Faculty Bibliography

Departing from traditional linguistic models, advances in deep learning have resulted in a new type of predictive (autoregressive) deep language models (DLMs). Using a self-supervised next-word prediction task, these models generate appropriate linguistic responses in a given context. In the current study, nine participants listened to a 30-min podcast while their brain responses were recorded using electrocorticography (ECoG). We provide empirical evidence that the human brain and autoregressive DLMs share three fundamental computational principles as they process the same natural narrative: (1) both are engaged in continuous next-word prediction before word onset; (2) both match their pre-onset predictions to the incoming word to calculate post-onset surprise; (3) both rely on contextual embeddings to represent words in natural contexts. Together, our findings suggest that autoregressive DLMs provide a new and biologically feasible computational framework for studying the neural basis of language.

Hearing one's own voice is critical for fluent speech production as it allows for the detection and correction of vocalization errors in real time. This behavior known as the auditory feedback control of speech is impaired in various neurological disorders ranging from stuttering to aphasia; however, the underlying neural mechanisms are still poorly understood. Computational models of speech motor control suggest that, during speech production, the brain uses an efference copy of the motor command to generate an internal estimate of the speech output. When actual feedback differs from this internal estimate, an error signal is generated to correct the internal estimate and update necessary motor commands to produce intended speech. We were able to localize the auditory error signal using electrocorticographic recordings from neurosurgical participants during a delayed auditory feedback (DAF) paradigm. In this task, participants hear their voice with a time delay as they produced words and sentences (similar to an echo on a conference call), which is well known to disrupt fluency by causing slow and stutter-like speech in humans. We observed a significant response enhancement in auditory cortex that scaled with the duration of feedback delay, indicating an auditory speech error signal. Immediately following auditory cortex, dorsal precentral gyrus (dPreCG), a region that has not been implicated in auditory feedback processing before, exhibited a markedly similar response enhancement, suggesting a tight coupling between the 2 regions. Critically, response enhancement in dPreCG occurred only during articulation of long utterances due to a continuous mismatch between produced speech and reafferent feedback. These results suggest that dPreCG plays an essential role in processing auditory error signals during speech production to maintain fluency.

Perception results from the interplay of sensory input and prior knowledge. Despite behavioral evidence that long-term priors powerfully shape perception, the neural mechanisms underlying these interactions remain poorly understood. We obtained direct cortical recordings in neurosurgical patients as they viewed ambiguous images that elicit constant perceptual switching. We observe top-down influences from the temporal to occipital cortex, during the preferred percept that is congruent with the long-term prior. By contrast, stronger feedforward drive is observed during the non-preferred percept, consistent with a prediction error signal. A computational model based on hierarchical predictive coding and attractor networks reproduces all key experimental findings. These results suggest a pattern of large-scale information flow change underlying long-term priors' influence on perception and provide constraints on theories about long-term priors' influence on perception.

Humans form lasting memories of stimuli that were only encountered once. This naturally occurs when listening to a story, however it remains unclear how and when memories are stored and retrieved during story-listening. Here, we first confirm in behavioral experiments that participants can learn about the structure of a story after a single exposure and are able to recall upcoming words when the story is presented again. We then track mnemonic information in high frequency activity (70-200 Hz) as patients undergoing electrocorticographic recordings listen twice to the same story. We demonstrate predictive recall of upcoming information through neural responses in auditory processing regions. This neural measure correlates with behavioral measures of event segmentation and learning. Event boundaries are linked to information flow from cortex to hippocampus. When listening for a second time, information flow from hippocampus to cortex precedes moments of predictive recall. These results provide insight on a fine-grained temporal scale into how episodic memory encoding and retrieval work under naturalistic conditions.

Despite increased awareness of the lack of gender equity in academia and a growing number of initiatives to address issues of diversity, change is slow, and inequalities remain. A major source of inequity is gender bias, which has a substantial negative impact on the careers, work-life balance, and mental health of underrepresented groups in science. Here, we argue that gender bias is not a single problem but manifests as a collection of distinct issues that impact researchers' lives. We disentangle these facets and propose concrete solutions that can be adopted by individuals, academic institutions, and society.

