Statistical learning of speech regularities can occur outside the focus of attention

Visual statistical learning requires attention

Statistical learning is a person's ability to automatically learn environmental regularities through passive exposure. Since the earliest studies of statistical learning in infants, it has been debated exactly how "passive" this learning can be (i.e., whether attention is needed for learning to occur). In Experiment 1 of the current study, participants performed a serial feature search task where they searched for a target shape among heterogenous nontarget shapes. Unbeknownst to the participants, one of these nontarget shapes was presented much more often in location. Even though the regularity concerned a nonsalient, nontarget item that did not receive any attentional priority during search, participants still learned its regularity (responding faster when it was presented at this high-probability location). While this may suggest that not much, if any, attention is needed for learning to occur, follow-up experiments showed that if an attentional strategy (i.e., color subset search or exogenous cueing) effectively prevents attention from being directed to this critical regularity, incidental learning is no longer observed. We conclude that some degree of attention to a regularity is needed for visual statistical learning to occur.

Clinical Diagnostic and Prognostic Value of Residual Language Learning Ability in Patients with Disorders of Consciousness

Recent research suggests that the detection of preserved cognitive function can assist in the diagnosis and prognosis of patients with disorders of consciousness (DoC). This study investigates EEG signals as indicators of neural activity associated with the processing of transitional probabilities during a learning paradigm in patients with DoC. By examining the sensitivity to transitional probabilities across levels of consciousness, we aim to assess the potential value of this indicator in clinical diagnosis and prognosis. We collected EEG recordings from 51 DoC patients (10 female) and 26 healthy controls (9 female). EEG activity was recorded while participants listened to artificial vocabulary speech sequences before and after the learning phase. Intertrial phase coherence (ITPC) was used to examine differences in neural responses in different learning phases. Results showed that minimally conscious patients showed a significant increase in the word-tracking response after the learning phase, similar to healthy controls. Moreover, their learning-mediated word-rate ITPC difference correlated significantly with their Coma Recovery Scale-Revised score and 6 month outcome. However, these correlations were absent in unresponsive wakefulness syndrome patients. Crucially, differences in vocabulary ITPC before and after the learning phase effectively discriminated between healthy controls and patients, as well as between minimally conscious and unresponsive wakefulness syndrome patients. Combining EEG indicators with clinical performance accurately predicted patients' prognosis. In conclusion, the language learning paradigm has the potential to contribute to both diagnosis and prognosis in this challenging population, thereby significantly reducing prognostic uncertainty in medical decision-making and benefiting the rehabilitation of DoC patients.

Implicit prediction as a consequence of statistical learning

The sensory input that we encounter while navigating through each day is highly structured, containing patterns that repeat over time. Statistical learning is the process of becoming attuned to these patterns and can facilitate online processing. These online facilitation effects are often ascribed to prediction, in which information about an upcoming event is represented before it occurs. However, previously observed facilitation effects could also be due to retrospective processing. Here, using a speech-based segmentation paradigm, we tested whether statistical learning leads to the prediction of upcoming syllables. Specifically, we probed for a behavioural hallmark of genuine prediction, in which a given prediction benefits online processing when confirmed, but incurs costs if disconfirmed. In line with the idea that prediction is a key outcome of statistical learning, we found a trade-off in which a greater benefit for processing predictable syllables was associated with a greater cost in processing syllables that occurred in a "mismatch" context, outside of their expected positions. This trade-off in making predictions was evident at both the participant and the item (i.e., individual syllable) level. Further, we found that prediction did not emerge indiscriminately to all syllables in the input stream, but was deployed selectively according to the trial-by-trial demands of the task. Explicit knowledge of a given word was not required for prediction to occur, suggesting that prediction operates largely implicitly. Overall, these results provide novel behavioural evidence that prediction arises as a natural consequence of statistical learning.

Neural tracking of auditory statistical regularities in adults with and without dyslexia

Listeners implicitly use statistical regularities to segment continuous sound input into meaningful units, eg transitional probabilities between syllables to segment a speech stream into separate words. Implicit learning of such statistical regularities in a novel stimulus stream is reflected in a synchronization of neural responses to the sequential stimulus structure. The present study aimed to test the hypothesis that neural tracking of the statistical stimulus structure is reduced in individuals with dyslexia who have weaker reading and spelling skills, and possibly also weaker statistical learning abilities in general, compared to healthy controls. To this end, adults with and without dyslexia were presented with continuous streams of (non-speech) tones, which were arranged into triplets, such that transitional probabilities between single tones were higher within triplets and lower between triplets. We found that the so-called Triplet Learning Index (ie the ratio of neural phase coherence at the triplet rate relative to the tone rate) was lower in adults with dyslexia compared to the control group. Moreover, a higher Triplet Learning Index was associated with better spelling skills. These results suggest that individuals with dyslexia have a rather broad deficit in processing structure in sound instead of a merely phonological deficit.

Taking time: Auditory statistical learning benefits from distributed exposure

In an auditory statistical learning paradigm, listeners learn to partition a continuous stream of syllables by discovering the repeating syllable patterns that constitute the speech stream. Here, we ask whether auditory statistical learning benefits from spaced exposure compared with massed exposure. In a longitudinal online study on Prolific, we exposed 100 participants to the regularities in a spaced way (i.e., with exposure blocks spread out over 3 days) and another 100 in a massed way (i.e., with all exposure blocks lumped together on a single day). In the exposure phase, participants listened to streams composed of pairs while responding to a target syllable. The spaced and massed groups exhibited equal learning during exposure, as indicated by a comparable response-time advantage for predictable target syllables. However, in terms of resulting long-term knowledge, we observed a benefit from spaced exposure. Following a 2-week delay period, we tested participants' knowledge of the pairs in a forced-choice test. While both groups performed above chance, the spaced group had higher accuracy. Our findings speak to the importance of the timing of exposure to structured input and also for statistical learning outside of the laboratory (e.g., in language development), and imply that current investigations of auditory statistical learning likely underestimate human statistical learning abilities.

