Publications

Abstract

Human language offers a variety of ways to create meaning, one of which is referring to entities, objects, or events in the world. One such meaning maker is understanding to whom or to what a pronoun in a discourse refers to. To understand a pronoun, the brain must access matching entities or concepts that have been encoded in memory from previous linguistic context. Models of language processing propose that internally stored linguistic concepts, accessed via exogenous cues such as phonological input of a word, are represented as (a)synchronous activities across a population of neurons active at specific frequency bands. Converging evidence suggests that delta band activity (1–3 Hz) is involved in temporal and representational integration during sentence processing. Moreover, recent advances in the neurobiology of memory suggest that recollection engages neural dynamics similar to those which occurred during memory encoding. Integrating from these two research lines, we here tested the hypothesis that neural dynamic patterns, especially in delta frequency range, underlying referential meaning representation, would be reinstated during pronoun resolution. By leveraging neural decoding techniques (i.e., representational similarity analysis) on a magnetoencephalogram data set acquired during a naturalistic story-listening task, we provide evidence that delta-band activity underlies referential meaning representation. Our findings suggest that, during spoken language comprehension, endogenous linguistic representations such as referential concepts may be proactively retrieved and represented via activation of their underlying dynamic neural patterns.

Abstract

When we understand language, we recognize words and combine them into sentences. In this article, we explore the hypothesis that listeners use probabilistic information about words to build syntactic structure. Recent work has shown that lexical probability and syntactic structure both modulate the delta-band (<4 Hz) neural signal. Here, we investigated whether the neural encoding of syntactic structure changes as a function of the distributional properties of a word. To this end, we analyzed MEG data of 24 native speakers of Dutch who listened to three fairytales with a total duration of 49 min. Using temporal response functions and a cumulative model-comparison approach, we evaluated the contributions of syntactic and distributional features to the variance in the delta-band neural signal. This revealed that lexical surprisal values (a distributional feature), as well as bottom-up node counts (a syntactic feature) positively contributed to the model of the delta-band neural signal. Subsequently, we compared responses to the syntactic feature between words with high- and low-surprisal values. This revealed a delay in the response to the syntactic feature as a consequence of the surprisal value of the word: high-surprisal values were associated with a delayed response to the syntactic feature by 150–190 ms. The delay was not affected by word duration, and did not have a lexical origin. These findings suggest that the brain uses probabilistic information to infer syntactic structure, and highlight an importance for the role of time in this process.

Abstract

From a brain's-eye-view, when a stimulus occurs and what it is are interrelated aspects of interpreting the perceptual world. Yet in practice, the putative perceptual inferences about sensory content and timing are often dichotomized and not investigated as an integrated process. We here argue that neural temporal dynamics can influence what is perceived, and in turn, stimulus content can influence the time at which perception is achieved. This computational principle results from the highly interdependent relationship of what and when in the environment. Both brain processes and perceptual events display strong temporal variability that is not always modeled; we argue that understanding—and, minimally, modeling—this temporal variability is key for theories of how the brain generates unified and consistent neural representations and that we ignore temporal variability in our analysis practice at the peril of both data interpretation and theory-building. Here, we review what and when interactions in the brain, demonstrate via simulations how temporal variability can result in misguided interpretations and conclusions, and outline how to integrate and synthesize what and when in theories and models of brain computation.

Abstract

Neural oscillations reflect fluctuations in excitability, which biases the percept of ambiguous sensory input. Why this bias occurs is still not fully understood. We hypothesized that neural populations representing likely events are more sensitive, and thereby become active on earlier oscillatory phases, when the ensemble itself is less excitable. Perception of ambiguous input presented during less-excitable phases should therefore be biased toward frequent or predictable stimuli that have lower activation thresholds. Here, we show such a frequency bias in spoken word recognition using psychophysics, magnetoencephalography (MEG), and computational modelling. With MEG, we found a double dissociation, where the phase of oscillations in the superior temporal gyrus and medial temporal gyrus biased word-identification behavior based on phoneme and lexical frequencies, respectively. This finding was reproduced in a computational model. These results demonstrate that oscillations provide a temporal ordering of neural activity based on the sensitivity of separable neural populations.

