Perceptual inference through global lexical similarity.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Psychology

Background:

• Existing theories on lexical semantic representation learning are disconnected, proposing either perceptual experience or language regularities as the primary source.
• A gap exists in understanding how humans integrate multimodal information for word meaning acquisition.

Purpose of the Study:

• To propose and validate a computational model that integrates both perceptual and linguistic information for learning lexical semantics.
• To demonstrate how global memory structures can exploit redundancies between language and perception to infer perceptual representations for words.

Main Methods:

• Development of a computational model leveraging global memory structure to link language and perception.
• Generation of inferred perceptual representations for words lacking direct perceptual experience.
• Testing the model against diverse datasets from grounded cognition experiments.

Main Results:

• The model successfully integrates language and perceptual data to learn lexical semantic representations.
• Inferred perceptual representations were generated for words without direct perceptual input.
• Model's predictions align with findings from various grounded cognition experiments.

Conclusions:

• The proposed model offers a unified framework for understanding lexical semantics by integrating language and perception.
• Perceptual simulation within a global memory model can explain diverse findings in grounded cognition research.
• This approach bridges the gap between abstract linguistic knowledge and concrete perceptual experience in word learning.