研究人员推出了 G-IdiomAlign，这是一个旨在评估大型语言模型在不同语言之间对齐习语能力的新基准。该基准使用维基词典的英文词义注释作为枢纽来锚定习语，解决了习语非组合性本质和表面形式基础薄弱带来的挑战。初步测试显示，大型语言模型常常表现出对字面翻译的偏见，尤其是在低资源语言方面，并且使用词义注释可以提高受控生成任务的性能，尽管仍有很大的改进空间。 AI
影响 该基准有望推动大型语言模型处理习语等细微语言现象的能力的提升，从而增强跨语言交流。
排序理由 该集群描述了一个用于评估大型语言模型能力的新学术基准，该基准在一篇研究论文中提出。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2606.18989v1 Announce Type: cross Abstract: Idioms are difficult to transfer across languages due to their non-compositionality and weak surface-form grounding, making literal mappings unreliable. We present G-IdiomAlign, a gloss-pivoted benchmark where each idiom is anchor…
Idioms are difficult to transfer across languages due to their non-compositionality and weak surface-form grounding, making literal mappings unreliable. We present G-IdiomAlign, a gloss-pivoted benchmark where each idiom is anchored by an English gloss from Wiktionary. We further…
Idioms are difficult to transfer across languages due to their non-compositionality and weak surface-form grounding, making literal mappings unreliable. We present G-IdiomAlign, a gloss-pivoted benchmark where each idiom is anchored by an English gloss from Wiktionary. We further…