Histoires Morales: A French Dataset for Assessing Moral Alignment
Thibaud Leteno, Irina Proskurina, Antoine Gourru, Julien Velcin, Charlotte Laclau, Guillaume Metzler, Christophe Gravier
2025-01-29
Summary
This paper talks about a new dataset called Histoires Morales, which is designed to test how well AI language models understand and align with French moral values and social norms. It's like creating a French version of an ethical guidebook for AI to learn from.
What's the problem?
AI language models are getting really good at understanding and using language, but they might not always understand the moral and cultural nuances of different societies. This is especially true for languages like French, where there hasn't been much research on how AI handles moral reasoning. It's like having a smart foreign exchange student who can speak perfect French but doesn't understand French cultural norms.
What's the solution?
The researchers created Histoires Morales, a collection of stories in French that describe various social situations and moral dilemmas. They took an existing English dataset called Moral Stories and carefully translated it into French, making sure it fit with French culture and values. They asked native French speakers to check and improve the translations and to make sure the moral values in the stories matched French norms. The dataset covers all sorts of situations, from how much to tip at a restaurant to how to be honest in relationships.
Why it matters?
This matters because as AI becomes more common in our daily lives, we need to make sure it understands and respects the moral values of different cultures. By creating this French dataset, researchers can now test and improve how well AI understands French moral reasoning. This could help make AI systems that are more culturally sensitive and ethically aware when used in French-speaking contexts. It's also a step towards making AI that can understand moral nuances across different languages and cultures, which is crucial as AI becomes more global.
Abstract
Aligning language models with human values is crucial, especially as they become more integrated into everyday life. While models are often adapted to user preferences, it is equally important to ensure they align with moral norms and behaviours in real-world social situations. Despite significant progress in languages like English and Chinese, French has seen little attention in this area, leaving a gap in understanding how LLMs handle moral reasoning in this language. To address this gap, we introduce Histoires Morales, a French dataset derived from Moral Stories, created through translation and subsequently refined with the assistance of native speakers to guarantee grammatical accuracy and adaptation to the French cultural context. We also rely on annotations of the moral values within the dataset to ensure their alignment with French norms. Histoires Morales covers a wide range of social situations, including differences in tipping practices, expressions of honesty in relationships, and responsibilities toward animals. To foster future research, we also conduct preliminary experiments on the alignment of multilingual models on French and English data and the robustness of the alignment. We find that while LLMs are generally aligned with human moral norms by default, they can be easily influenced with user-preference optimization for both moral and immoral data.