CALM: Unleashing the Cross-Lingual Self-Aligning Ability of Language Model Question Answering

TL;DR

This paper introduces CALM, a method that improves cross-lingual consistency in language models by aligning knowledge across languages using self-selected responses and preference optimization, enhancing multilingual question answering.

Contribution

CALM is a novel approach that leverages self-alignment and preference optimization to improve cross-lingual knowledge consistency in language models.

Findings

CALM improves accuracy in multilingual QA tasks.

Increasing languages in training enhances model consistency.

CALM outperforms baseline methods in zero-shot settings.

Abstract

Large Language Models (LLMs) are pretrained on extensive multilingual corpora to acquire both language-specific cultural knowledge and general knowledge. Ideally, while LLMs should provide consistent responses to culture-independent questions across languages, we observe significant performance disparities. To address this, we explore the Cross-Lingual Self-Aligning ability of Language Models (CALM) to align knowledge across languages. Specifically, for a given question, we sample multiple responses across different languages and select the most self-consistent response as the target, leaving the remaining responses as negative examples. We then employ direct preference optimization (DPO) to align the model's knowledge across different languages. Evaluations on the MEDQA and X-CSQA datasets demonstrate CALM's effectiveness in enhancing cross-lingual knowledge question answering, both in zero-shot and retrieval-augmented settings. We also found that increasing the number of languages involved in CALM training leads to higher accuracy and consistency. We offer a qualitative analysis of how cross-lingual consistency can enhance knowledge alignment and explore the method's generalizability.