Abstract

Humans excel at extracting structurally-determined meaning from speech despite inherent physical variability. This study explores the brain’s ability to predict and understand spoken language robustly. It investigates the relationship between structural and statistical language knowledge in brain dynamics, focusing on phase and amplitude modulation. Using syntactic features from constituent hierarchies and surface statistics from a transformer model as predictors of forward encoding models, we reconstructed cross-frequency neural dynamics from MEG data during audiobook listening. Our findings challenge a strict separation of linguistic structure and statistics in the brain, with both aiding neural signal reconstruction. Syntactic features have a more temporally spread impact, and both word entropy and the number of closing syntactic constituents are linked to the phase-amplitude coupling of neural dynamics, implying a role in temporal prediction and cortical oscillation alignment during speech processing. Our results indicate that structured and statistical information jointly shape neural dynamics during spoken language comprehension and suggest an integration process via a cross-frequency coupling mechanism

Abstract

Electrophysiological brain activity has been shown to synchronize with the quasi-regular repetition of grammatical phrases in connected speech—so-called phrase-rate neural tracking. Current debate centers around whether this phenomenon is best explained in terms of the syntactic properties of phrases or in terms of syntax-external information, such as the sequential repetition of parts of speech. As these two factors were confounded in previous studies, much of the literature is compatible with both accounts. Here, we used electroencephalography (EEG) to determine if and when the brain is sensitive to both types of information. Twenty native speakers of Mandarin Chinese listened to isochronously presented streams of monosyllabic words, which contained either grammatical two-word phrases (e.g., catch fish, sell house) or non-grammatical word combinations (e.g., full lend, bread far). Within the grammatical conditions, we varied two structural factors: the position of the head of each phrase and the type of attachment. Within the non-grammatical conditions, we varied the consistency with which parts of speech were repeated. Tracking was quantified through evoked power and inter-trial phase coherence, both derived from the frequency-domain representation of EEG responses. As expected, neural tracking at the phrase rate was stronger in grammatical sequences than in non-grammatical sequences without syntactic structure. Moreover, it was modulated by both attachment type and head position, revealing the structure-sensitivity of phrase-rate tracking. We additionally found that the brain tracks the repetition of parts of speech in non-grammatical sequences. These data provide an integrative perspective on the current debate about neural tracking effects, revealing that the brain utilizes regularities computed over multiple levels of linguistic representation in guiding rhythmic computation.

Abstract

Research into the role of brain oscillations in basic perceptual and cognitive functions has suggested that the alpha rhythm reflects functional inhibition while the beta rhythm reflects neural ensemble (re)activation. However, little is known regarding the generalization of these proposed fundamental operations to linguistic processes, such as speech comprehension and production. Here, we recorded magnetoencephalography in participants performing a novel rule-switching paradigm. Specifically, Dutch native speakers had to produce an alternative exemplar from the same category or a feature of a given target word embedded in spoken sentences (e.g., for the word “tuna”, an exemplar from the same category—“seafood”—would be “shrimp”, and a feature would be “pink”). A cue indicated the task rule—exemplar or feature—either before (pre-cue) or after (retro-cue) listening to the sentence. Alpha power during the working memory delay was lower for retro-cue compared with that for pre-cue in the left hemispheric language-related regions. Critically, alpha power negatively correlated with reaction times, suggestive of alpha facilitating task performance by regulating inhibition in regions linked to lexical retrieval. Furthermore, we observed a different spatiotemporal pattern of beta activity for exemplars versus features in the right temporoparietal regions, in line with the proposed role of beta in recruiting neural networks for the encoding of distinct categories. Overall, our study provides evidence for the generalizability of the role of alpha and beta oscillations from perceptual to more “complex, linguistic processes” and offers a novel task to investigate links between rule-switching, working memory, and word production.

Abstract

Since the cognitive revolution, language and action have been compared as cognitive systems, with cross-domain convergent views recently gaining renewed interest in biology, neuroscience, and cognitive science. Language and action are both combinatorial systems whose mode of combination has been argued to be hierarchical, combining elements into constituents of increasingly larger size. This structural similarity has led to the suggestion that they rely on shared cognitive and neural resources. In this article, we compare the conceptual and formal properties of hierarchy in language and action using set theory. We show that the strong compositionality of language requires a particular formalism, a magma, to describe the algebraic structure corresponding to the set of hierarchical structures underlying sentences. When this formalism is applied to actions, it appears to be both too strong and too weak. To overcome these limitations, which are related to the weak compositionality and sequential nature of action structures, we formalize the algebraic structure corresponding to the set of actions as a trace monoid. We aim to capture the different system properties of language and action in terms of the distinction between hierarchical sets and hierarchical sequences and discuss the implications for the way both systems could be represented in the brain.

