发表机构
Seoul National University; NVIDIA（首尔国立大学; 英伟达）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
研究从眼动预测阅读理解中仅注视模型的提升，基于LEXIC-Base提出两种机制注入难度信号，在一站式阅读理解任务中实验，两种机制使未见文本AUROC增益，LEXIC-Concat提升未见读者表现，LEXIC-Res存在架构边界问题。
AI 中文摘要
在最近的EyeBench基准测试中，从眼动预测阅读理解存在明显差距：使用预训练语言模型的文本感知模型AUROC达到56%-63%，而仅注视模型则处于随机水平。我们研究仅注视模型在轻量级、无语言模型条件下能提升多少。基于EyeBench的AhnCNN基线LEXIC-Base，我们提出两种机制将三个预计算的单词级难度信号（GPT-2惊奇度、词频和词长）注入每个注视输入：直接连接LEXIC-Concat和残差机制LEXIC-Res。在一站式阅读理解任务中，通过K=5种子集成训练十折，两种机制在未见文本上均产生统计学上一致的AUROC增益，LEXIC-Concat还提升了未见读者的表现。我们在LEXIC-Res中追踪到架构边界，预测头针对训练读者进行校准，向分布外读者转移时存在不足。
英文摘要
Predicting comprehension from eye movements could support adaptive reading interfaces. We present LEXIC, a compact recurrent model that predicts response correctness from fixation sequences, word frequency, and character length. It has 41.6K parameters and requires no language-model inference. Mean area under the receiver operating characteristic curve (AUROC) reaches 0.529 for Unseen Text and 0.554 for Unseen Reader on OneStop. Matched comparisons with AhnCNN show gains from both encoder redesign and lexical augmentation in these settings. Separate training and evaluation on SB-SAT also yield higher mean AUROC than AhnCNN. LEXIC occupies only 166 KB of model weights and requires 1.4 ms per trial on a single CPU core, supporting lightweight on-device inference.
CommentsSubmitted to ICCE-Asia 2026