The role of conscious attention in auditory statistical learning: Evidence from patients with impaired consciousness

The need for attention to enable statistical learning is debated. Testing individuals with impaired consciousness offers valuable insight, but very few studies have been conducted due to the difficulties inherent in such studies. Here, we examined the ability of patients with varying levels of disorders of consciousness (DOC) to extract statistical regularities from an artificial language composed of randomly concatenated pseudowords by measuring frequency tagging in EEG. The objectives were firstly, to assess the automaticity of the segmentation process and the correlations between the level of covert consciousness and statistical learning capacities; secondly, to identify potential new diagnostic indicators. We observed that segmentation abilities were preserved in some minimally conscious patients, suggesting that auditory statistical learning is an inherently automatic low-level process. Due to significant inter-individual variability, word segmentation might not be robust enough for clinical use. In contrast, temporal accuracy of auditory syllable responses correlates strongly with coma severity.

Chunk Duration Limits the Learning of Multiword Chunks: Behavioral and Electroencephalography Evidence from Statistical Learning

Language comprehension involves the grouping of words into larger multiword chunks. This is required to recode information into sparser representations to mitigate memory limitations and counteract forgetting. It has been suggested that electrophysiological processing time windows constrain the formation of these units. Specifically, the period of rhythmic neural activity (i.e., low-frequency neural oscillations) may set an upper limit of 2-3 sec. Here, we assess whether learning of new multiword chunks is also affected by this neural limit. We applied an auditory statistical learning paradigm of an artificial language while manipulating the duration of to-be-learned chunks. Participants listened to isochronous sequences of disyllabic pseudowords from which they could learn hidden three-word chunks based on transitional probabilities. We presented chunks of 1.95, 2.55, and 3.15 sec that were created by varying the pause interval between pseudowords. In a first behavioral experiment, we tested learning using an implicit target detection task. We found better learning for chunks of 2.55 sec as compared to longer durations in line with an upper limit of the proposed time constraint. In a second experiment, we recorded participants' electroencephalogram during the exposure phase to use frequency tagging as a neural index of statistical learning. Extending the behavioral findings, results show a significant decline in neural tracking for chunks exceeding 3 sec as compared to both shorter durations. Overall, we suggest that language learning is constrained by endogenous time constraints, possibly reflecting electrophysiological processing windows.

Early Caregiver Predictability Shapes Neural Indices of Statistical Learning Later in Infancy

Caregivers play an outsized role in shaping early life experiences and development, but we often lack mechanistic insight into how exactly caregiver behavior scaffolds the neurodevelopment of specific learning processes. Here, we capitalized on the fact that caregivers differ in how predictable their behavior is to ask if infants' early environmental input shapes their brains' later ability to learn about predictable information. As part of an ongoing longitudinal study in South Africa, we recorded naturalistic, dyadic interactions between 103 (46 females and 57 males) infants and their primary caregivers at 3-6 months of age, from which we calculated the predictability of caregivers' behavior, following caregiver vocalization and overall. When the same infants were 6-12-months-old they participated in an auditory statistical learning task during EEG. We found evidence of learning-related change in infants' neural responses to predictable information during the statistical learning task. The magnitude of statistical learning-related change in infants' EEG responses was associated with the predictability of their caregiver's vocalizations several months earlier, such that infants with more predictable caregiver vocalization patterns showed more evidence of statistical learning later in the first year of life. These results suggest that early experiences with caregiver predictability influence learning, providing support for the hypothesis that the neurodevelopment of core learning and memory systems is closely tied to infants' experiences during key developmental windows.

Lexical Surprisal Shapes the Time Course of Syntactic Structure Building

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.

Visual linguistic statistical learning is traceable through neural entrainment

The human brain can detect statistical regularities in the environment across a wide variety of contexts. The importance of this process is well-established not just in language acquisition but across different modalities; in addition, several neural correlates of statistical learning have been identified. A current technique for tracking the emergence of regularity learning and localizing its neural background is frequency tagging (FT). FT can detect neural entrainment not only to the frequency of stimulus presentation but also to that of a hidden structure. Auditory learning paradigms with linguistic and nonlinguistic stimuli, along with a visual paradigm using nonlinguistic stimuli, have already been tested with FT. To complete the picture, we conducted an FT experiment using written syllables as stimuli and a hidden triplet structure. Both behavioral and neural entrainment data showed evidence of structure learning. In addition, we localized two electrode clusters related to the process, which spread across the frontal and parieto-occipital areas, similar to previous findings. Accordingly, we conclude that fast-paced visual linguistic regularities can be acquired and are traceable through neural entrainment. In comparison with the literature, our findings support the view that statistical learning involves a domain-general network.

The relationship between neural phase entrainment and statistical word-learning: A scoping review

Statistical language-learning, the capacity to extract regularities from a continuous speech stream, arguably involves the ability to segment the stream before the discrete constituents can be stored in memory. According to recent accounts, the segmentation process is reflected in the alignment of neural activity to the statistical structure embedded in the input. However, the degree to which it can predict the subsequent leaning outcome is currently unclear. As this is a relatively new avenue of research on statistical learning, a scoping review approach was adopted to identify and explore the current body of evidence on the use of neural phase entrainment as a measure of online neural statistical language-learning and its relation to the learning outcome, as well as the design characteristics of these studies. All included studies (11) observed entrainment to the underlying statistical pattern with exposure to the structured speech stream. A significant association between entrainment and learning outcome was observed in six of the studies. We discuss these findings in light of what neural entrainment in statistical word-learning experiments might represent, and speculate that it might reflect a general auditory processing mechanism, rather than segmentation of the speech stream per se. Lastly, as we find the current selection of studies to provide inconclusive evidence for neural entrainment's role in statistical learning, future research avenues are proposed.

The Brain's Sensitivity to Real-world Statistical Regularity Does Not Require Full Attention

Predictive coding accounts of perception state that the brain generates perceptual predictions in the service of processing incoming sensory data. These predictions are hypothesized to be afforded by the brain's ability to internalize useful patterns, that is, statistical regularities, from the environment. We have previously argued that the N300 ERP component serves as an index of the brain's use of representations of (real-world) statistical regularities. However, we do not yet know whether overt attention is necessary in order for this process to engage. We addressed this question by presenting stimuli of either high or low real-world statistical regularity in terms of their representativeness (good/bad exemplars of natural scene categories) to participants who either fully attended the stimuli or were distracted by another task (attended/distracted conditions). Replicating past work, N300 responses were larger to bad than to good scene exemplars, and furthermore, we demonstrate minimal impacts of distraction on N300 effects. Thus, it seems that overtly focused attention is not required to maintain the brain's sensitivity to real-world statistical regularity. Furthermore, in an exploratory analysis, we showed that providing additional, artificial regularities, formed by altering the proportions of good and bad exemplars within blocks, further enhanced the N300 effect in both attended and distracted conditions, shedding light on the relationship between statistical regularities learned in the real world and those learned within the context of an experiment.

