Papers by Philip C. Treleaven
Cultural Alignment in Large Language Models: An Explanatory Analysis Based on Hofstede’s Cultural Dimensions (2025.coling-main)
Copied to clipboard
| Challenge: | Large language models (LLMs) are deployed in many countries, but they fail to account for cultural variances among their potential users. |
| Approach: | They propose to use Hofstede’s cultural dimension framework to quantify cultural alignment using latent variable analysis to evaluate large language models against cultural dimensions of regions like the United States, China, and Arab countries. |
| Outcome: | The proposed model is compared against LLMs in the United States, China, and Arab countries and demonstrates that all models struggle to grasp cultural values, while GPT-4 shows a unique capability to adapt to cultural nuances, particularly in Chinese settings. |