Probing Pre-Trained Language Models for Cross-Cultural Differences in Values
This addresses the issue of cultural misalignment in AI models for users in cross-cultural applications, though it is incremental as it builds on prior bias studies.
The paper tackles the problem of how pre-trained language models encode cross-cultural values, finding that while they capture differences across cultures, these only weakly align with established value surveys.
Language embeds information about social, cultural, and political values people hold. Prior work has explored social and potentially harmful biases encoded in Pre-Trained Language models (PTLMs). However, there has been no systematic study investigating how values embedded in these models vary across cultures. In this paper, we introduce probes to study which values across cultures are embedded in these models, and whether they align with existing theories and cross-cultural value surveys. We find that PTLMs capture differences in values across cultures, but those only weakly align with established value surveys. We discuss implications of using mis-aligned models in cross-cultural settings, as well as ways of aligning PTLMs with value surveys.