Dissociating language and thought in large language models.

Large language models (LLMs) have come closest among all models to date to mastering human language, yet opinions about their linguistic and cognitive capabilities remain split. Here, we evaluate LLMs using a distinction between formal linguistic competence (knowledge of linguistic rules and patterns) and functional linguistic competence (understanding and using language in the world). We ground this distinction in human neuroscience, which has shown that formal and functional competence rely on different neural mechanisms. Although LLMs are surprisingly good at formal competence, their performance on functional competence tasks remains spotty and often requires specialized fine-tuning and/or coupling with external modules. We posit that models that use language in human-like ways would need to master both of these competence types, which, in turn, could require the emergence of separate mechanisms specialized for formal versus functional linguistic competence.

Functional characterization of the language network of polyglots and hyperpolyglots with precision fMRI.

How do polyglots-individuals who speak five or more languages-process their languages, and what can this population tell us about the language system? Using fMRI, we identified the language network in each of 34 polyglots (including 16 hyperpolyglots with knowledge of 10+ languages) and examined its response to the native language, non-native languages of varying proficiency, and unfamiliar languages. All language conditions engaged all areas of the language network relative to a control condition. Languages that participants rated as higher proficiency elicited stronger responses, except for the native language, which elicited a similar or lower response than a non-native language of similar proficiency. Furthermore, unfamiliar languages that were typologically related to the participants' high-to-moderate-proficiency languages elicited a stronger response than unfamiliar unrelated languages. The results suggest that the language network's response magnitude scales with the degree of engagement of linguistic computations (e.g. related to lexical access and syntactic-structure building). We also replicated a prior finding of weaker responses to native language in polyglots than non-polyglot bilinguals. These results contribute to our understanding of how multiple languages coexist within a single brain and provide new evidence that the language network responds more strongly to stimuli that more fully engage linguistic computations.

Watching videos of a drawing hand improves students' understanding of the normal probability distribution.

Understanding normal probability distributions is a crucial objective in mathematics and statistics education. Drawing upon cognitive psychology research, this study explores the use of drawings and visualizations as effective scaffolds to enhance students' comprehension. Although much research has documented the helpfulness of drawing as a research tool to reveal students' knowledge states, its direct utility in advancing higher-order cognitive processes remains understudied. In Study 1, qualitative methods were utilized to identify common misunderstandings among students regarding canonical depictions of the normal probability distribution. Building on these insights, Study 2 experimentally compared three instructional videos (static slides, dynamic drawing, and dynamic drawings done by a visible hand). The hand drawing video led to better learning than the other versions. Study 3 examined whether the benefits from observing a hand drawing could be reproduced by a dynamic cursor moving around otherwise static slides (without the presence of a hand). Results showed no significant learning difference between observing a hand drawing and a moving cursor, both outperforming a control. This research links the cognitive process of drawing with its educational role and provides insights into its potential to enhance memory, cognition, and inform instructional methods.

Functional characterization of the language network of polyglots and hyperpolyglots with precision fMRI.

How do polyglots-individuals who speak five or more languages-process their languages, and what can this population tell us about the language system? Using fMRI, we identified the language network in each of 34 polyglots (including 16 hyperpolyglots with knowledge of 10+ languages) and examined its response to the native language, non-native languages of varying proficiency, and unfamiliar languages. All language conditions engaged all areas of the language network relative to a control condition. Languages that participants rated as higher-proficiency elicited stronger responses, except for the native language, which elicited a similar or lower response than a non-native language of similar proficiency. Furthermore, unfamiliar languages that were typologically related to the participants' high-to-moderate-proficiency languages elicited a stronger response than unfamiliar unrelated languages. The results suggest that the language network's response magnitude scales with the degree of engagement of linguistic computations (e.g., related to lexical access and syntactic-structure building). We also replicated a prior finding of weaker responses to native language in polyglots than non-polyglot bilinguals. These results contribute to our understanding of how multiple languages co-exist within a single brain and provide new evidence that the language network responds more strongly to stimuli that more fully engage linguistic computations.

What are large language models supposed to model?

Do large language models (LLMs) constitute a computational account of how humans process language? And if so, what is the role of (psycho)linguistic theory in understanding the relationship between artificial and biological minds? The answer depends on choosing among several, fundamentally distinct ways of interpreting these models as hypotheses about humans.

Robust effects of working memory demand during naturalistic language comprehension in language-selective cortex.

