# The Geometry of Culture: Analyzing Meaning through Word Embeddings

**Authors:** Austin C. Kozlowski, Matt Taddy, James A. Evans  
**arXiv:** [1803.09288](https://arxiv.org/abs/1803.09288)  
**Date:** 2018-03-25  
**Summary page:** https://www.summarizepaper.com/en/paper/1803.09288v1/

## Key points

- Authors introduce neural-network word embedding models for large-scale text analysis
- Word embeddings represent semantic relations between words as geometric relationships in high-dimensional space
- Models offer richer insights into cultural associations and categories compared to previous methods
- Dimensions induced by word differences in vector spaces closely correspond to dimensions of cultural meaning
- Words projected onto these dimensions reflect widely shared cultural connotations
- Various applications of word embeddings for macro-cultural investigation demonstrated, including coevolution of gender and class associations in the United States over the 20th century
- Comparative analysis of historic distinctions between markers of gender and class in the U.S. and Britain conducted
- High-dimensional models call for a shift towards "high-dimensional theorizing" regarding meanings, identities, and cultural processes

## Summary

In their paper titled "The Geometry of Culture: Analyzing Meaning through Word Embeddings," authors Austin C. Kozlowski, Matt Taddy, and James A. Evans introduce a new methodological tool called neural-network word embedding models for large-scale text analysis. They demonstrate how these models offer richer insights into cultural associations and categories compared to previous methods. Word embeddings are representations of semantic relations between words as geometric relationships between vectors in a high-dimensional space. This approach operationalizes a relational model of meaning that aligns with contemporary theories of identity and culture. The authors show that the dimensions induced by word differences in these vector spaces closely correspond to dimensions of cultural meaning. By projecting words onto these dimensions, the authors find that they reflect widely shared cultural connotations when compared to surveyed responses and labeled historical data. To test the stability of these associations, they pilot a method and then proceed to demonstrate various applications of word embeddings for macro-cultural investigation. One such application includes a longitudinal analysis of the coevolution of gender and class associations in the United States over the 20th century. Additionally, the authors conduct a comparative analysis of historic distinctions between markers of gender and class in the U.S. and Britain. Based on their findings, the authors argue that the success of these high-dimensional models calls for a shift towards "high-dimensional theorizing" regarding meanings, identities, and cultural processes. Overall, this paper highlights how neural-network word embedding models provide valuable insights into cultural associations and categories by representing semantic relations between words as geometric relationships in high-dimensional spaces. The applications showcased in this study shed light on the coevolution of gender and class associations while emphasizing the need for more comprehensive approaches to understanding meanings, identities, and cultural processes.