Does Cross-Cultural Alignment Change the Commonsense Morality of Language Models?

TL;DR

This paper investigates how aligning Japanese language models with predominantly English preference data affects their cultural morality, revealing that some moral aspects transfer while others do not.

Contribution

It highlights the impact of cross-cultural alignment on language models' morality, emphasizing the limitations of English-centric preference data for non-English cultures.

Findings

Fine-tuned models outperform SFT models in morality tasks.

English preference data transfer partially improves Japanese model morality.

Some cultural aspects of morality are not transferred through English-based alignment.

Abstract

Alignment of the language model with human preferences is a common approach to making a language model useful to end users. However, most alignment work is done in English, and human preference datasets are dominated by English, reflecting only the preferences of English-speaking annotators. Nevertheless, it is common practice to use the English preference data, either directly or by translating it into the target language, when aligning a multilingual language model. The question is whether such an alignment strategy marginalizes the preference of non-English speaking users. To this end, we investigate the effect of aligning Japanese language models with (mostly) English resources. In particular, we focus on evaluating whether the commonsense morality of the resulting fine-tuned models is aligned with Japanese culture using the JCommonsenseMorality (JCM) and ETHICS datasets. The experimental results show that the fine-tuned model outperforms the SFT model. However, it does not demonstrate the same level of improvement as a model fine-tuned using the JCM, suggesting that while some aspects of commonsense morality are transferable, others may not be.