Statistical learning at a virtual cocktail party

Statistical learning - the ability to extract distributional regularities from input - is suggested to be key to language acquisition. Yet, evidence for the human capacity for statistical learning comes mainly from studies conducted in carefully controlled settings without auditory distraction. While such conditions permit careful examination of learning, they do not reflect the naturalistic language learning experience, which is replete with auditory distraction - including competing talkers. Here, we examine how statistical language learning proceeds in a virtual cocktail party environment, where the to-be-learned input is presented alongside a competing speech stream with its own distributional regularities. During exposure, participants in the Dual Talker group concurrently heard two novel languages, one produced by a female talker and one by a male talker, with each talker virtually positioned at opposite sides of the listener (left/right) using binaural acoustic manipulations. Selective attention was manipulated by instructing participants to attend to only one of the two talkers. At test, participants were asked to distinguish words from part-words for both the attended and the unattended languages. Results indicated that participants' accuracy was significantly higher for trials from the attended vs. unattended language. Further, the performance of this Dual Talker group was no different compared to a control group who heard only one language from a single talker (Single Talker group). We thus conclude that statistical learning is modulated by selective attention, being relatively robust against the additional cognitive load provided by competing speech, emphasizing its efficiency in naturalistic language learning situations.

Limited but specific engagement of the mature language network during linguistic statistical learning

Statistical learning (SL) is the ability to detect and learn regularities from input and is foundational to language acquisition. Despite the dominant role of SL as a theoretical construct for language development, there is a lack of direct evidence supporting the shared neural substrates underlying language processing and SL. It is also not clear whether the similarities, if any, are related to linguistic processing, or statistical regularities in general. The current study tests whether the brain regions involved in natural language processing are similarly recruited during auditory, linguistic SL. Twenty-two adults performed an auditory linguistic SL task, an auditory nonlinguistic SL task, and a passive story listening task as their neural activation was monitored. Within the language network, the left posterior temporal gyrus showed sensitivity to embedded speech regularities during auditory, linguistic SL, but not auditory, nonlinguistic SL. Using a multivoxel pattern similarity analysis, we uncovered similarities between the neural representation of auditory, linguistic SL, and language processing within the left posterior temporal gyrus. No other brain regions showed similarities between linguistic SL and language comprehension, suggesting that a shared neurocomputational process for auditory SL and natural language processing within the left posterior temporal gyrus is specific to linguistic stimuli.

Does animacy affect visual statistical learning? Revisiting the effects of selective attention and animacy on visual statistical learning

Animates receive preferential attentional processing over inanimates because, from an evolutionary perspective, animates are important to human survival. We investigated whether animacy affects visual statistical learning-the detection and extraction of regularities in visual information from our rich, dynamic, and complex environment. Participants completed a selective-attention task, in which regularities were embedded in two visual streams, an attended and an unattended visual stream. The attended visual stream always consisted of line-drawings of non-objects, while the unattended visual stream consisted of line-drawings of either animates or inanimates. Participants then completed a triplet-discrimination task, which assessed their ability to extract regularities from the attended and unattended visual streams. We also assessed participants' awareness of regularities in the visual statistical learning task, and asked if any learning strategies were used. We were specifically interested in whether the animacy status of line-drawings in the unattended visual stream would affect visual statistical learning. There were four key findings. First, selective attention modulates visual statistical learning, with greater visual statistical learning for attended than for unattended information. Second, animacy does not affect visual statistical learning, with no differences found in visual statistical learning performance between the animate and inanimate condition. Third, awareness of regularities was associated with visual statistical learning of attended information. Fourth, participants used strategies (e.g., naming or labelling stimuli) during the visual statistical learning task. Further research is required to understand whether visual statistical learning is one of the adaptive functions that evolved from ancestral environments.

Of words and whistles: Statistical learning operates similarly for identical sounds perceived as speech and non-speech

Statistical learning is an ability that allows individuals to effortlessly extract patterns from the environment, such as sound patterns in speech. Some prior evidence suggests that statistical learning operates more robustly for speech compared to non-speech stimuli, supporting the idea that humans are predisposed to learn language. However, any apparent statistical learning advantage for speech could be driven by signal acoustics, rather than the subjective perception per se of sounds as speech. To resolve this issue, the current study assessed whether there is a statistical learning advantage for ambiguous sounds that are subjectively perceived as speech-like compared to the same sounds perceived as non-speech, thereby controlling for acoustic features. We first induced participants to perceive sine-wave speech (SWS)-a degraded form of speech not immediately perceptible as speech-as either speech or non-speech. After this induction phase, participants were exposed to a continuous stream of repeating trisyllabic nonsense words, composed of SWS syllables, and then completed an explicit familiarity rating task and an implicit target detection task to assess learning. Critically, participants showed robust and equivalent performance on both measures, regardless of their subjective speech perception. In contrast, participants who perceived the SWS syllables as more speech-like showed better detection of individual syllables embedded in speech streams. These results suggest that speech perception facilitates processing of individual sounds, but not the ability to extract patterns across sounds. Our findings suggest that statistical learning is not influenced by the perceived linguistic relevance of sounds, and that it may be conceptualized largely as an automatic, stimulus-driven mechanism.

