多语言激活引导用于多语言语言模型
Cross-Lingual Activation Steering for Multilingual Language Models
AI总结：
本文提出CLAS方法，通过激活引导提升多语言模型在非主导语言上的性能，无需修改模型权重。
摘要：
大型语言模型表现出强大的多语言能力，但在主导语言和非主导语言之间仍存在显著性能差距。先前的工作将这种差距归因于多语言表示中共享神经元和语言特定神经元之间的不平衡。我们提出了跨语言激活引导（CLAS），这是一种无需训练的推理时间干预，可以有选择地调节神经元激活。我们在分类和生成基准上评估了CLAS，分别实现了2.3%（准确率）和3.4%（F1）的平均提升，同时保持高资源语言的性能。我们发现有效的转移通过功能分歧而不是严格对齐来运作；性能提升与语言簇分离度的增加相关。我们的结果表明，有针对性的激活引导可以在不修改模型权重的情况下解锁现有模型中的潜在多语言能力。
英文摘要：
Large language models exhibit strong multilingual capabilities, yet significant performance gaps persist between dominant and non-dominant languages. Prior work attributes this gap to imbalances between shared and language-specific neurons in multilingual representations. We propose Cross-Lingual Activation Steering (CLAS), a training-free inference-time intervention that selectively modulates neuron activations. We evaluate CLAS on classification and generation benchmarks, achieving average improvements of 2.3% (Acc.) and 3.4% (F1) respectively, while maintaining high-resource language performance. We discover that effective transfer operates through functional divergence rather than strict alignment; performance gains correlate with increased language cluster separation. Our results demonstrate that targeted activation steering can unlock latent multilingual capacity in existing models without modification to model weights.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
