Precision fMRI reveals that the language-selective network supports both phrase-structure building and lexical access during language production

Language in Brains, Minds, and Machines

It has long been argued that only humans could produce and understand language. But now, for the first time, artificial language models (LMs) achieve this feat. Here we survey the new purchase LMs are providing on the question of how language is implemented in the brain. We discuss why, a priori, LMs might be expected to share similarities with the human language system. We then summarize evidence that LMs represent linguistic information similarly enough to humans to enable relatively accurate brain encoding and decoding during language processing. Finally, we examine which LM properties-their architecture, task performance, or training-are critical for capturing human neural responses to language and review studies using LMs as in silico model organisms for testing hypotheses about language. These ongoing investigations bring us closer to understanding the representations and processes that underlie our ability to comprehend sentences and express thoughts in language.

Linguistic inputs must be syntactically parsable to fully engage the language network

Human language comprehension is remarkably robust to ill-formed inputs (e.g., word transpositions). This robustness has led some to argue that syntactic parsing is largely an illusion, and that incremental comprehension is more heuristic, shallow, and semantics-based than is often assumed. However, the available data are also consistent with the possibility that humans always perform rule-like symbolic parsing and simply deploy error correction mechanisms to reconstruct ill-formed inputs when needed. We put these hypotheses to a new stringent test by examining brain responses to a) stimuli that should pose a challenge for syntactic reconstruction but allow for complex meanings to be built within local contexts through associative/shallow processing (sentences presented in a backward word order), and b) grammatically well-formed but semantically implausible sentences that should impede semantics-based heuristic processing. Using a novel behavioral syntactic reconstruction paradigm, we demonstrate that backward-presented sentences indeed impede the recovery of grammatical structure during incremental comprehension. Critically, these backward-presented stimuli elicit a relatively low response in the language areas, as measured with fMRI. In contrast, semantically implausible but grammatically well-formed sentences elicit a response in the language areas similar in magnitude to naturalistic (plausible) sentences. In other words, the ability to build syntactic structures during incremental language processing is both necessary and sufficient to fully engage the language network. Taken together, these results provide strongest to date support for a generalized reliance of human language comprehension on syntactic parsing.

Tracking Components of Bilingual Language Control in Speech Production: An fMRI Study Using Functional Localizers

When bilingual speakers switch back to speaking in their native language (L1) after having used their second language (L2), they often experience difficulty in retrieving words in their L1. This phenomenon is referred to as the L2 after-effect. We used the L2 after-effect as a lens to explore the neural bases of bilingual language control mechanisms. Our goal was twofold: first, to explore whether bilingual language control draws on domain-general or language-specific mechanisms; second, to investigate the precise mechanism(s) that drive the L2 after-effect. We used a precision fMRI approach based on functional localizers to measure the extent to which the brain activity that reflects the L2 after-effect overlaps with the language network (Fedorenko et al., 2010) and the domain-general multiple demand network (Duncan, 2010), as well as three task-specific networks that tap into interference resolution, lexical retrieval, and articulation. Forty-two Polish-English bilinguals participated in the study. Our results show that the L2 after-effect reflects increased engagement of domain-general but not language-specific resources. Furthermore, contrary to previously proposed interpretations, we did not find evidence that the effect reflects increased difficulty related to lexical access, articulation, and the resolution of lexical interference. We propose that difficulty of speech production in the picture naming paradigm-manifested as the L2 after-effect-reflects interference at a nonlinguistic level of task schemas or a general increase of cognitive control engagement during speech production in L1 after L2.

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.

Distributed Sensitivity to Syntax and Semantics throughout the Language Network

Human language is expressive because it is compositional: The meaning of a sentence (semantics) can be inferred from its structure (syntax). It is commonly believed that language syntax and semantics are processed by distinct brain regions. Here, we revisit this claim using precision fMRI methods to capture separation or overlap of function in the brains of individual participants. Contrary to prior claims, we find distributed sensitivity to both syntax and semantics throughout a broad frontotemporal brain network. Our results join a growing body of evidence for an integrated network for language in the human brain within which internal specialization is primarily a matter of degree rather than kind, in contrast with influential proposals that advocate distinct specialization of different brain areas for different types of linguistic functions.