Rhythmically Modulating Neural Entrainment during Exposure to Regularities Influences Statistical Learning

The ability to discover regularities in the environment, such as syllable patterns in speech, is known as statistical learning. Previous studies have shown that statistical learning is accompanied by neural entrainment, in which neural activity temporally aligns with repeating patterns over time. However, it is unclear whether these rhythmic neural dynamics play a functional role in statistical learning or whether they largely reflect the downstream consequences of learning, such as the enhanced perception of learned words in speech. To better understand this issue, we manipulated participants' neural entrainment during statistical learning using continuous rhythmic visual stimulation. Participants were exposed to a speech stream of repeating nonsense words while viewing either (1) a visual stimulus with a "congruent" rhythm that aligned with the word structure, (2) a visual stimulus with an incongruent rhythm, or (3) a static visual stimulus. Statistical learning was subsequently measured using both an explicit and implicit test. Participants in the congruent condition showed a significant increase in neural entrainment over auditory regions at the relevant word frequency, over and above effects of passive volume conduction, indicating that visual stimulation successfully altered neural entrainment within relevant neural substrates. Critically, during the subsequent implicit test, participants in the congruent condition showed an enhanced ability to predict upcoming syllables and stronger neural phase synchronization to component words, suggesting that they had gained greater sensitivity to the statistical structure of the speech stream relative to the incongruent and static groups. This learning benefit could not be attributed to strategic processes, as participants were largely unaware of the contingencies between the visual stimulation and embedded words. These results indicate that manipulating neural entrainment during exposure to regularities influences statistical learning outcomes, suggesting that neural entrainment may functionally contribute to statistical learning. Our findings encourage future studies using non-invasive brain stimulation methods to further understand the role of entrainment in statistical learning.

Separate but not independent: Behavioral pattern separation and statistical learning are differentially affected by aging

Our brains are capable of discriminating similar inputs (pattern separation) and rapidly generalizing across inputs (statistical learning). Are these two processes dissociable in behavior? Here, we asked whether cognitive aging affects them in a differential or parallel manner. Older and younger adults were tested on their ability to discriminate between similar trisyllabic words and to extract trisyllabic words embedded in a continuous speech stream. Older adults demonstrated intact statistical learning on an implicit, reaction time-based measure and an explicit, familiarity-based measure of learning. However, they performed poorly in discriminating similar items presented in isolation, both for episodically-encoded items and for statistically-learned regularities. These results indicate that pattern separation and statistical learning are dissociable and differentially affected by aging. The acquisition of implicit representations of statistical regularities operates robustly into old age, whereas pattern separation influences the expression of statistical learning with high representational fidelity and is subject to age-related decline.

Chunking Versus Transitional Probabilities: Differentiating Between Theories of Statistical Learning

There are two main approaches to how statistical patterns are extracted from sequences: The transitional probability approach proposes that statistical learning occurs through the computation of probabilities between items in a sequence. The chunking approach, including models such as PARSER and TRACX, proposes that units are extracted as chunks. Importantly, the chunking approach suggests that the extraction of full units weakens the processing of subunits while the transitional probability approach suggests that both units and subunits should strengthen. Previous findings using sequentially organized, auditory stimuli or spatially organized, visual stimuli support the chunking approach. However, one limitation of prior studies is that most assessed learning with the two-alternative forced-choice task. In contrast, this pre-registered experiment examined the two theoretical approaches in sequentially organized, visual stimuli using an online self-paced task-arguably providing a more sensitive index of learning as it occurs-and a secondary offline familiarity judgment task. During the self-paced task, abstract shapes were covertly organized into eight triplets (ABC) where one in every eight was altered (BCA) from the canonical structure in a way that disrupted the full unit while preserving a subunit (BC). Results from the offline familiarity judgment task revealed that the altered triplets were perceived as highly familiar, suggesting the learned representations were relatively flexible. More importantly, results from the online self-paced task demonstrated that processing for subunits, but not unit-initial stimuli, was impeded in the altered triplet. The pattern of results is in line with the chunking approach to statistical learning and, more specifically, the TRACX model.

No evidence for spatial suppression due to across-trial distractor learning in visual search

Previous studies have shown that during visual search, participants are able to implicitly learn across-trial regularities regarding target locations and use these to improve search performance. The present study asks whether such across-trial visual statistical learning also extends to the location of salient distractors. In Experiments 1 and 2, distractor regularities were paired so that a specific distractor location was 100% predictive of another specific distractor location on the next trial. Unlike previous findings that employed target regularities, the current results show no difference in search times between predictable and unpredictable trials. In Experiments 3-5 the distractor location was presented in a structured order (a sequence) for one group of participants, while it was presented randomly for the other group. Again, there was no learning effect of the across-trial regularities regarding the salient distractor locations. Across five experiments, we demonstrated that participants were unable to exploit across-trial spatial regularities regarding the salient distractors. These findings point to important boundary conditions for the modulation of visual attention by statistical regularities and they highlight the need to differentiate between different types of statistical regularities.

Foreign speech sound discrimination and associative word learning lead to a fast reconfiguration of resting-state networks

Learning new words in an unfamiliar language is a complex endeavor that requires the orchestration of multiple perceptual and cognitive functions. Although the neural mechanisms governing word learning are becoming better understood, little is known about the predictive value of resting-state (RS) metrics for foreign word discrimination and word learning attainment. In addition, it is still unknown which of the multistep processes involved in word learning have the potential to rapidly reconfigure RS networks. To address these research questions, we used electroencephalography (EEG), measured forty participants, and examined scalp-based power spectra, source-based spectral density maps and functional connectivity metrics before (RS1), in between (RS2) and after (RS3) a series of tasks which are known to facilitate the acquisition of new words in a foreign language, namely word discrimination, word-referent mapping and semantic generalization. Power spectra at the scalp level consistently revealed a reconfiguration of RS networks as a function of foreign word discrimination (RS1 vs. RS2) and word learning (RS1 vs. RS3) tasks in the delta, lower and upper alpha, and upper beta frequency ranges. Otherwise, functional reconfigurations at the source level were restricted to the theta (spectral density maps) and to the lower and upper alpha frequency bands (spectral density maps and functional connectivity). Notably, scalp RS changes related to the word discrimination tasks (difference between RS2 and RS1) correlated with word discrimination abilities (upper alpha band) and semantic generalization performance (theta and upper alpha bands), whereas functional changes related to the word learning tasks (difference between RS3 and RS1) correlated with word discrimination scores (lower alpha band). Taken together, these results highlight that foreign speech sound discrimination and word learning have the potential to rapidly reconfigure RS networks at multiple functional scales.

