Toward a brain-based componential semantic representation

Area of Science:

• Cognitive Neuroscience
• Computational Linguistics
• Psycholinguistics

Background:

• Traditional componential theories of lexical semantics rely on features that are often complex, leaving the nature of primitive semantic components unclear.
• Neuroimaging advances highlight the embodied nature of conceptual representations, suggesting a link between neural systems and experienced concepts.
• Existing models struggle with issues like feature selection, abstract concept representation, and context effects.

Purpose of the Study:

• To develop a componential model of semantic representation grounded in functional divisions of the human brain.
• To propose a set of approximately 65 neurobiologically informed experiential attributes.
• To provide normative data and demonstrate the model's ability to capture semantic similarity and distinguish conceptual categories.

Main Methods:

• Proposed a set of ~65 experiential attributes based on sensory, motor, spatial, temporal, affective, social, and cognitive experiences.
• Collected normative data on attribute salience for English nouns, verbs, and adjectives.
• Utilized attribute vectors for category differentiation, semantic similarity analysis, and cluster analysis, comparing results with distributional (latent semantic) analysis.

Main Results:

• Attribute vectors effectively distinguished a priori conceptual categories and captured semantic similarity.
• Demonstrated robust quantitative differences between concrete object categories across numerous attribute dimensions.
• Cluster analyses revealed novel category distinctions independent of predefined labels, showing greater category separation than latent semantic analysis.

Conclusions:

• The proposed experiential attribute model provides a biologically plausible account of semantic representation, linking semantic content to large-scale brain networks.
• This approach offers solutions to longstanding semantic theory problems, including feature selection, abstract concept representation, and context effects.
• The model presents a viable alternative to traditional verbal-feature-based componential models, emphasizing embodied cognition.