With the emergence of low-cost wearable hardware for eye activity analysis comes the opportunity to use pupil and blink behavior during conversations to improve human computer interaction. Conversations in general can be decomposed into four segments, i.e. listening, speaking, thinking (transition from listening to speaking) and waiting (transition from speaking to listening). However, pupil and blink behavior during these segments of conversations is not yet fully understood. In this paper, pupillary response and blink in conversations are analyzed at the conversation-level and segment-level. 24 participants undertook two natural conversational tasks with different levels of communication load, while their eye activity was continuously recorded by eyewear. More intense conversation was found to impose a significant increase in pupil diameter, that is, higher communication load produces larger increases in pupil size, like other mental tasks. During conversations, pupil diameter increased more during speaking than during listening among conversations, while blink showed no difference between listening and speaking. Comparing two the transition segments, pupil size was found to be larger after speaking than before speaking, while more blinks occurred before speaking than after speaking. These findings on eye activity in conversations could be used to infer a user's communication load or predict their conversational state, which can be applied in turn-taking prediction, interruption management systems or computer-mediated conversational systems.

随着用于眼睛活动分析的低成本可穿戴硬件的出现，在对话过程中有机会利用瞳孔和眨眼行为来改善人机交互性。一般而言，对话可以分解为四个部分，即听，说，思考（从听到说的过渡）和等待（从说话到听的过渡）。但是，在这些对话阶段中的瞳孔和眨眼行为尚未完全了解。在本文中，在会话级别和段级别分析了瞳孔反应和会话中的眨眼。24名参与者进行了两个自然的会话任务，具有不同的通信负荷水平，而他们的眼睛活动被眼镜连续记录下来。人们发现，更激烈的交谈会导致瞳孔直径显着增加，也就是说，与其他智力任务一样，较高的交流负荷会导致瞳孔尺寸增加较大。在对话中，说话期间瞳孔直径的增加比在对话中聆听期间的直径增加更多，而眨眼则表明听和说之间没有差异。比较两个过渡段，发现说话后的瞳孔尺寸比说话前大，而说话前的眨眼次数多于说话后。这些关于对话中眼睛活动的发现可用于推断用户的通信负荷或预测他们的对话状态，这些结果可应用于转弯预测，中断管理系统或计算机介导的对话系统中。