To understand language, we must infer structured meanings from real-time auditory or visual signals. Researchers have long focused on word-by-word structure building in working memory as a mechanism that might enable this feat. However, some have argued that language processing does not typically involve rich word-by-word structure building, and/or that apparent working memory effects are underlyingly driven by surprisal (how predictable a word is in context). Consistent with this alternative, some recent behavioral studies of naturalistic language processing that control for surprisal have not shown clear working memory effects. In this fMRI study, we investigate a range of theory-driven predictors of word-by-word working memory demand during naturalistic language comprehension in humans of both sexes under rigorous surprisal controls. In addition, we address a related debate about whether the working memory mechanisms involved in language comprehension are language-specialized or domain-general. To do so, in each participant, we functionally localize (a) the language-selective network and (b) the 'multiple demand' network, which supports working memory across domains. Results show robust surprisal-independent effects of memory demand in the language network and no effect of memory demand in the multiple demand network. Our findings thus support the view that language comprehension involves computationally demanding word-by-word structure building operations in working memory, in addition to any prediction-related mechanisms. Further, these memory operations appear to be primarily carried out by the same neural resources that store linguistic knowledge, with no evidence of involvement of brain regions known to support working memory across domains.SIGNIFICANCE STATEMENT:This study uses fMRI to investigate signatures of working memory (WM) demand during naturalistic story listening, using a broad range of theoretically motivated estimates of WM demand. Results support a strong effect of WM demand in the brain that is distinct from effects of word predictability. Further, these WM demands register primarily in language-selective regions, rather than in 'multiple demand' regions that have previously been associated with WM in non-linguistic domains. Our findings support a core role for WM in incremental language processing, using WM resources that are specialized for language.

Probabilistic atlas for the language network based on precision fMRI data from >800 individuals.

Two analytic traditions characterize fMRI language research. One relies on averaging activations across individuals. This approach has limitations: because of inter-individual variability in the locations of language areas, any given voxel/vertex in a common brain space is part of the language network in some individuals but in others, may belong to a distinct network. An alternative approach relies on identifying language areas in each individual using a functional 'localizer'. Because of its greater sensitivity, functional resolution, and interpretability, functional localization is gaining popularity, but it is not always feasible, and cannot be applied retroactively to past studies. To bridge these disjoint approaches, we created a probabilistic functional atlas using fMRI data for an extensively validated language localizer in 806 individuals. This atlas enables estimating the probability that any given location in a common space belongs to the language network, and thus can help interpret group-level activation peaks and lesion locations, or select voxels/electrodes for analysis. More meaningful comparisons of findings across studies should increase robustness and replicability in language research.

Differential Tracking of Linguistic vs. Mental State Content in Naturalistic Stimuli by Language and Theory of Mind (ToM) Brain Networks.

Language and social cognition, especially the ability to reason about mental states, known as theory of mind (ToM), are deeply related in development and everyday use. However, whether these cognitive faculties rely on distinct, overlapping, or the same mechanisms remains debated. Some evidence suggests that, by adulthood, language and ToM draw on largely distinct-though plausibly interacting-cortical networks. However, the broad topography of these networks is similar, and some have emphasized the importance of social content / communicative intent in the linguistic signal for eliciting responses in the language areas. Here, we combine the power of individual-subject functional localization with the naturalistic-cognition inter-subject correlation approach to illuminate the language-ToM relationship. Using functional magnetic resonance imaging (fMRI), we recorded neural activity as participants (n = 43) listened to stories and dialogues with mental state content (+linguistic, +ToM), viewed silent animations and live action films with mental state content but no language (-linguistic, +ToM), or listened to an expository text (+linguistic, -ToM). The ToM network robustly tracked stimuli rich in mental state information regardless of whether mental states were conveyed linguistically or non-linguistically, while tracking a +linguistic / -ToM stimulus only weakly. In contrast, the language network tracked linguistic stimuli more strongly than (a) non-linguistic stimuli, and than (b) the ToM network, and showed reliable tracking even for the linguistic condition devoid of mental state content. These findings suggest that in spite of their indisputably close links, language and ToM dissociate robustly in their neural substrates-and thus plausibly cognitive mechanisms-including during the processing of rich naturalistic materials.

The Small and Efficient Language Network of Polyglots and Hyper-polyglots.

Acquiring a foreign language is challenging for many adults. Yet certain individuals choose to acquire sometimes dozens of languages and often just for fun. Is there something special about the minds and brains of such polyglots? Using robust individual-level markers of language activity, measured with fMRI, we compared native language processing in polyglots versus matched controls. Polyglots (n = 17, including nine "hyper-polyglots" with proficiency in 10-55 languages) used fewer neural resources to process language: Their activations were smaller in both magnitude and extent. This difference was spatially and functionally selective: The groups were similar in their activation of two other brain networks-the multiple demand network and the default mode network. We hypothesize that the activation reduction in the language network is experientially driven, such that the acquisition and use of multiple languages makes language processing generally more efficient. However, genetic and longitudinal studies will be critical to distinguish this hypothesis from the one whereby polyglots' brains already differ at birth or early in development. This initial characterization of polyglots' language network opens the door to future investigations of the cognitive and neural architecture of individuals who gain mastery of multiple languages, including changes in this architecture with linguistic experiences.

