Computational mechanisms for the meanings of words in reading

阅读理解是一个将作为输入的正字法表征映射到作为输出的语义表征的计算过程。计算过程包括两个过程：直接视觉过程（正字法->语义学：orth->sem）和语音中介过程（正字法->语音学->语义学：orth->phon->sem）。关键是两个过程的分工。模拟实验使用人工神经网络模型，学习日语阅读大声和阅读理解。该模型显示了一些阅读效果，在很大程度上与日本熟练读者的命名词汇中的这些效果相当。应用损伤方法后，按照orth->sem或orth->phon->sem过程分离模型。在阅读汉字时（尤其是orth->sem映射一致的词），orth->sem比orth->phon->sem对语义正确激活的贡献更大。然而，阅读假名却导致了相反的结果。这些结果表明，在阅读日语的两个过程之间的分工不同的脚本。

Reading comprehension is a computational process which maps orthographic representation as an input onto semantic representation as an output. The computation involves two processes : direct visual process (orthography->semantics : orth->sem, hereafter) and phonologically mediated process (orthography->phonology->semantics : orth->phon->sem). The key is the division of labor between the two processes. Simulation experiment was conducted using an artificial neural network model that learned Japanese reading aloud and reading comprehension. The model showed some reading effects, largely comparable to these effects in the naming latencies of Japanese skilled readers. After lesion method had been applied, the model was isolated either the orth->sem or the orth->phon->sem process. When reading Kanji (especially words with consistent orth->sem mappings), the orth->sem had more contribution to the correct activation of semantics than the orth->phon->sem. Reading Kana, however, leaded to the opposite results. These results suggest that the division of labor between the two processes in reading Japanese varied according to the scripts.