跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.18264 · v2

MamaReason: Unfinished Reasoning and Answer Recovery in Clinical Language Models

MamaReason:临床语言模型中未完成的推理与答案恢复

首次提交 2026-11-13 22:58
版本更新 2026-11-18 02:05
雷达观测 2026-11-18 12:34
arXiv ↗PDF ↗

摘要

Answer continuation often returns a diagnosis already stated in unfinished clinical reasoning. We study this behavior in MamaReason, adapted from dense Gemma 4 31B, on MamaBench's 217 paired maternal and child-health presentations. An explicit-commitment extractor identifies a diagnosis in 75 of 94 failed RL first passes and 75 of 93 failed SFT first passes. The continuation matches that diagnosis in 67/75 RL cases and 58/75 SFT cases, using diagnosis strings alone. Reading these commitments without additional generation recovers 30 of the 40 RL pairs restored by continuation under the archived correctness labels. Recurrence controls distinguish routine mentions from repeated answer selection. An offline second-commitment trigger measures the tradeoff between earlier readout and paired accuracy. The resulting decomposition separates diagnostic content already present in a trace, completion of an accepted answer, and paired diagnostic correctness. For clinical language-model evaluation, these are distinct outcomes that become visible when the same reasoning trace is examined through multiple readouts.

AI 论文解读

基于摘要 · 非全文审阅

作者在母婴健康配对呈现上研究临床语言模型“答案延续”行为,发现延续往往只是复述未完成推理中已出现的诊断,并将诊断内容、答案补全与配对正确性拆分为不同结果。

正在读取账号与研究记录…

版本差异

仅对比实际观测的元数据,不代表 PDF 正文或实验变化。版本可能不连续;带删除线为移除,下划线为新增。

作者变化

Thanni Adewuyi,Adepoju Jeremiah Moyondafoluwa,Dorcas Amapu,Dieko Whenu,Sylvia Essessien,Sandra Oko-Oba,Pelumi Onafadeji,Lily Aimas-Love,Samuel Omolaiye,AbdulQahhar Okenla,Abiodun Adereni

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v2 · 2026-11-18 02:05 · 最新观测版本

MamaReason: Unfinished Reasoning and Answer Recovery in Clinical Language Models

Answer continuation often returns a diagnosis already stated in unfinished clinical reasoning. We study this behavior in MamaReason, adapted from dense Gemma 4 31B, on MamaBench's 217 paired maternal and child-health presentations. An explicit-commitment extractor identifies a diagnosis in 75 of 94 failed RL first passes and 75 of 93 failed SFT first passes. The continuation matches that diagnosis in 67/75 RL cases and 58/75 SFT cases, using diagnosis strings alone. Reading these commitments without additional generation recovers 30 of the 40 RL pairs restored by continuation under the archived correctness labels. Recurrence controls distinguish routine mentions from repeated answer selection. An offline second-commitment trigger measures the tradeoff between earlier readout and paired accuracy. The resulting decomposition separates diagnostic content already present in a trace, completion of an accepted answer, and paired diagnostic correctness. For clinical language-model evaluation, these are distinct outcomes that become visible when the same reasoning trace is examined through multiple readouts.

查看此版本解读 →查看此版本原文 ↗
v1 · 2026-11-13 22:58

MamaReason: Unfinished Reasoning and Answer Recovery in Clinical Language Models

Answer continuation often returns a diagnosis already stated in unfinished clinical reasoning. We study this behavior in MamaReason, adapted from dense Gemma 4 31B, on MamaBench's 217 paired maternal and child-health presentations. An explicit-commitment extractor identifies a diagnosis in 75 of 94 failed RL first passes and 75 of 93 failed SFT first passes. The continuation matches that diagnosis in 67/75 RL cases and 58/75 SFT cases, using diagnosis strings alone. Reading these commitments without additional generation recovers 30 of the 40 RL pairs restored by continuation under the archived correctness labels. Recurrence controls distinguish routine mentions from repeated answer selection. An offline second-commitment trigger measures the tradeoff between earlier readout and paired accuracy. The resulting decomposition separates diagnostic content already present in a trace, completion of an accepted answer, and paired diagnostic correctness. For clinical language-model evaluation, these are distinct outcomes that become visible when the same reasoning trace is examined through multiple readouts.

查看此版本解读 →查看此版本原文 ↗