Language is primarily a tool for communication rather than thought

Language is a defining characteristic of our species, but the function, or functions, that it serves has been debated for centuries. Here we bring recent evidence from neuroscience and allied disciplines to argue that in modern humans, language is a tool for communication, contrary to a prominent view that we use language for thinking. We begin by introducing the brain network that supports linguistic ability in humans. We then review evidence for a double dissociation between language and thought, and discuss several properties of language that suggest that it is optimized for communication. We conclude that although the emergence of language has unquestionably transformed human culture, language does not appear to be a prerequisite for complex thought, including symbolic thought. Instead, language is a powerful tool for the transmission of cultural knowledge; it plausibly co-evolved with our thinking and reasoning capacities, and only reflects, rather than gives rise to, the signature sophistication of human cognition.

The language network as a natural kind within the broader landscape of the human brain

Language behaviour is complex, but neuroscientific evidence disentangles it into distinct components supported by dedicated brain areas or networks. In this Review, we describe the 'core' language network, which includes left-hemisphere frontal and temporal areas, and show that it is strongly interconnected, independent of input and output modalities, causally important for language and language-selective. We discuss evidence that this language network plausibly stores language knowledge and supports core linguistic computations related to accessing words and constructions from memory and combining them to interpret (decode) or generate (encode) linguistic messages. We emphasize that the language network works closely with, but is distinct from, both lower-level - perceptual and motor - mechanisms and higher-level systems of knowledge and reasoning. The perceptual and motor mechanisms process linguistic signals, but, in contrast to the language network, are sensitive only to these signals' surface properties, not their meanings; the systems of knowledge and reasoning (such as the system that supports social reasoning) are sometimes engaged during language use but are not language-selective. This Review lays a foundation both for in-depth investigations of these different components of the language processing pipeline and for probing inter-component interactions.

Discriminating nonfluent/agrammatic and logopenic PPA variants with automatically extracted morphosyntactic measures from connected speech

Morphosyntactic assessments are important for characterizing individuals with nonfluent/agrammatic variant primary progressive aphasia (nfvPPA). Yet, standard tests are subject to examiner bias and often fail to differentiate between nfvPPA and logopenic variant PPA (lvPPA). Moreover, relevant neural signatures remain underexplored. Here, we leverage natural language processing tools to automatically capture morphosyntactic disturbances and their neuroanatomical correlates in 35 individuals with nfvPPA relative to 10 healthy controls (HC) and 26 individuals with lvPPA. Participants described a picture, and ensuing transcripts were analyzed via part-of-speech tagging to extract sentence-related features (e.g., subordinating and coordinating conjunctions), verbal-related features (e.g., tense markers), and nominal-related features (e.g., subjective and possessive pronouns). Gradient boosting machines were used to classify between groups using all features. We identified the most discriminant morphosyntactic marker via a feature importance algorithm and examined its neural correlates via voxel-based morphometry. Individuals with nfvPPA produced fewer morphosyntactic elements than the other two groups. Such features robustly discriminated them from both individuals with lvPPA and HCs with an AUC of .95 and .82, respectively. The most discriminatory feature corresponded to subordinating conjunctions was correlated with cortical atrophy within the left posterior inferior frontal gyrus across groups (pFWE < .05). Automated morphosyntactic analysis can efficiently differentiate nfvPPA from lvPPA. Also, the most sensitive morphosyntactic markers correlate with a core atrophy region of nfvPPA. Our approach, thus, can contribute to a key challenge in PPA diagnosis.

Diverging neural dynamics for syntactic structure building in naturalistic speaking and listening

The neural correlates of sentence production are typically studied using task paradigms that differ considerably from the experience of speaking outside of an experimental setting. In this fMRI study, we aimed to gain a better understanding of syntactic processing in spontaneous production versus naturalistic comprehension in three regions of interest (BA44, BA45, and left posterior middle temporal gyrus). A group of participants (n = 16) was asked to speak about the events of an episode of a TV series in the scanner. Another group of participants (n = 36) listened to the spoken recall of a participant from the first group. To model syntactic processing, we extracted word-by-word metrics of phrase-structure building with a top-down and a bottom-up parser that make different hypotheses about the timing of structure building. While the top-down parser anticipates syntactic structure, sometimes before it is obvious to the listener, the bottom-up parser builds syntactic structure in an integratory way after all of the evidence has been presented. In comprehension, neural activity was found to be better modeled by the bottom-up parser, while in production, it was better modeled by the top-down parser. We additionally modeled structure building in production with two strategies that were developed here to make different predictions about the incrementality of structure building during speaking. We found evidence for highly incremental and anticipatory structure building in production, which was confirmed by a converging analysis of the pausing patterns in speech. Overall, this study shows the feasibility of studying the neural dynamics of spontaneous language production.