Activity in the fronto-parietal multiple-demand network is robustly associated with individual differences in working memory and fluid intelligence.

Numerous brain lesion and fMRI studies have linked individual differences in executive abilities and fluid intelligence to brain regions of the fronto-parietal "multiple-demand" (MD) network. Yet, fMRI studies have yielded conflicting evidence as to whether better executive abilities are associated with stronger or weaker MD activations and whether this relationship is restricted to the MD network. Here, in a large-sample (n = 216) fMRI investigation, we found that stronger activity in MD regions - functionally defined in individual participants - was robustly associated with more accurate and faster responses on a spatial working memory task performed in the scanner, as well as fluid intelligence measured independently (n = 114). In line with some prior claims about a relationship between language and fluid intelligence, we also found a weak association between activity in the brain regions of the left fronto-temporal language network during an independent passive reading task, and performance on the working memory task. However, controlling for the level of MD activity abolished this relationship, whereas the MD activity-behavior association remained highly reliable after controlling for the level of activity in the language network. Finally, we demonstrate how unreliable MD activity measures, coupled with small sample sizes, could falsely lead to the opposite, negative, association that has been reported in some prior studies. Taken together, these results demonstrate that a core component of individual differences variance in executive abilities and fluid intelligence is selectively and robustly positively associated with the level of activity in the MD network, a result that aligns well with lesion studies.

No evidence for differences among language regions in their temporal receptive windows.

The "core language network" consists of left frontal and temporal regions that are selectively engaged in linguistic processing. Whereas functional differences among these regions have long been debated, many accounts propose distinctions in terms of representational grain-size-e.g., words vs. phrases/sentences-or processing time-scale, i.e., operating on local linguistic features vs. larger spans of input. Indeed, the topography of language regions appears to overlap with a cortical hierarchy reported by Lerner et al. (2011) wherein mid-posterior temporal regions are sensitive to low-level features of speech, surrounding areas-to word-level information, and inferior frontal areas-to sentence-level information and beyond. However, the correspondence between the language network and this hierarchy of "temporal receptive windows" (TRWs) is difficult to establish because the precise anatomical locations of language regions vary across individuals. To directly test this correspondence, we first identified language regions in each participant with a well-validated task-based localizer, which confers high functional resolution to the study of TRWs (traditionally based on stereotactic coordinates); then, we characterized regional TRWs with the naturalistic story listening paradigm of Lerner et al. (2011), which augments task-based characterizations of the language network by more closely resembling comprehension "in the wild". We find no region-by-TRW interactions across temporal and inferior frontal regions, which are all sensitive to both word-level and sentence-level information. Therefore, the language network as a whole constitutes a unique stage of information integration within a broader cortical hierarchy.

Broca's Area Is Not a Natural Kind.

Theories of human cognition prominently feature 'Broca's area', which causally contributes to a myriad of mental functions. However, Broca's area is not a monolithic, multipurpose unit - it is structurally and functionally heterogeneous. Some functions engaging (subsets of) this area share neurocognitive resources, whereas others rely on separable circuits. A decade of converging evidence has now illuminated a fundamental distinction between two subregions of Broca's area that likely play computationally distinct roles in cognition: one belongs to the domain-specific 'language network', the other to the domain-general 'multiple-demand (MD) network'. Claims about Broca's area should be (re)cast in terms of these (and other, as yet undetermined) functional components, to establish a cumulative research enterprise where empirical findings can be replicated and theoretical proposals can be meaningfully compared and falsified.

An Attempt to Conceptually Replicate the Dissociation between Syntax and Semantics during Sentence Comprehension.

Is sentence structure processed by the same neural and cognitive resources that are recruited for processing word meanings, or do structure and meaning rely on distinct resources? Linguistic theorizing and much behavioral evidence suggest tight integration between lexico-semantic and syntactic representations and processing. However, most current proposals of the neural architecture of language continue to postulate a distinction between the two. One of the earlier and most cited pieces of neuroimaging evidence in favor of this dissociation comes from a paper by Dapretto and Bookheimer (1999). Using a sentence-meaning judgment task, Dapretto & Bookheimer observed two distinct peaks within the left inferior frontal gyrus (LIFG): one was more active during a lexico-semantic manipulation, and the other during a syntactic manipulation. Although the paper is highly cited, no attempt has been made, to our knowledge, to replicate the original finding. We report an fMRI study that attempts to do so. Using a combination of whole-brain, group-level ROI, and participant-specific functional ROI approaches, we fail to replicate the original dissociation. In particular, whereas parts of LIFG respond reliably more strongly during lexico-semantic than syntactic processing, no part of LIFG (including in the region defined around the peak reported by Dapretto & Bookheimer) shows the opposite pattern. We speculate that the original result was a false positive, possibly driven by a small subset of participants or items that biased a fixed-effects analysis with low power.

