arXiv cs.CL· Silvia Casola, Ryan Soh-Eun Shim, Felicia K\"orner, Yuchen Mao, Barbara Plank·· 12 小时前AI 评分22
SteerEval：推理时干预增强神经摘要指标的多语言泛化能力
SteerEval: Inference-time Interventions Strengthen Multilingual Generalization in Neural Summarization Metrics
AI 导读
SteerEval 提出在推理时将多语言神经摘要指标的激活导向英语枢轴语言，可提升其与人类判断的相关性。研究覆盖基于编码器和解码器的指标，测试时干预方法在多种语言上均有效。该工作已被 EMNLP 2026 接收。
正文 · 原文
Abstract:An increasing body of work has leveraged multilingual language models for Natural Language Generation tasks such as summarization. A major empirical bottleneck in this area is the shortage of accurate and robust evaluation metrics for many languages, which hinders progress. Recent studies suggest that multilingual language models often use English as an internal pivot language, and that misalignment with this pivot can lead to degraded downstream performance. Motivated by the hypothesis that this mismatch could also apply to multilingual neural metrics, we ask whether steering their activations toward an English pivot can improve correlation with human judgments. We experiment with encoder- and decoder-based metrics and find that test-time intervention methods are effective across the board, increasing metric effectiveness for diverse languages.
| Comments: | Accepted to EMNLP 2026. Please cite the anthology BibTeX |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2601.15809 [cs.CL] |
| (or arXiv:2601.15809v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2601.15809 arXiv-issued DOI via DataCite |
Submission history
From: Silvia Casola [view email]
[v1]
Thu, 22 Jan 2026 09:49:29 UTC (3,068 KB)
[v2]
Fri, 23 Oct 2026 15:28:31 UTC (1,783 KB)
来源：arXiv cs.CL · arxiv.org