Abstract

In the cognitive, computational, and neuro-sciences, practitioners often reason about what computational models represent or learn, as well as what algorithm is instantiated. The putative goal of such reasoning is to generalize claims about the model in question, to claims about the mind and brain, and the neurocognitive capacities of those systems. Such inference is often based on a model’s performance on a task, and whether that performance approximates human behavior or brain activity. Here we demonstrate how such argumentation problematizes the relationship between models and their targets; we place emphasis on artificial neural networks (ANNs), though any theory-brain relationship that falls into the same schema of reasoning is at risk. In this paper, we model inferences from ANNs to brains and back within a formal framework — metatheoretical calculus — in order to initiate a dialogue on both how models are broadly understood and used, and on how to best formally characterize them and their functions. To these ends, we express claims from the published record about models’ successes and failures in first-order logic. Our proposed formalization describes the decision-making processes enacted by scientists to adjudicate over theories. We demonstrate that formalizing the argumentation in the literature can uncover potential deep issues about how theory is related to phenomena. We discuss what this means broadly for research in cognitive science, neuroscience, and psychology; what it means for models when they lose the ability to mediate between theory and data in a meaningful way; and what this means for the metatheoretical calculus our fields deploy when performing high-level scientific inference.

Abstract

To understand language, we need to recognize words and combine them into phrases and sentences. During this process, responses to the words themselves are changed. In a step towards understanding how the brain builds sentence structure, the present study concerns the neural readout of this adaptation. We ask whether low-frequency neural readouts associated with words change as a function of being in a sentence. To this end, we analyzed an MEG dataset by Schoffelen et al. (2019) of 102 human participants (51 women) listening to sentences and word lists, the latter lacking any syntactic structure and combinatorial meaning. Using temporal response functions and a cumulative model-fitting approach, we disentangled delta- and theta-band responses to lexical information (word frequency), from responses to sensory- and distributional variables. The results suggest that delta-band responses to words are affected by sentence context in time and space, over and above entropy and surprisal. In both conditions, the word frequency response spanned left temporal and posterior frontal areas; however, the response appeared later in word lists than in sentences. In addition, sentence context determined whether inferior frontal areas were responsive to lexical information. In the theta band, the amplitude was larger in the word list condition around 100 milliseconds in right frontal areas. We conclude that low-frequency responses to words are changed by sentential context. The results of this study speak to how the neural representation of words is affected by structural context, and as such provide insight into how the brain instantiates compositionality in language.

Abstract

When we comprehend language from speech, the phase of the neural response aligns with particular features of the speech input, resulting in a phenomenon referred to as neural tracking. In recent years, a large body of work has demonstrated the tracking of the acoustic envelope and abstract linguistic units at the phoneme and word levels, and beyond. However, the degree to which speech tracking is driven by acoustic edges of the signal, or by internally-generated linguistic units, or by the interplay of both, remains contentious. In this study, we used naturalistic story-listening to investigate (1) whether phoneme-level features are tracked over and above acoustic edges, (2) whether word entropy, which can reflect sentence- and discourse-level constraints, impacted the encoding of acoustic and phoneme-level features, and (3) whether the tracking of acoustic edges was enhanced or suppressed during comprehension of a first language (Dutch) compared to a statistically familiar but uncomprehended language (French). We first show that encoding models with phoneme-level linguistic features, in addition to acoustic features, uncovered an increased neural tracking response; this signal was further amplified in a comprehended language, putatively reflecting the transformation of acoustic features into internally generated phoneme-level representations. Phonemes were tracked more strongly in a comprehended language, suggesting that language comprehension functions as a neural filter over acoustic edges of the speech signal as it transforms sensory signals into abstract linguistic units. We then show that word entropy enhances neural tracking of both acoustic and phonemic features when sentence- and discourse-context are less constraining. When language was not comprehended, acoustic features, but not phonemic ones, were more strongly modulated, but in contrast, when a native language is comprehended, phoneme features are more strongly modulated. Taken together, our findings highlight the flexible modulation of acoustic, and phonemic features by sentence and discourse-level constraint in language comprehension, and document the neural transformation from speech perception to language comprehension, consistent with an account of language processing as a neural filter from sensory to abstract representations.