Statistical learning in patients in the minimally conscious state

When listening to speech, cortical activity can track mentally constructed linguistic units such as words, phrases, and sentences. Recent studies have also shown that the neural responses to mentally constructed linguistic units can predict the outcome of patients with disorders of consciousness (DoC). In healthy individuals, cortical tracking of linguistic units can be driven by both long-term linguistic knowledge and online learning of the transitional probability between syllables. Here, we investigated whether statistical learning could occur in patients in the minimally conscious state (MCS) and patients emerged from the MCS (EMCS) using electroencephalography (EEG). In Experiment 1, we presented to participants an isochronous sequence of syllables, which were composed of either 4 real disyllabic words or 4 reversed disyllabic words. An inter-trial phase coherence analysis revealed that the patient groups showed similar word tracking responses to real and reversed words. In Experiment 2, we presented trisyllabic artificial words that were defined by the transitional probability between words, and a significant word-rate EEG response was observed for MCS patients. These results suggested that statistical learning can occur with a minimal conscious level. The residual statistical learning ability in MCS patients could potentially be harnessed to induce neural plasticity.

Updating Contextual Sensory Expectations for Adaptive Behavior

The brain has the extraordinary capacity to construct predictive models of the environment by internalizing statistical regularities in the sensory inputs. The resulting sensory expectations shape how we perceive and react to the world; at the neural level, this relates to decreased neural responses to expected than unexpected stimuli ("expectation suppression"). Crucially, expectations may need revision as context changes. However, existing research has often neglected this issue. Further, it is unclear whether contextual revisions apply selectively to expectations relevant to the task at hand, hence serving adaptive behavior. The present fMRI study examined how contextual visual expectations spread throughout the cortical hierarchy as we update our beliefs. We created a volatile environment: two alternating contexts contained different sequences of object images, thereby producing context-dependent expectations that needed revision when the context changed. Human participants of both sexes attended a training session before scanning to learn the contextual sequences. The fMRI experiment then tested for the emergence of contextual expectation suppression in two separate tasks, respectively, with task-relevant and task-irrelevant expectations. Effects of contextual expectation emerged progressively across the cortical hierarchy as participants attuned themselves to the context: expectation suppression appeared first in the insula, inferior frontal gyrus, and posterior parietal cortex, followed by the ventral visual stream, up to early visual cortex. This applied selectively to task-relevant expectations. Together, the present results suggest that an insular and frontoparietal executive control network may guide the flexible deployment of contextual sensory expectations for adaptive behavior in our complex and dynamic world.SIGNIFICANCE STATEMENT The world is structured by statistical regularities, which we use to predict the future. This is often accompanied by suppressed neural responses to expected compared with unexpected events ("expectation suppression"). Crucially, the world is also highly volatile and context-dependent: expected events may become unexpected when the context changes, thus raising the crucial need for belief updating. However, this issue has generally been neglected. By setting up a volatile environment, we show that expectation suppression emerges first in executive control regions, followed by relevant sensory areas, only when observers use their expectations to optimize behavior. This provides surprising yet clear evidence on how the brain controls the updating of sensory expectations for adaptive behavior in our ever-changing world.

No statistical learning advantage in children over adults: Evidence from behaviour and neural entrainment

Explicit recognition measures of statistical learning (SL) suggest that children and adults have similar linguistic SL abilities. However, explicit tasks recruit additional cognitive processes that are not directly relevant for SL and may thus underestimate children's true SL capacities. In contrast, implicit tasks and neural measures of SL should be less influenced by explicit, higher-level cognitive abilities and thus may be better suited to capturing developmental differences in SL. Here, we assessed SL to six minutes of an artificial language in English-speaking children (n = 56, 24 females, M = 9.98 years) and adults (n = 44; 31 females, M = 22.97 years), using explicit and implicit behavioural measures and an EEG measure of neural entrainment. With few exceptions, children and adults showed largely similar performance on the behavioural explicit and implicit tasks, replicating prior work. Children and adults also demonstrated robust neural entrainment to both words and syllables, with a similar time course of word-level entrainment, reflecting learning of the hidden word structure. These results demonstrate that children and adults have similar linguistic SL abilities, even when learning is assessed through implicit performance-based and neural measures.

Statistical Learning of Language: A Meta-Analysis Into 25 Years of Research

Statistical learning is a key concept in our understanding of language acquisition. Ample work has highlighted its role in numerous linguistic functions-yet statistical learning is not a unitary construct, and its consistency across different language properties remains unclear. In a meta-analysis of auditory-linguistic statistical learning research spanning the last 25 years, we evaluated how learning varies across different language properties in infants, children, and adults and surveyed the methodological trends in the literature. We found robust learning across stimuli (syllables, words, etc.) in infants, and across stimuli and structures (adjacent dependencies, non-adjacent dependencies, etc.) in adults, with larger effect sizes when multiple cues were present. However, the analysis also showed significant publication bias and revealed a tendency toward using a narrow range of simplified language properties, including in the strength of the transitional probabilities used during training. Bayes factor analyses revealed prevalent data insensitivity of moderators commonly hypothesized to impact learning, such as the amount of exposure and transitional probability strength, which contradict core theoretical assumptions in the field. Methodological factors, such as the tasks used at test, also significantly impacted effect sizes in adults and children, suggesting that choice of task may critically constrain current theories of how statistical learning operates. Collectively, our results suggest that auditory-linguistic statistical learning has the kind of robustness needed to play a foundational role in language acquisition, but that more research is warranted to reveal its full potential.

Linking the neural basis of distributional statistical learning with transitional statistical learning: The paradox of attention

Statistical learning, the process of tracking distributional information and discovering embedded patterns, is traditionally regarded as a form of implicit learning. However, recent studies proposed that both implicit (attention-independent) and explicit (attention-dependent) learning systems are involved in statistical learning. To understand the role of attention in statistical learning, the current study investigates the cortical processing of distributional patterns in speech across local and global contexts. We then ask how these cortical responses relate to statistical learning behavior in a word segmentation task. We found Event-Related Potential (ERP) evidence of pre-attentive processing of both the local (mismatching negativity) and global distributional information (late discriminative negativity). However, as speech elements became less frequent and more surprising, some participants showed an involuntary attentional shift, reflected in a P3a response. Individuals who displayed attentive neural tracking of distributional information showed faster learning in a speech statistical learning task. These results suggest that an involuntary attentional shift might play a facilitatory, but not essential, role in statistical learning.

