Language encodes geographical information.

Area of Science:

• Computational linguistics
• Cognitive science
• Geography

Background:

• Traditional methods for estimating city population and geographical location rely on data collection.
• Understanding how humans process and encode geographical information in language is crucial for cognitive science.

Purpose of the Study:

• To investigate if computational linguistic techniques can accurately estimate population counts and geographical coordinates (longitude, latitude) for major US cities.
• To determine if geographical proximity of cities correlates with semantic similarity in newspaper texts.
• To compare computational estimates with human estimations of city attributes.

Main Methods:

• Latent Semantic Analysis (LSA) was applied to newspaper texts to generate city similarity ratings.
• Multidimensional Scaling (MDS) was used to visualize city relationships based on LSA similarity.
• A first-order co-occurrence algorithm was employed to replicate findings.
• Population counts and geographical coordinates were estimated computationally and compared to actual data and human estimates.

Main Results:

• MDS coordinates derived from LSA showed a significant correlation with actual city longitude and latitude.
• Cities located near each other exhibited similar semantic contexts in the analyzed texts.
• Computational estimates for geographical location and population size were comparable to those made by human participants.
• The findings were robust and replicated using a co-occurrence algorithm.

Conclusions:

• Language inherently encodes geographical information that influences human understanding.
• Computational linguistic methods, such as LSA, can effectively extract and represent geographical and demographic data from text.
• This research bridges computational linguistics, cognitive science, and geography by demonstrating a quantifiable link between language use and spatial cognition.