We describe the spatiotemporal course of cortical high-gamma activity, hippocampal ripple activity and interictal epileptiform discharges during an associative memory task in 15 epilepsy patients undergoing invasive EEG. Successful encoding trials manifested significantly greater high-gamma activity in hippocampus and frontal regions. Successful cued recall trials manifested sustained high-gamma activity in hippocampus compared to failed responses. Hippocampal ripple rates were greater during successful encoding and retrieval trials. Interictal epileptiform discharges during encoding were associated with 15% decreased odds of remembering in hippocampus (95% confidence interval 6-23%). Hippocampal interictal epileptiform discharges during retrieval predicted 25% decreased odds of remembering (15-33%). Odds of remembering were reduced by 25-52% if interictal epileptiform discharges occurred during the 500-2000-ms window of encoding or by 41% during retrieval. During encoding and retrieval, hippocampal interictal epileptiform discharges were followed by a transient decrease in ripple rate. We hypothesize that interictal epileptiform discharges impair associative memory in a regionally and temporally specific manner by decreasing physiological hippocampal ripples necessary for effective encoding and recall. Because dynamic memory impairment arises from pathological interictal epileptiform discharge events competing with physiological ripples, interictal epileptiform discharges represent a promising therapeutic target for memory remediation in patients with epilepsy.

Sensory input arrives in continuous sequences that humans experience as segmented units, e.g., words and events. The brain's ability to discover regularities is called statistical learning. Structure can be represented at multiple levels, including transitional probabilities, ordinal position, and identity of units. To investigate sequence encoding in cortex and hippocampus, we recorded from intracranial electrodes in human subjects as they were exposed to auditory and visual sequences containing temporal regularities. We find neural tracking of regularities within minutes, with characteristic profiles across brain areas. Early processing tracked lower-level features (e.g., syllables) and learned units (e.g., words), while later processing tracked only learned units. Learning rapidly shaped neural representations, with a gradient of complexity from early brain areas encoding transitional probability, to associative regions and hippocampus encoding ordinal position and identity of units. These findings indicate the existence of multiple, parallel computational systems for sequence learning across hierarchically organized cortico-hippocampal circuits.

Functional human brain mapping is commonly performed during invasive monitoring with intracranial electroencephalographic (iEEG) electrodes prior to resective surgery for drug- resistant epilepsy. The current gold standard, electrocortical stimulation mapping (ESM), is time -consuming, sometimes elicits pain, and often induces after discharges or seizures. Moreover, there is a risk of overestimating eloquent areas due to propagation of the effects of stimulation to a broader network of language cortex. Passive iEEG spatial-temporal functional mapping (STFM) has recently emerged as a potential alternative to ESM. However, investigators have observed less correspondence between STFM and ESM maps of language than between their maps of motor function. We hypothesized that incongruities between ESM and STFM of language function may arise due to propagation of the effects of ESM to cortical areas having strong effective connectivity with the site of stimulation. We evaluated five patients who underwent invasive monitoring for seizure localization, whose language areas were identified using ESM. All patients performed a battery of language tasks during passive iEEG recordings. To estimate the effective connectivity of stimulation sites with a broader network of task-activated cortical sites, we measured cortico-cortical evoked potentials (CCEPs) elicited across all recording sites by single-pulse electrical stimulation at sites where ESM was performed at other times. With the combination of high gamma power as well as CCEPs results, we trained a logistic regression model to predict ESM results at individual electrode pairs. The average accuracy of the classifier using both STFM and CCEPs results combined was 87.7%, significantly higher than the one using STFM alone (71.8%), indicating that the correspondence between STFM and ESM results is greater when effective connectivity between ESM stimulation sites and task-activated sites is taken into consideration. These findings, though based on a small number of subjects to date, provide preliminary support for the hypothesis that incongruities between ESM and STFM may arise in part from propagation of stimulation effects to a broader network of cortical language sites activated by language tasks, and suggest that more studies, with larger numbers of patients, are needed to understand the utility of both mapping techniques in clinical practice.