arXiv cs.CL· Haocheng Mei, Yuchuan Tian, Ye Yuan, Enhua Wu, Chao Xu, Allen Chen, Kai Han, Yunhe Wang·· 14 小时前AI 评分18
用 Span Readout 与选择性适配改进 JEV 蕴含重排序
Improving JEV Entailment Reranking with Span Readouts and Selective Adaptation
AI 导读
研究将 mean-span readout 与选择性主干适配结合用于 JEV 蕴含重排序：mean-span pooling 以每个 NLI 选项内部上下文状态的平均值替代选项结尾指针，且不增加 head 参数；选择性适配保留 attention 与 gated DeltaNet 投影中的低秩更新、移除前馈适配器。
正文 · 原文
Abstract:JEV decision models pair a language model backbone with a lightweight scoring head to select among natural-language options. Direct selection compares candidate answers within one request, whereas entailment reranking evaluates each candidate in a separate natural language inference (NLI) request over fixed relation options. We examine whether the readout and adaptation choices used for direct selection remain effective for cross-request reranking. We combine a mean-span readout with selective backbone adaptation. Mean-span pooling replaces the option-ending pointer with an average of the contextualized states inside each NLI option, without adding head parameters. Selective adaptation retains low-rank updates in attention and gated DeltaNet projections while removing feed-forward adapters. On MC11, which averages accuracy over eleven reranking configurations from seven sources, selective adaptation improves NeoHorse-1-4B accuracy over the all-module baseline while cutting trainable parameters by more than half, and mean-span pooling yields further consistent gains at the same parameter count.
| Comments: | 25 pages, 3 figures. Project page: this https URL |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.35654 [cs.CL] |
| (or arXiv:2610.35654v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.35654 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Haocheng Mei [view email]
[v1]
Mon, 26 Oct 2026 16:02:25 UTC (40 KB)
来源：arXiv cs.CL · arxiv.org