# Lost in Translation: Large Language Models in Non-English Content Analysis

**Authors:** Gabriel Nicholas, Aliya Bhatia  
**arXiv:** [2306.07377](https://arxiv.org/abs/2306.07377)  
**Date:** 2023-06-12  
**Summary page:** https://www.summarizepaper.com/en/paper/2306.07377v1/

## Key points

- Large language models are dominant for analyzing and generating language online
- These models are more effective in English than in other languages
- Automated systems that mediate online interactions are not proficient in languages other than English
- Multilingual language models have been developed to address this gap
- The report titled "Lost in Translation: Large Language Models in Non-English Content Analysis" explores multilingual language models
- Part I of the report explains how large language models function and highlights the disparity in available data between English and other languages
- Part II discusses challenges associated with content analysis using large language models, including bias, accuracy, cultural nuances, and ethical considerations
- Part III offers recommendations for companies, researchers, and policymakers involved in researching, developing, and deploying large and multilingual language models.

## Summary

In recent years, large language models have emerged as the dominant approach for analyzing and generating language online. However, these models are primarily designed for and work more effectively in English than in other languages. This poses a challenge as automated systems that mediate our online interactions, such as chatbots, content moderation systems, and search engines, are not as proficient in languages other than English. To address this gap, researchers and technology companies have developed multilingual language models. These models aim to extend the capabilities of large language models into languages other than English. In this report titled "Lost in Translation: Large Language Models in Non-English Content Analysis," Gabriel Nicholas and Aliya Bhatia explore how these multilingual language models work and examine their capabilities and limitations. The report is divided into three parts. Part I provides a technical explanation of how large language models function and highlights the disparity in available data between English and other languages. It also explains how multilingual language models attempt to bridge this gap by leveraging diverse datasets. Part II delves into the challenges associated with content analysis using large language models, particularly focusing on multilingual language models. The authors discuss issues such as bias, accuracy, cultural nuances, and ethical considerations when applying these models to non-English content analysis. Finally, Part III offers recommendations for companies, researchers, and policymakers involved in researching, developing, and deploying large and multilingual language models.