Abstract

Brain oscillations are prevalent in all species and are involved in numerous perceptual operations. α oscillations are thought to facilitate processing through the inhibition of task-irrelevant networks, while β oscillations are linked to the putative reactivation of content representations. Can the proposed functional role of α and β oscillations be generalized from low-level operations to higher-level cognitive processes? Here we address this question focusing on naturalistic spoken language comprehension. Twenty-two (18 female) Dutch native speakers listened to stories in Dutch and French while MEG was recorded. We used dependency parsing to identify three dependency states at each word: the number of (1) newly opened dependencies, (2) dependencies that remained open, and (3) resolved dependencies. We then constructed forward models to predict α and β power from the dependency features. Results showed that dependency features predict α and β power in language-related regions beyond low-level linguistic features. Left temporal, fundamental language regions are involved in language comprehension in α, while frontal and parietal, higher-order language regions, and motor regions are involved in β. Critically, α- and β-band dynamics seem to subserve language comprehension tapping into syntactic structure building and semantic composition by providing low-level mechanistic operations for inhibition and reactivation processes. Because of the temporal similarity of the α-β responses, their potential functional dissociation remains to be elucidated. Overall, this study sheds light on the role of α and β oscillations during naturalistic spoken language comprehension, providing evidence for the generalizability of these dynamics from perceptual to complex linguistic processes.

Abstract

In this study, we compared the performance of a long short-term memory (LSTM) neural network to the behavior of human participants on a language task that requires hierarchically structured knowledge. We show that humans interpret ambiguous noun phrases, such as second blue ball, in line with their hierarchical constituent structure. LSTMs, instead, only do
so after unambiguous training, and they do not systematically generalize to novel items. Overall, the results of our simulations indicate that a model can behave hierarchically without relying on hierarchical constituent structure.

Abstract

How a system represents information tightly constrains the kinds of problems it can solve. Humans routinely solve problems that appear to require abstract representations of stimulus properties and relations. How we acquire such representations has central importance in an account of human cognition. We briefly describe a theory of how a system can learn invariant responses to instances of similarity and relative magnitude, and how structured, relational representations can be learned from initially unstructured inputs. Two operations, comparing distributed representations and learning from the concomitant network dynamics in time, underpin the ability to learn these representations and to respond to invariance in the environment. Comparing analog representations of absolute magnitude produces invariant signals that carry information about similarity and relative magnitude. We describe how a system can then use this information to bootstrap learning structured (i.e., symbolic) concepts of relative magnitude from experience without assuming such representations a priori.

Abstract

Psychology endeavors to develop theories of human capacities and behaviors on the basis of a variety of methodologies and dependent measures. We argue that one of the most divisive factors in psychological science is whether researchers choose to use computational modeling of theories (over and above data) during the scientific-inference process. Modeling is undervalued yet holds promise for advancing psychological science. The inherent demands of computational modeling guide us toward better science by forcing us to conceptually analyze, specify, and formalize intuitions that otherwise remain unexamined—what we dub open theory. Constraining our inference process through modeling enables us to build explanatory and predictive theories. Here, we present scientific inference in psychology as a path function in which each step shapes the next. Computational modeling can constrain these steps, thus advancing scientific inference over and above the stewardship of experimental practice (e.g., preregistration). If psychology continues to eschew computational modeling, we predict more replicability crises and persistent failure at coherent theory building. This is because without formal modeling we lack open and transparent theorizing. We also explain how to formalize, specify, and implement a computational model, emphasizing that the advantages of modeling can be achieved by anyone with benefit to all.

Abstract

Whether neural networks can capture relational knowledge is a matter of long-standing controversy. Recently, some researchers have argued that (1) classic connectionist models can handle relational structure and (2) the success of deep learning approaches to natural language processing suggests that structured representations are unnecessary to model human language. We tested the Story Gestalt model, a classic connectionist model of text comprehension, and a Sequence-to-Sequence with Attention model, a modern deep learning architecture for natural language processing. Both models were trained to answer questions about stories based on abstract thematic roles. Two simulations varied the statistical structure of new stories while keeping their relational structure intact. The performance of each model fell below chance at least under one manipulation. We argue that both models fail our tests because they can't perform dynamic binding. These results cast doubts on the suitability of traditional neural networks for explaining relational reasoning and language processing phenomena.

Abstract

Neuronal oscillations putatively track speech in order to optimize sensory processing. However, it is unclear how isochronous brain oscillations can track pseudo-rhythmic speech input. Here we propose that oscillations can track pseudo-rhythmic speech when considering that speech time is dependent on content-based predictions flowing from internal language models. We show that temporal dynamics of speech are dependent on the predictability of words in a sentence. A computational model including oscillations, feedback, and inhibition is able to track pseudo-rhythmic speech input. As the model processes, it generates temporal phase codes, which are a candidate mechanism for carrying information forward in time. The model is optimally sensitive to the natural temporal speech dynamics and can explain empirical data on temporal speech illusions. Our results suggest that speech tracking does not have to rely only on the acoustics but could also exploit ongoing interactions between oscillations and constraints flowing from internal language models.