Selected Languages are All You Need for Cross-lingual Truthfulness Transfer
This addresses truthfulness challenges in multilingual AI applications, offering a practical method for improving model performance across diverse languages, though it is incremental as it builds on existing multilingual alignment techniques.
The paper tackles the problem of truthfulness gaps in multilingual large language models, especially for languages distant from English, by proposing Fact-aware Multilingual Selective Synergy (FaMSS), which selects an optimal language subset and uses translation instruction tuning to reduce representation disparity and boost cross-lingual truthfulness transfer.
Truthfulness stands out as an essential challenge for Large Language Models (LLMs). Although many works have developed various ways for truthfulness enhancement, they seldom focus on truthfulness in multilingual scenarios. Meanwhile, contemporary multilingual aligning technologies struggle to balance numerous languages and often exhibit serious truthfulness gaps across different languages, especially those that differ greatly from English. In our work, we extend truthfulness evaluation to multilingual contexts and propose a practical method for cross-lingual truthfulness transfer called Fact-aware Multilingual Selective Synergy (FaMSS). FaMSS is able to select an optimal subset of all tested languages by language bias and transfer contributions, and then employ translation instruction tuning for cross-lingual truthfulness transfer. Experimental results demonstrate that our approach can effectively reduce the multilingual representation disparity and boost cross-lingual truthfulness transfer of LLMs.