We investigate the role of information-theoretic measures for compound word reading in two languages: Mandarin Chinese and English. For each language, we report the results of two analyses: a time-to-event analysis using piece-wise additive mixed models (PAMMs) and a causal inference analysis with causal additive models (CAMs). We use the PAMM analyses to gain insight into the temporal profile of the effects of information-theoretic measures in the word naming task. For both English and Mandarin Chinese, we report early effects of the entropy of both constituents, as well as temporally widespread effects of point-wise mutual information (PMI). The CAM analyses provide further insight into the relations between lexical-distributional variables. The image that emerges from the CAM analyses is that the information-theoretic measures entropy and PMI are embedded in a carefully balanced system in which lexical-distributional properties that lead to processing difficulties are offset by lexical-distributional properties that guarantee successful communication. The information-theoretic measures have a central position in this system, and are causally influenced not only by frequency, but also by the effects of other lower-level lexical-distributional variables such as visual complexity, and phonology to orthography consistency.

我们研究了信息理论方法在两种语言（普通话和英语）中复合单词阅读中的作用。对于每种语言，我们报告两个分析的结果：使用分段加法混合模型（PAMM）进行事件分析和使用因果加法模型（CAM）进行因果推论分析。我们使用PAMM分析来深入了解单词命名任务中信息理论量度的时间变化。对于英语和普通话，我们都报告了这两种成分的熵的早期影响，以及逐点相互信息（PMI）在时间上的广泛影响。CAM分析可进一步了解词汇分布变量之间的关系。从CAM分析得出的图像是，信息理论测度的熵和PMI嵌入在精心平衡的系统中，在该系统中，导致处理困难的词汇分布属性被保证成功通信的词汇分布属性所抵消。信息理论测度在该系统中处于中心位置，并且不仅因频率而受到因果关系的影响，而且还受其他较低层次的词汇分布变量（如视觉复杂性和音韵学对拼写法一致性）的影响。