Exploring Large Language Models on Cross-Cultural Values in Connection with Training Methodology

TL;DR

This paper investigates how open-source large language models understand and judge diverse cultural values across countries, highlighting biases and the impact of training methods on their socio-cultural judgment capabilities.

Contribution

It provides a comprehensive analysis of how training methodology influences LLMs' understanding of cultural values and identifies biases towards Western culture.

Findings

LLMs can judge socio-cultural norms similar to humans.

Bias toward Western cultural values exists in LLM judgments.

Increasing model size improves understanding of social values.

Abstract

Large language models (LLMs) closely interact with humans, and thus need an intimate understanding of the cultural values of human society. In this paper, we explore how open-source LLMs make judgments on diverse categories of cultural values across countries, and its relation to training methodology such as model sizes, training corpus, alignment, etc. Our analysis shows that LLMs can judge socio-cultural norms similar to humans but less so on social systems and progress. In addition, LLMs tend to judge cultural values biased toward Western culture, which can be improved with training on the multilingual corpus. We also find that increasing model size helps a better understanding of social values, but smaller models can be enhanced by using synthetic data. Our analysis reveals valuable insights into the design methodology of LLMs in connection with their understanding of cultural values.