ACL

TELLME: Test-Enhanced Learning for Language Model Enrichment

TELLME：面向语言模型增强的测试增强学习

Minjun Kim, Inho Won, Hyeonseok Lim, MinKyu Kim, Junghun Yuk, Wooyoung Go, Jongyoul Park, Jungyeul Park, KyungTae Lim

机构 * Korea Advanced Institute of Science and Technology（韩国科学技术院） ； Seoul National University of Science and Technology（首尔科学技术大学） ； National Security Research Institute（国家安全研究院） 纠错

AI总结 本研究提出TELLME方法，将测试增强学习（TEL）原理与持续预训练（CPT）结合，缓解大语言模型领域自适应中数据与成本问题，在金融领域表现优于现有方法，长期记忆保留提升显著。

Comments Findings of the Association for Computational Linguistics: EACL 2026

Journal ref Findings of the Association for Computational Linguistics: EACL 2026, pages 1655-1677

MedScore: Generalizable Factuality Evaluation of Free-Form Medical Answers by Domain-adapted Claim Decomposition and Verification

Heyuan Huang, Alexandra DeLucia, Vijay Murari Tiyyala, Mark Dredze

机构 * Center for Language and Speech Processing（语言与语音处理中心） ； Johns Hopkins University（约翰霍普金斯大学） 纠错

Comments Added generalizability experiment and examples on non-medical free-form answer. Added ablation study for MedCorp verification corpus and MedScore decomposition prompt

Journal ref Findings of the Association for Computational Linguistics: ACL 2026, pages 14149-14180, San Diego, California, United States. Association for Computational Linguistics