Driving and suppressing the human language network using large language models

Transformer models such as GPT generate human-like language and are predictive of human brain responses to language. Here, using functional-MRI-measured brain responses to 1,000 diverse sentences, we first show that a GPT-based encoding model can predict the magnitude of the brain response associated with each sentence. We then use the model to identify new sentences that are predicted to drive or suppress responses in the human language network. We show that these model-selected novel sentences indeed strongly drive and suppress the activity of human language areas in new individuals. A systematic analysis of the model-selected sentences reveals that surprisal and well-formedness of linguistic input are key determinants of response strength in the language network. These results establish the ability of neural network models to not only mimic human language but also non-invasively control neural activity in higher-level cortical areas, such as the language network.

Common and distinct patterns underlying different linguistic tasks: multivariate disconnectome symptom mapping in poststroke patients

Numerous studies have been devoted to neural mechanisms of a variety of linguistic tasks (e.g. speech comprehension and production). To date, however, whether and how the neural patterns underlying different linguistic tasks are similar or differ remains elusive. In this study, we compared the neural patterns underlying 3 linguistic tasks mainly concerning speech comprehension and production. To address this, multivariate regression approaches with lesion/disconnection symptom mapping were applied to data from 216 stroke patients with damage to the left hemisphere. The results showed that lesion/disconnection patterns could predict both poststroke scores of speech comprehension and production tasks; these patterns exhibited shared regions on the temporal pole of the left hemisphere as well as unique regions contributing to the prediction for each domain. Lower scores in speech comprehension tasks were associated with lesions/abnormalities in the superior temporal gyrus and middle temporal gyrus, while lower scores in speech production tasks were associated with lesions/abnormalities in the left inferior parietal lobe and frontal lobe. These results suggested an important role of the ventral and dorsal stream pathways in speech comprehension and production (i.e. supporting the dual stream model) and highlighted the applicability of the novel multivariate disconnectome-based symptom mapping in cognitive neuroscience research.

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.

Domain-general and language-specific contributions to speech production in a second language: an fMRI study using functional localizers

For bilinguals, speaking in a second language (L2) compared to the native language (L1) is usually more difficult. In this study we asked whether the difficulty in L2 production reflects increased demands imposed on domain-general or core language mechanisms. We compared the brain response to speech production in L1 and L2 within two functionally-defined networks in the brain: the Multiple Demand (MD) network and the language network. We found that speech production in L2 was linked to a widespread increase of brain activity in the domain-general MD network. The language network did not show a similarly robust differences in processing speech in the two languages, however, we found increased response to L2 production in the language-specific portion of the left inferior frontal gyrus (IFG). To further explore our results, we have looked at domain-general and language-specific response within the brain structures postulated to form a Bilingual Language Control (BLC) network. Within this network, we found a robust increase in response to L2 in the domain-general, but also in some language-specific voxels including in the left IFG. Our findings show that L2 production strongly engages domain-general mechanisms, but only affects language sensitive portions of the left IFG. These results put constraints on the current model of bilingual language control by precisely disentangling the domain-general and language-specific contributions to the difficulty in speech production in L2.

Left-Hemisphere Cortical Language Regions Respond Equally to Observed Dialogue and Monologue

Much of the language we encounter in our everyday lives comes in the form of conversation, yet the majority of research on the neural basis of language comprehension has used input from only one speaker at a time. Twenty adults were scanned while passively observing audiovisual conversations using functional magnetic resonance imaging. In a block-design task, participants watched 20 s videos of puppets speaking either to another puppet (the dialogue condition) or directly to the viewer (the monologue condition), while the audio was either comprehensible (played forward) or incomprehensible (played backward). Individually functionally localized left-hemisphere language regions responded more to comprehensible than incomprehensible speech but did not respond differently to dialogue than monologue. In a second task, participants watched videos (1-3 min each) of two puppets conversing with each other, in which one puppet was comprehensible while the other's speech was reversed. All participants saw the same visual input but were randomly assigned which character's speech was comprehensible. In left-hemisphere cortical language regions, the time course of activity was correlated only among participants who heard the same character speaking comprehensibly, despite identical visual input across all participants. For comparison, some individually localized theory of mind regions and right-hemisphere homologues of language regions responded more to dialogue than monologue in the first task, and in the second task, activity in some regions was correlated across all participants regardless of which character was speaking comprehensibly. Together, these results suggest that canonical left-hemisphere cortical language regions are not sensitive to differences between observed dialogue and monologue.