The power of rhythms: how steady-state evoked responses reveal early neurocognitive development

Electroencephalography (EEG) is a non-invasive and painless recording of cerebral activity, particularly well-suited for studying young infants, allowing the inspection of cerebral responses in a constellation of different ways. Of particular interest for developmental cognitive neuroscientists is the use of rhythmic stimulation, and the analysis of steady-state evoked potentials (SS-EPs) - an approach also known as frequency tagging. In this paper we rely on the existing SS-EP early developmental literature to illustrate the important advantages of SS-EPs for studying the developing brain. We argue that (1) the technique is both objective and predictive: the response is expected at the stimulation frequency (and/or higher harmonics), (2) its high spectral specificity makes the computed responses particularly robust to artifacts, and (3) the technique allows for short and efficient recordings, compatible with infants' limited attentional spans. We additionally provide an overview of some recent inspiring use of the SS-EP technique in adult research, in order to argue that (4) the SS-EP approach can be implemented creatively to target a wide range of cognitive and neural processes. For all these reasons, we expect SS-EPs to play an increasing role in the understanding of early cognitive processes. Finally, we provide practical guidelines for implementing and analyzing SS-EP studies.

Formant-invariant voice and pitch representations are pre-attentively formed from constantly varying speech and non-speech stimuli

The present study investigated whether listeners can form abstract voice representations while ignoring constantly changing phonological information and if they can use the resulting information to facilitate voice change detection. Further, the study aimed at understanding whether the use of abstraction is restricted to the speech domain or can be deployed also in non‐speech contexts. We ran an electroencephalogram (EEG) experiment including one passive and one active oddball task, each featuring a speech and a rotated speech condition. In the speech condition, participants heard constantly changing vowels uttered by a male speaker (standard stimuli) which were infrequently replaced by vowels uttered by a female speaker with higher pitch (deviant stimuli). In the rotated speech condition, participants heard rotated vowels, in which the natural formant structure of speech was disrupted. In the passive task, the mismatch negativity was elicited after the presentation of the deviant voice in both conditions, indicating that listeners could successfully group together different stimuli into a formant‐invariant voice representation. In the active task, participants showed shorter reaction times (RTs), higher accuracy and a larger P3b in the speech condition with respect to the rotated speech condition. Results showed that whereas at a pre‐attentive level the cognitive system can track pitch regularities while presumably ignoring constantly changing formant information both in speech and in rotated speech, at an attentive level the use of such information is facilitated for speech. This facilitation was also testified by a stronger synchronisation in the theta band (4–7 Hz), potentially pointing towards differences in encoding/retrieval processes.

Assessing the Sensitivity of EEG-Based Frequency-Tagging as a Metric for Statistical Learning

Statistical learning (SL) is hypothesized to play an important role in language development. However, the measures typically used to assess SL, particularly at the level of individual participants, are largely indirect and have low sensitivity. Recently, a neural metric based on frequency-tagging has been proposed as an alternative measure for studying SL. We tested the sensitivity of frequency-tagging measures for studying SL in individual participants in an artificial language paradigm, using non-invasive electroencephalograph (EEG) recordings of neural activity in humans. Importantly, we used carefully constructed controls to address potential acoustic confounds of the frequency-tagging approach, and compared the sensitivity of EEG-based metrics to both explicit and implicit behavioral tests of SL. Group-level results confirm that frequency-tagging can provide a robust indication of SL for an artificial language, above and beyond potential acoustic confounds. However, this metric had very low sensitivity at the level of individual participants, with significant effects found only in 30% of participants. Comparison of the neural metric to previously established behavioral measures for assessing SL showed a significant yet weak correspondence with performance on an implicit task, which was above-chance in 70% of participants, but no correspondence with the more common explicit 2-alternative forced-choice task, where performance did not exceed chance-level. Given the proposed ubiquitous nature of SL, our results highlight some of the operational and methodological challenges of obtaining robust metrics for assessing SL, as well as the potential confounds that should be taken into account when using the frequency-tagging approach in EEG studies.

Simple statistical regularities presented during sleep are detected but not retained

In recent years, there has been growing interest and excitement over the newly discovered cognitive capacities of the sleeping brain, including its ability to form novel associations. These recent discoveries raise the possibility that other more sophisticated forms of learning may also be possible during sleep. In the current study, we tested whether sleeping humans are capable of statistical learning - the process of becoming sensitive to repeating, hidden patterns in environmental input, such as embedded words in a continuous stream of speech. Participants' EEG was recorded while they were presented with one of two artificial languages, composed of either trisyllabic or disyllabic nonsense words, during slow-wave sleep. We used an EEG measure of neural entrainment to assess whether participants became sensitive to the repeating regularities during sleep-exposure to the language. We further probed for long-term memory representations by assessing participants' performance on implicit and explicit tests of statistical learning during subsequent wake. In the disyllabic-but not trisyllabic-language condition, participants' neural entrainment to words increased over time, reflecting a gradual gain in sensitivity to the embedded regularities. However, no significant behavioural effects of sleep-exposure were observed after the nap, for either language. Overall, our results indicate that the sleeping brain can detect simple, repeating pairs of syllables, but not more complex triplet regularities. However, the online detection of these regularities does not appear to produce any durable long-term memory traces that persist into wake - at least none that were revealed by our current measures and sample size. Although some perceptual aspects of statistical learning are preserved during sleep, the lack of memory benefits during wake indicates that exposure to a novel language during sleep may have limited practical value.

Dynamics of nonlinguistic statistical learning: From neural entrainment to the emergence of explicit knowledge

Humans are highly attuned to patterns in the environment. This ability to detect environmental patterns, referred to as statistical learning, plays a key role in many diverse aspects of cognition. However, the spatiotemporal neural mechanisms underlying implicit statistical learning, and how these mechanisms may relate or give rise to explicit learning, remain poorly understood. In the present study, we investigated these different aspects of statistical learning by using an auditory nonlinguistic statistical learning paradigm combined with magnetoencephalography. Twenty-four healthy volunteers were exposed to structured and random tone sequences, and statistical learning was quantified by neural entrainment. Already early during exposure, participants showed strong entrainment to the embedded tone patterns. A significant increase in entrainment over exposure was detected only in the structured condition, reflecting the trajectory of learning. While source reconstruction revealed a wide range of brain areas involved in this process, entrainment in areas around the left pre-central gyrus as well as right temporo-frontal areas significantly predicted behavioral performance. Sensor level results confirmed this relationship between neural entrainment and subsequent explicit knowledge. These results give insights into the dynamic relation between neural entrainment and explicit learning of triplet structures, suggesting that these two aspects are systematically related yet dissociable. Neural entrainment reflects robust, implicit learning of underlying patterns, whereas the emergence of explicit knowledge, likely built on the implicit encoding of structure, varies across individuals and may depend on factors such as sufficient exposure time and attention.