Functionally distinct language and Theory of Mind networks are synchronized at rest and during language comprehension.

Communication requires the abilities to generate and interpret utterances and to infer the beliefs, desires, and goals of others ("Theory of Mind"; ToM). These two abilities have been shown to dissociate: individuals with aphasia retain the ability to think about others' mental states; and individuals with autism are impaired in social reasoning, but their basic language processing is often intact. In line with this evidence from brain disorders, functional MRI (fMRI) studies have shown that linguistic and ToM abilities recruit distinct sets of brain regions. And yet, language is a social tool that allows us to share thoughts with one another. Thus, the language and ToM brain networks must share information despite being implemented in distinct neural circuits. Here, we investigated potential interactions between these networks during naturalistic cognition using functional correlations in fMRI. The networks were functionally defined in individual participants, in terms of preference for sentences over nonwords for language, and for belief inference over physical-event processing for ToM, with both a verbal and a nonverbal paradigm. Although, across experiments, interregion correlations within each network were higher than between-network correlations, we also observed above-baseline synchronization of blood oxygenation level-dependent signal fluctuations between the two networks during rest and story comprehension. This synchronization was functionally specific: neither network was synchronized with the executive control network (functionally defined in terms of preference for a harder over easier version of an executive task). Thus, coordination between the language and ToM networks appears to be an inherent and specific characteristic of their functional architecture. NEW & NOTEWORTHY Humans differ from nonhuman primates in their abilities to communicate linguistically and to infer others' mental states. Although linguistic and social abilities appear to be interlinked onto- and phylogenetically, they are dissociated in the adult human brain. Yet successful communication requires language and social reasoning to work in concert. Using functional MRI, we show that language regions are synchronized with social regions during rest and language comprehension, pointing to a possible mechanism for internetwork interaction.

Can neuroimaging help aphasia researchers? Addressing generalizability, variability, and interpretability.

Neuroimaging studies of individuals with brain damage seek to link brain structure and activity to cognitive impairments, spontaneous recovery, or treatment outcomes. To date, such studies have relied on the critical assumption that a given anatomical landmark corresponds to the same functional unit(s) across individuals. However, this assumption is fallacious even across neurologically healthy individuals. Here, we discuss the severe implications of this issue, and argue for an approach that circumvents it, whereby: (i) functional brain regions are defined separately for each subject using fMRI, allowing for inter-individual variability in their precise location; (ii) the response profile of these subject-specific regions are characterized using various other tasks; and (iii) the results are averaged across individuals, guaranteeing generalizabliity. This method harnesses the complementary strengths of single-case studies and group studies, and it eliminates the need for post hoc "reverse inference" from anatomical landmarks back to cognitive operations, thus improving data interpretability.

Domain-General Brain Regions Do Not Track Linguistic Input as Closely as Language-Selective Regions.

Language comprehension engages a cortical network of left frontal and temporal regions. Activity in this network is language-selective, showing virtually no modulation by nonlinguistic tasks. In addition, language comprehension engages a second network consisting of bilateral frontal, parietal, cingulate, and insular regions. Activity in this "multiple demand" (MD) network scales with comprehension difficulty, but also with cognitive effort across a wide range of nonlinguistic tasks in a domain-general fashion. Given the functional dissociation between the language and MD networks, their respective contributions to comprehension are likely distinct, yet such differences remain elusive. Prior neuroimaging studies have suggested that activity in each network covaries with some linguistic features that, behaviorally, influence on-line processing and comprehension. This sensitivity of the language and MD networks to local input characteristics has often been interpreted, implicitly or explicitly, as evidence that both networks track linguistic input closely, and in a manner consistent across individuals. Here, we used fMRI to directly test this assumption by comparing the BOLD signal time courses in each network across different people (n = 45, men and women) listening to the same story. Language network activity showed fewer individual differences, indicative of closer input tracking, whereas MD network activity was more idiosyncratic and, moreover, showed lower reliability within an individual across repetitions of a story. These findings constrain cognitive models of language comprehension by suggesting a novel distinction between the processes implemented in the language and MD networks.SIGNIFICANCE STATEMENT Language comprehension recruits both language-specific mechanisms and domain-general mechanisms that are engaged in many cognitive processes. In the human cortex, language-selective mechanisms are implemented in the left-lateralized "core language network", whereas domain-general mechanisms are implemented in the bilateral "multiple demand" (MD) network. Here, we report the first direct comparison of the respective contributions of these networks to naturalistic story comprehension. Using a novel combination of neuroimaging approaches we find that MD regions track stories less closely than language regions. This finding constrains the possible contributions of the MD network to comprehension, contrasts with accounts positing that this network has continuous access to linguistic input, and suggests a new typology of comprehension processes based on their extent of input tracking.