Author Correction: An fMRI Dataset for Concept Representation with Semantic Feature Annotations

Shaonan Wang, Yunhao Zhang, Xiaohan Zhang, Jingyuan Sun, Nan Lin, Jiajun Zhang, Chengqing ZongScientific 2023 · DOI · OpenAlex

Introduction: Mapping the Landscape of Human Thought

How does the human brain transform sensory input or linguistic symbols into meaningful concepts? While cognitive neuroscience has made significant progress in identifying the neural correlates of concrete objects—such as "apple" or "hammer"—the representation of abstract concepts like "justice," "faith," or "velocity" remains a frontier. Most existing neuroimaging datasets are constrained by small stimulus sets or a narrow focus on physical entities, creating a "generalizability gap." To address this, researchers have released a comprehensive fMRI dataset capturing brain activity for 672 distinct concepts, spanning both the tangible and the intangible.

Bridging the Concrete-Abstract Divide

The primary challenge in studying concept representation is the sheer diversity of human thought. Previous studies often relied on a few dozen nouns, which may not reflect the complex web of semantic associations used in daily life. This new dataset provides a high-resolution benchmark by pairing neural imaging with 54 crowdsourced semantic feature annotations. By doing so, it offers a multi-dimensional view of how the brain encodes the "building blocks" of meaning across a wide range of semantic categories.

Methodology: A Multi-Modal Approach

The researchers employed a robust experimental design to ensure the data captured conceptual meaning rather than just perceptual features:

  • Stimulus Design: Participants engaged with 672 individual concepts. To isolate the underlying concept from specific modalities, each was probed using a combination of words and images.
  • Neural Recording: High-resolution fMRI data was collected from 11 participants as they processed these stimuli.
  • Semantic Feature Crowdsourcing: To decode the neural patterns, the team collected annotations for 54 semantic features per concept. These features are rooted in componential theories of representation, which posit that concepts are constructed from various experiential dimensions, including sensory/motor (e.g., color, motion), spatial/temporal (scale and duration), affective/social (emotional resonance), and cognitive (intellectual complexity).

Dataset Overview

The resulting dataset represents a significant leap in scale and depth for the field:

  • Scope: 672 concepts, a substantial increase over traditional datasets.
  • Data Structure: fMRI scans from 11 subjects paired with 54-dimensional feature vectors for every concept.
  • Diversity: A balanced inclusion of concrete and abstract concepts across diverse semantic domains.

Key Insights: The "Building Block" Framework

The dataset enables a "bottom-up" analysis of human cognition. Rather than searching for a single localized "node" for a word like "freedom," researchers can investigate how brain regions track specific features—such as "social" or "temporal" components—across the entire set of 672 concepts.

This approach is particularly valuable for determining whether abstract concepts recruit the same neural "hardware" as concrete ones or if they rely on distinct systems involved in language and social reasoning. Technical validation confirms that the data is of sufficient quality to investigate the neural representation of individual concepts, rather than just broad categories.

Future Directions and Applications

This dataset serves as a bridge between traditional neuroscience and modern computational linguistics. Potential applications include:

  1. Comparative AI Research: Comparing human brain representations with the internal "embeddings" of Large Language Models (LLMs).
  2. Brain-Decoding Algorithms: Developing models that can predict conceptual thought based on neural activity patterns.
  3. Clinical Insights: Exploring how neurological or psychiatric conditions might alter the brain's "semantic map."

While the sample size of 11 participants is standard for intensive fMRI studies, and crowdsourcing introduces some subjectivity, the scale and feature richness of this dataset provide a powerful new tool for understanding the architecture of the human mind.

Summary updated

Abstract

The neural representation of concepts is a focus of many cognitive neuroscience studies. Prior works studying concept representation with neural imaging data have been largely limited to concrete concepts. The use of relatively small and constrained sets of stimuli leaves open the question of whether the findings can generalize other concepts. We share an fMRI dataset in which 11 participants thought of 672 individual concepts, including both concrete and abstract concepts. The concepts were probed using words paired with images in which the words were selected to cover a wide range of semantic categories. Furthermore, according to the componential theories of concept representation, we collected the 54 semantic features of the 672 concepts comprising sensory, motor, spatial, temporal, affective, social, and cognitive experiences by crowdsourcing annotations. The quality assessment results verify this as a high-quality neuroimaging dataset. Such a dataset is well suited to study how the brain represents different semantic features and concepts, creating the essential condition to investigate the neural representation of individual concepts.
BibTeX
@inproceedings{wang2023author,
  author = {Shaonan Wang and Yunhao Zhang and Xiaohan Zhang and Jingyuan Sun and Nan Lin and Jiajun Zhang and Chengqing Zong},
  title = {{Author Correction: An fMRI Dataset for Concept Representation with Semantic Feature Annotations}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41597-023-02480-w},
}

Loading related content...

Figure