# Word differences in news media of lower and higher peace countries revealed by natural language processing and machine learning

**Authors:** Larry S. Liebovitch, William Powers, Lin Shi, Allegra Chen-Carrel, Philippe Loustaunau, Peter T. Coleman  
**arXiv:** [2305.12537](https://arxiv.org/abs/2305.12537)  
**Date:** 2023-05-21  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.12537v1/

## Key points

- Language plays a crucial role in causing and reflecting social processes that lead to conflict or peace.
- Hate speech can incite violence, while peace speech can support processes that maintain peace.
- The study used peace indices, machine learning, and online news sources to identify words linked to lower-peace versus higher-peace countries.
- Little consensus exists on numerical values of peace indices due to varied measured social properties.
- Greater agreement is found for extreme cases of lower-peace and higher-peace countries.
- A data-driven approach with natural language processing and machine learning accurately classified countries based on peacefulness levels.
- Machine learning model trained on word frequencies from extreme cases computed a quantitative peace index for intermediate-peace countries successfully.
- Words related to "games," "teams," and "play" were unexpectedly representative of higher levels of peace.
- Future studies could focus on identifying social processes reflected in linguistic differences for deeper understanding.

## Summary

Language plays a crucial role in both causing and reflecting the social processes that lead to conflict or peace. Hate speech can incite violence and destruction, while peace speech can reflect and support the social processes that maintain peace. This study utilized existing peace indices, machine learning techniques, and online news sources to identify words associated with lower-peace versus higher-peace countries. The researchers found little consensus on the numerical values of peace indices due to different measured social properties. However, there is greater agreement for extreme cases of lower-peace and higher-peace countries. Using a data-driven approach with natural language processing and machine learning, the study identified words that accurately classified countries as lower-peace or higher-peace. Instead of relying on a theoretical framework to predict likely words in these countries, the researchers trained a machine learning model on word frequencies from extreme cases. This model then computed a quantitative peace index for intermediate-peace countries, successfully capturing their level of peacefulness even though they were not part of the training set. The study demonstrates how natural language processing and machine learning can generate new quantitative measures of social systems by analyzing linguistic differences. Surprisingly, analysis revealed unexpected findings such as words related to "games," "teams," and "play" being representative of higher levels of peace. Moving forward, future studies could focus on identifying the social processes reflected in these linguistic differences. Overall, this research highlights the importance of language in shaping perceptions and behaviors related to conflict and peace. By identifying key words associated with different levels of peace using advanced technologies like natural language processing and machine learning, researchers aim to deepen our understanding of how language influences societal dynamics and contributes to maintaining peaceful cultures.