Cognitive, computational, and neuropsychological study on reading aloud Japanese kanji words.

我们提出，汉字和假名阅读的基本过程是在Seidenberg和McClelland（1989）提出的三角形框架中捕获的，该框架假设单词的正字法、音系和意义是在并行分布式处理的基础上相互计算的。我们的建议，即相同的架构和算法适用于汉字和假名字符串，在正常读者的表现中得到了支持。在大声朗读两字汉字时，正常阅读者对具有非典型字音对应的低频词的阅读速度比高频词或具有一致/典型对应的词慢。他们可以流利地大声读出两个汉字的非词汇。正常读者对低可想象词的阅读速度比高可想象词的阅读速度慢，但可想象效应仅在低熟悉度、非典型字音对应的汉字时才出现。在大声朗读假名非词时，当假名非词与高可想像性的汉字谐音时，正常读者对假名非词（即与真词谐音的正字法非词）的识别速度比非谐音非词快。这些影响表明，汉字和假名串的音系是直接从正字法中计算出来的，当直接计算效率不高时，有意义的支持。正常阅读者的表现性质表明，语义障碍的日语表面失读症患者在朗读低熟悉度、非典型字音对应的汉字时表现出严重的朗读缺陷，而在朗读高熟悉度、典型字音对应的汉字时表现相对保留。此外，日语语音障碍患者在大声朗读非同音非词时表现出严重的缺陷，但在朗读具有高熟悉度和/或高可想象度词的假同音同音非词时表现相对较好，无论这些非词包括何种文字。我们的假设被表面和语音阅读障碍患者的报告所证实。此外，连接主义模型成功模拟了正常阅读和受损阅读的一些模式，该模型在额外的语义模仿激活下，从汉字和假名字符串的正字法计算音系。少

We propose that the basic process of kanji and kana reading is captured in a triangle framework suggested by Seidenberg & McClelland (1989), which assumes that the orthography, phonology and meaning of words are interactively computed on the basis of parallel distributed processing. Our proposal, that the identical architecture and algorithm apply to both kanji and kana strings, found support in normal readers' performance. In reading aloud two-character kanji words, normal readers were slower at low-frequency words with atypical character-sound correspondences than at either high-frequency words or words with consistent/typical correspondences. They could read aloud two-character kanji nonwords fluently. Normal readers were slower at low-imageable words than at high-imageable words, but the imageability effect emerged only when the targets were low-familiar kanji words with atypical character-sound correspondences. In reading aloud kana nonwords, normal readers were faster at pseudoho … More mophones (viz. orthographic non-words homophonic with real words) than at non-homophonic nonwords when the pseudohomophones were homophonic with high-imageable kanji words. These effects indicate that the phonology of kanji and kana strings is computed directly from orthography, with support from meaning when direct computation is ineffi-cient.The nature of normal readers' performance suggested that Japanese surface dyslexic patients suffer-ing from semantic impairment should show severe deficit in reading aloud low-familiar kanji words with atypical character-sound correspondences, but relatively preserved performance for high-familiar words or words with typical character-sound correspondences. Furthermore, Japanese phonological dyslexic patients suffering from phonological impairment should be expected to show severe deficit in reading aloud non-homophonic nonwords, but relatively better performance for pseudohomophones homophonic with high-familiar and/or high-imageable words, in whatever script the nonwords comprise. Our assumptions were confirmed by reports of surface and phonological dyslexic patients. Furthermore, a connectionist model, which compute phonology from orthography of kanji and kana strings under additional input mimicing activation derived from semantic, successfully simulated some observed pattern of normal and impaired reading. Less