Spatial Representations Without Spatial Computations.

Area of Science:

• Cognitive Science
• Neuroscience
• Artificial Intelligence

Background:

• Cognitive maps are traditionally considered fundamentally spatial, relying on perceptual processes and dedicated brain cells.
• This view suggests spatial representations are exclusively tied to environmental navigation and sensory input.

Purpose of the Study:

• To challenge the notion that cognitive maps are solely based on spatial perception.
• To investigate if nonspatial learning mechanisms, particularly through language, can form spatial representations.
• To determine if linguistic information influences geographical judgments and cognitive map distortions.

Main Methods:

• Utilized cognitively plausible artificial intelligence models to retrieve spatial representations from natural language.
• Analyzed linguistic data to identify patterns influencing geographical judgments.
• Compared model performance with human participants judging city positions on real maps.

Main Results:

• Demonstrated that spatial representations, like geographical maps, can be accurately retrieved from natural language using nonspatial associative learning.
• Showed that linguistic information explains specific distortions in human geographical position judgments, even when using real maps.
• Found that language experience can encode and reproduce cognitive maps without a dedicated spatial system.

Conclusions:

• Language experience plays a critical role in forming cognitive maps, challenging the necessity of a purely spatial system.
• Cognitive map formation results from an interplay between spatial and nonspatial learning principles, with language being a key nonspatial component.
• Findings suggest a more flexible and integrated model of spatial representation in the brain, influenced by diverse learning experiences.