Alexander G. Huth

Visual Semantic Encoding and Identification of Naturalistic Movies via High-Density Diffuse Optical Tomography

This study uses HD-DOT to encode and decode semantic representations from naturalistic movies, showing it can capture high-dimensional semantic maps similar to fMRI.

The Premotor Language Area Encodes a Full Acoustic-to-semantic Speech Hierarchy

This study identifies a premotor language area (PML) in the ventral premotor cortex that contains a full speech-processing hierarchy from acoustic to semantic features.

The human language processing system straightens natural speech

The paper investigates how temporal prediction shapes neural speech representations, finding that brain responses to natural speech become progressively straighter along the cortical hierarchy.

Opposing BOLD signals and oxygen metabolism largely arise from statistical uncertainty in metabolic estimates

The paper reanalyzes BOLD and metabolic data, finding that apparent discordance between BOLD signals and metabolic estimates is due to statistical uncertainty rather than physiological reversal.

Decoding concept representations in aphasia after stroke

This study decodes concept representations in aphasia using fMRI to generate continuous word sequences, showing that concept representations are largely preserved despite aphasia.

Rapid cortical mapping with cross-participant encoding models

This study introduces cross-participant encoding models for rapid cortical mapping, showing improved accuracy and efficiency compared to within-participant models.

Evaluating scientific theories as predictive models in language neuroscience

This paper introduces QA encoding models that convert qualitative language theories into interpretable brain response predictors, outperforming baselines in fMRI and ECoG data.

Efficient uniform sampling explains non-uniform memory of narrative stories

The paper proposes a uniform sampling model to explain non-uniform memory of narrative stories, showing it outperforms surprise-based theories.

A natural language fMRI dataset for voxelwise encoding models

The paper provides a natural language fMRI dataset with BOLD responses and tools for voxelwise encoding models to study language comprehension.

A unifying computational account of temporal context effects in language across the human cortex

This paper introduces a framework to improve deep learning model interpretability in language processing by simulating brain responses and identifying temporal context effects in human cortex.

Semantic reconstruction of continuous language from non-invasive brain recordings

This paper introduces a non-invasive brain-computer interface that decodes continuous natural language from fMRI data, demonstrating it can reconstruct speech and silent videos without requiring subject cooperation.

Voxelwise encoding models show that cerebellar language representations are highly conceptual

The paper uses voxelwise encoding models to show cerebellar language responses are primarily conceptual and socially biased compared to cortex.

Cortical Representations of Concrete and Abstract Concepts in Language Combine Visual and Linguistic Representations

The paper shows that cortical regions near visual systems combine visual and linguistic information for concepts, while language regions use mostly linguistic information, with concrete concepts containing more visual information.

Attentional Modulation of Hierarchical Speech Representations in a Multitalker Environment

The paper investigates how attention modulates speech representations in the brain during multitalker listening tasks, finding attentional effects across multiple processing stages.

Interpretable multi-timescale models for predicting fMRI responses to continuous natural speech

This paper develops interpretable multi-timescale models to predict fMRI responses to natural speech, revealing how different brain regions encode information at varying temporal scales.

Incorporating Context into Language Encoding Models for fMRI

Encoding models using contextual representations from an LSTM language model improve human fMRI brain response prediction over non-contextual word embeddings in nearly every brain area and map cortical context sensitivity.

Voxelwise encoding models with non-spherical multivariate normal priors

This paper introduces Tikhonov regression for fMRI data modeling, showing it can improve predictive models by using non-spherical Gaussian priors instead of spherical ridge regression.