be when making claims about differential roles of different brain areas. But the value of this is undercut by the fact that these interaction analyses are still based on univariate analyses of lesion-behavior relationships in each region. The fact that many lesions to one region will extend to one or more of the other regions is simply ignored (as in most VLSM studies). This unrealistic model is just inherently limited (Mah et al., 2014). A multivariate approach to lesion-symptom mapping would be needed to make progress in teasing out differential contributions of different regions. Furthermore, one of the three interactions is not statistically significant, and another one (involving the semantic measure) is not well motivated because the authors present no analysis of the category fluency task, and therefore no principled reason to expect it to be associated with one or another semantic region. Regarding that finding, they end up making a reverse inference on p. 9, and although they cite Schwartz et al. (2011), they don't explain that that paper already showed differential roles of these two regions in a lesion-symptom mapping study. Finally, there are no interactions that actually address the segregation of syntax and semantics promised in the title.
Some other issues to consider:
Speech rate is used as a covariate to control for non-semantic factors influencing category word fluency, but it cannot possibly serve that purpose. There are many factors influencing speech rate, especially motor factors, and completely different factors contributing to word fluency performance, especially executive. The bottom line is category word fluency is really not a very helpful measure because there are too many contributing factors.
There seems to be inadequate lesion coverage in the TP ROI.
Although the uncorrected voxelwise maps are reassuring with respect to the main ROI analysis, the fact that they are uncorrected means that they don't really have any evidentiary value.
It is problematic to combine two sentence comprehension measures without showing that they are on an identical scale or adjusting them accordingly.
-
##Preprint Review
This preprint was reviewed using eLife’s Preprint Review service, which provides public peer reviews of manuscripts posted on bioRxiv for the benefit of the authors, readers, potential readers, and others interested in our assessment of the work. This review applies only to version 2 of the manuscript.
###Summary:
The reviewers feel that the authors are addressing an interesting and important issue in cognitive neuroscience. Nevertheless, serious shortcomings in methods and analytic approaches, and in interpretation, were flagged by all three reviewers.
-