Prediction during language comprehension: what is next?

Prediction is often regarded as an integral aspect of incremental language comprehension, but little is known about the cognitive architectures and mechanisms that support it. We review studies showing that listeners and readers use all manner of contextual information to generate multifaceted predictions about upcoming input. The nature of these predictions may vary between individuals owing to differences in language experience, among other factors. We then turn to unresolved questions which may guide the search for the underlying mechanisms. (i) Is prediction essential to language processing or an optional strategy? (ii) Are predictions generated from within the language system or by domain-general processes? (iii) What is the relationship between prediction and memory? (iv) Does prediction in comprehension require simulation via the production system? We discuss promising directions for making progress in answering these questions and for developing a mechanistic understanding of prediction in language.

Driving and suppressing the human language network using large language models

Transformer models such as GPT generate human-like language and are highly predictive of human brain responses to language. Here, using fMRI-measured brain responses to 1,000 diverse sentences, we first show that a GPT-based encoding model can predict the magnitude of brain response associated with each sentence. Then, we use the model to identify new sentences that are predicted to drive or suppress responses in the human language network. We show that these model-selected novel sentences indeed strongly drive and suppress activity of human language areas in new individuals. A systematic analysis of the model-selected sentences reveals that surprisal and well-formedness of linguistic input are key determinants of response strength in the language network. These results establish the ability of neural network models to not only mimic human language but also noninvasively control neural activity in higher-level cortical areas, like the language network.

Resting State Network Segregation Modulates Age-Related Differences in Language Production

Older adults typically exhibit decline in language production. However, how the brain supports or fails to support these processes is unclear. Moreover, there are competing hypotheses about the nature of age-related neural changes and whether age-related increases in neural activity reflect compensation or a decline in neural efficiency. In the current study, we investigated the neural bases of language production focusing on resting state functional connectivity. We hypothesized that language production performance, functional connectivity, and their relationship would differ as a function of age. Consistent with prior work, older age was associated with worse language production performance. Functional connectivity analyses showed that network segregation within the left hemisphere language network was maintained across adulthood. However, increased age was associated with lower whole brain network segregation. Moreover, network segregation was related to language production ability. In both network analyses, there were significant interactions with age-higher network segregation was associated with better language production abilities for younger and middle-aged adults, but not for older adults. Interestingly, there was a stronger relationship between language production and the whole brain network segregation than between production and the language network. These results highlight the utility of network segregation measures as an index of brain function, with higher network segregation associated with better language production ability. Moreover, these results are consistent with stability in the left hemisphere language network across adulthood and suggest that dedifferentiation among brain networks, outside of the language network, is a hallmark of aging and may contribute to age-related language production difficulties.

ROSE: A Neurocomputational Architecture for Syntax

A comprehensive model of natural language processing in the brain must accommodate four components: representations, operations, structures and encoding. It further requires a principled account of how these different components mechanistically, and causally, relate to each another. While previous models have isolated regions of interest for structure-building and lexical access, and have utilized specific neural recording measures to expose possible signatures of syntax, many gaps remain with respect to bridging distinct scales of analysis that map onto these four components. By expanding existing accounts of how neural oscillations can index various linguistic processes, this article proposes a neurocomputational architecture for syntax, termed the ROSE model (Representation, Operation, Structure, Encoding). Under ROSE, the basic data structures of syntax are atomic features, types of mental representations (R), and are coded at the single-unit and ensemble level. Elementary computations (O) that transform these units into manipulable objects accessible to subsequent structure-building levels are coded via high frequency broadband γ activity. Low frequency synchronization and cross-frequency coupling code for recursive categorial inferences (S). Distinct forms of low frequency coupling and phase-amplitude coupling (δ-θ coupling via pSTS-IFG; θ-γ coupling via IFG to conceptual hubs in lateral and ventral temporal cortex) then encode these structures onto distinct workspaces (E). Causally connecting R to O is spike-phase/LFP coupling; connecting O to S is phase-amplitude coupling; connecting S to E is a system of frontotemporal traveling oscillations; connecting E back to lower levels is low-frequency phase resetting of spike-LFP coupling. This compositional neural code has important implications for algorithmic accounts, since it makes concrete predictions for the appropriate level of study for psycholinguistic parsing models. ROSE is reliant on neurophysiologically plausible mechanisms, is supported at all four levels by a range of recent empirical research, and provides an anatomically precise and falsifiable grounding for the basic property of natural language syntax: hierarchical, recursive structure-building.