Optimizing steady-state responses to index statistical learning: Response to Benjamin and colleagues

Neural entrainment refers to the tendency of neural activity to align with an ongoing rhythmic stimulus. Measures of neural entrainment have been increasingly leveraged as a tool to understand how the brain tracks different types of regularities in sensory input. However, the methods used to quantify neural entrainment are varied, with numerous analytic decision points whose consequences have not been well-characterized. In a valuable contribution to this field, Benjamin, Dehaene-Lambertz and Flo (submitted) systematically compare various methodological approaches for studying neural entrainment. They demonstrate that the use of overlapping epochs, in which sliding time windows are extracted and analyzed, results in an artifactual inflation of entrainment estimates at the frequency of overlap. Here, in response to this updated best practice recommendation, we reanalyzed three previously published datasets that had been previously analyzed with overlapping epochs. Although our main results and conclusions are unaltered from those originally reported, we agree with Benjamin and colleagues that overlapping epochs should generally be avoided in classic analyses of steady-state experiments, which aim to quantify overall peaks in phase or power across an entire experimental duration. However, we present a case that overlapping epochs may be beneficial in fine-grained analyses of neural entrainment over time. The use of overlapping epochs in such analyses could improve temporal resolution without complicating interpretability of the results in cases where the question of interest relates to relative changes in neural entrainment over time within a given frequency.

Remarks on the analysis of steady-state responses: Spurious artifacts introduced by overlapping epochs

Periodic and stable sensory input can result in rhythmic and stable neural responses, a phenomenon commonly referred to as neural entrainment. Although the use of neural entrainment to investigate the regularities the brain tracks has increased in recent years, the methods used for its quantification are not well-defined in the literature. Here we argue that some strategies used in previous papers, are inadequate for the study of steady-state response, and lead to methodological artefacts. The aim of this commentary is to discuss these articles and to propose alternative measures of neural entrainment. Specifically, we applied four possible alternatives and two epoching approaches reported in the literature to quantify neural entrainment on simulated datasets. Our results demonstrate that overlapping epochs, as used in the original Batterink and colleagues articles, inevitably lead to a methodological artefact at the frequency corresponding to the overlap. We therefore strongly discourage this approach and encourage the re-analysis of data based on overlapping epochs. Additionally, we argue that the use of time-frequency decomposition to compute phase coherence at low frequencies to reveal neural entrainment is not optimal.

Neurophysiological tracking of speech-structure learning in typical and dyslexic readers

Statistical learning, or the ability to extract statistical regularities from the sensory environment, plays a critical role in language acquisition and reading development. Here we employed electroencephalography (EEG) with frequency-tagging measures to track the temporal evolution of speech-structure learning in individuals with reading difficulties due to developmental dyslexia and in typical readers. We measured EEG while participants listened to (a) a structured stream of repeated tri-syllabic pseudowords, (b) a random stream of the same isochronous syllables, and (c) a series of tri-syllabic real Dutch words. Participants' behavioral learning outcome (pseudoword recognition) was measured after training. We found that syllable-rate tracking was comparable between the two groups and stable across both the random and structured streams of syllables. More importantly, we observed a gradual emergence of the tracking of tri-syllabic pseudoword structures in both groups. Compared to the typical readers, however, in the dyslexic readers this implicit speech structure learning seemed to build up at a slower pace. A brain-behavioral correlation analysis showed that slower learners (i.e., participants who were slower in establishing the neural tracking of pseudowords) were less skilled in phonological awareness. Moreover, those who showed stronger neural tracking of real words tended to be less fluent in the visual-verbal conversion of linguistic symbols. Taken together, our study provides an online neurophysiological approach to track the progression of implicit learning processes and gives insights into the learning difficulties associated with dyslexia from a dynamic perspective.

Statistical learning and prosodic bootstrapping differentially affect neural synchronization during speech segmentation

Neural oscillations constitute an intrinsic property of functional brain organization that facilitates the tracking of linguistic units at multiple time scales through brain-to-stimulus alignment. This ubiquitous neural principle has been shown to facilitate speech segmentation and word learning based on statistical regularities. However, there is no common agreement yet on whether speech segmentation is mediated by a transition of neural synchronization from syllable to word rate, or whether the two time scales are concurrently tracked. Furthermore, it is currently unknown whether syllable transition probability contributes to speech segmentation when lexical stress cues can be directly used to extract word forms. Using Inter-Trial Coherence (ITC) analyses in combinations with Event-Related Potentials (ERPs), we showed that speech segmentation based on both statistical regularities and lexical stress cues was accompanied by concurrent neural synchronization to syllables and words. In particular, ITC at the word rate was generally higher in structured compared to random sequences, and this effect was particularly pronounced in the flat condition. Furthermore, ITC at the syllable rate dynamically increased across the blocks of the flat condition, whereas a similar modulation was not observed in the stressed condition. Notably, in the flat condition ITC at both time scales correlated with each other, and changes in neural synchronization were accompanied by a rapid reconfiguration of the P200 and N400 components with a close relationship between ITC and ERPs. These results highlight distinct computational principles governing neural synchronization to pertinent linguistic units while segmenting speech under different listening conditions.

Maternal stress predicts neural responses during auditory statistical learning in 26-month-old children: An event-related potential study

Exposure to high levels of early life stress have been associated with long-term difficulties in learning, behavior, and health, with particular impact evident in the language domain. While some have proposed that the increased stress of living in a low-income household mediates observed associations between socioeconomic status (SES) and child outcomes, considerable individual differences have been observed. The extent to which specific variables associated with socioeconomic status - in particular exposure to stressful life events - influence the neurocognitive mechanisms underlying language acquisition are not well understood. Auditory statistical learning, or the ability to segment a continuous auditory stream based on its statistical properties, develops during early infancy and is one mechanism thought to underlie language learning. The present study used an event-related potential (ERP) paradigm to test whether maternal stress, adjusting for socioeconomic variables (e.g., family income, maternal education) was associated with neurocognitive processes underlying statistical learning in a sample of 26-month-old children (n = 23) from predominantly low- to middle-income backgrounds. Event-related potentials were recorded while children listened to a continuous stream of tri-tone "words" in which tone elements varied in transitional probability. "Tone-words" were presented in random order, such that Tone 1 always predicted Tones 2 and 3 (transitional probability for Tone 3 = 1.0), but Tone 1 appeared randomly. A larger P2 amplitude was observed in response to Tone 3 compared to Tone 1, demonstrating that children implicitly tracked differences in transitional probabilities during passive listening. Maternal reports of stress at 26 months, adjusting for SES, were negatively associated with difference in P2 amplitude between Tones 1 and 3. These findings suggest that maternal stress, within a low-SES context, is associated with the manner in which children process statistical properties of auditory input.

Learning boosts the decoding of sound sequences in rat auditory cortex

Continuous acoustic streams, such as speech signals, can be chunked into segments containing reoccurring patterns (e.g., words). Noninvasive recordings of neural activity in humans suggest that chunking is underpinned by low-frequency cortical entrainment to the segment presentation rate, and modulated by prior segment experience (e.g., words belonging to a familiar language). Interestingly, previous studies suggest that also primates and rodents may be able to chunk acoustic streams. Here, we test whether neural activity in the rat auditory cortex is modulated by previous segment experience. We recorded subdural responses using electrocorticography (ECoG) from the auditory cortex of 11 anesthetized rats. Prior to recording, four rats were trained to detect familiar triplets of acoustic stimuli (artificial syllables), three were passively exposed to the triplets, while another four rats had no training experience. While low-frequency neural activity peaks were observed at the syllable level, no triplet-rate peaks were observed. Notably, in trained rats (but not in passively exposed and naïve rats), familiar triplets could be decoded more accurately than unfamiliar triplets based on neural activity in the auditory cortex. These results suggest that rats process acoustic sequences, and that their cortical activity is modulated by the training experience even under subsequent anesthesia.

•

Rats could behaviourally differentiate acoustic stimulus triplets after training.

•

Learning relatively increased auditory cortical entrainment to triplets.

•

Learning improved decoding of familiar stimuli based on auditory cortical activity.

Syllables in Sync Form a Link: Neural Phase-locking Reflects Word Knowledge during Language Learning

Language is composed of small building blocks, which combine to form larger meaningful structures. To understand language, we must process, track, and concatenate these building blocks into larger linguistic units as speech unfolds over time. An influential idea is that phase-locking of neural oscillations across different levels of linguistic structure provides a mechanism for this process. Building on this framework, the goal of the current study was to determine whether neural phase-locking occurs more robustly to novel linguistic items that are successfully learned and encoded into memory, compared to items that are not learned. Participants listened to a continuous speech stream composed of repeating nonsense words while their EEG was recorded and then performed a recognition test on the component words. Neural phase-locking to individual words during the learning period strongly predicted the strength of subsequent word knowledge, suggesting that neural phase-locking indexes the subjective perception of specific linguistic items during real-time language learning. These findings support neural oscillatory models of language, demonstrating that words that are successfully perceived as functional units are tracked by oscillatory activity at the matching word rate. In contrast, words that are not learned are processed merely as a sequence of unrelated syllables and thus not tracked by corresponding word-rate oscillations.

Preverbal Infants Discover Statistical Word Patterns at Similar Rates as Adults: Evidence From Neural Entrainment

The discovery of words in continuous speech is one of the first challenges faced by infants during language acquisition. This process is partially facilitated by statistical learning, the ability to discover and encode relevant patterns in the environment. Here, we used an electroencephalogram (EEG) index of neural entrainment to track 6-month-olds’ ( N = 25) segmentation of words from continuous speech. Infants’ neural entrainment to embedded words increased logarithmically over the learning period, consistent with a perceptual shift from isolated syllables to wordlike units. Moreover, infants’ neural entrainment during learning predicted postlearning behavioral measures of word discrimination ( n = 18). Finally, the logarithmic increase in entrainment to words was comparable in infants and adults, suggesting that infants and adults follow similar learning trajectories when tracking probability information among speech sounds. Statistical-learning effects in infants and adults may reflect overlapping neural mechanisms, which emerge early in life and are maintained throughout the life span.

Statistical learning in the absence of explicit top-down attention

Recently it has been shown that statistical learning of regularities presented in a display can bias attentional selection, such that attentional capture by salient objects is reduced by suppressing the location where these distractors are likely to appear. The role of attention in learning these contingencies is not immediately clear. Specifically, it is not known whether attention needs to be directed to the contingencies present in the display for learning to occur. In the current study we investigated whether participants can learn statistical regularities present in the environment even when these regularities are not relevant for the participant and are not part of their top-down goals. We used the additional singleton paradigm in which a color singleton was presented much more often in one location than in all other locations. We show that after being exposed to these regularities regarding the location of the color singleton during an unrelated task in which there are no targets nor distractors, participants showed a suppression effect from the previously learned contingencies when switching to a task in which they search for a target and suppress a distractor. We conclude that visual statistical learning can occur in the absence of top-down attention.

Tracking the Effects of Top-Down Attention on Word Discrimination Using Frequency-tagged Neuromagnetic Responses

Discrimination of words from nonspeech sounds is essential in communication. Still, how selective attention can influence this early step of speech processing remains elusive. To answer that question, brain activity was recorded with magnetoencephalography in 12 healthy adults while they listened to two sequences of auditory stimuli presented at 2.17 Hz, consisting of successions of one randomized word (tagging frequency = 0.54 Hz) and three acoustically matched nonverbal stimuli. Participants were instructed to focus their attention on the occurrence of a predefined word in the verbal attention condition and on a nonverbal stimulus in the nonverbal attention condition. Steady-state neuromagnetic responses were identified with spectral analysis at sensor and source levels. Significant sensor responses peaked at 0.54 and 2.17 Hz in both conditions. Sources at 0.54 Hz were reconstructed in supratemporal auditory cortex, left superior temporal gyrus (STG), left middle temporal gyrus, and left inferior frontal gyrus. Sources at 2.17 Hz were reconstructed in supratemporal auditory cortex and STG. Crucially, source strength in the left STG at 0.54 Hz was significantly higher in verbal attention than in nonverbal attention condition. This study demonstrates speech-sensitive responses at primary auditory and speech-related neocortical areas. Critically, it highlights that, during word discrimination, top-down attention modulates activity within the left STG. This area therefore appears to play a crucial role in selective verbal attentional processes for this early step of speech processing.