Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Normal and Impaired Reading of Japanese Kanji and Kana

Takao Fushimi (1), Mutsuo Ijuin (1), Naoko Sakuma (1), Masayuki Tanaka (2), Tadahisa Kondo (3), Shigeaki Amano (3), Karalyn Patterson (4), Itaru F. Tatsumi (4)

(1) Tokyo Metropolitan Institute of Gerontology, Japan
(2) Kyoto University Primate Research Institute,
(3) NTT Communication Science Laboratories,
(4) MRC Cognition and Brain Sciences Unit, Cambridge, UK

Two kinds of scripts are used in the written forms of Japanese words: morphographic kanji and phonographic kana. Whereas each kana character invariably represents a single pronunciation, the majority of kanji characters have two or more legitimate pronunciations, with one appropriate to the character in any given word. Furthermore, each kanji character has meaning while a kana character does not. On the basis of these and other differences between kanji and kana, some traditional views assume that, in reading aloud, kanji is processed by a semantic/lexical system while kana is processed by a phonological/rule system.

We review accumulating evidence from our research that argues against these traditional views. (1) In reading aloud twocharacter kanji words, normal readers are slower on lowfrequency words with statistically atypical character-sound correspondences than either high-frequency words or words with statistically typical correspondences. (2) Normal readers are easily capable of reading aloud two-character kanji nonwords. (3) Normal readers are slower on low-imageability words than high-imageability words, but the imageability effect emerges only for low-familiarity kanji words with atypical character-sound correspondences. (4) Although Japanese surface dyslexia has been described as a selective reading disorder on kanji words, recently reported cases reveal good kanji performance for high-frequency words and words with statistically typical correspondences, despite a profound deficit on low-frequency words with atypical character-sound correspondences.

(5) In reading aloud kana nonwords, normal readers are faster on pseudohomophones (orthographic nonwords with a familiar phonological pattern, created by transcribing kanji words into kana strings) than nonwords not homophonic with any words, but this significant advantage emerges only when the pseudohomophones share their pronunciations with high-imageability words. (6) Although Japanese phonological dyslexia has been considered a selective reading disorder on kana nonwords, a recently reported case showed good performance on pseudohomophones with the identical pronunciation to high-familiarity and high-imageability words, despite a profound deficit on nonhomophonic nonwords.

These data suggest that phonology of both kanji and kana strings is computed directly from orthography, with additional reliance on semantics when the direct computation is inefficient.


Full Paper

Bibliographic reference.  Fushimi, Takao / Ijuin, Mutsuo / Sakuma, Naoko / Tanaka, Masayuki / Kondo, Tadahisa / Amano, Shigeaki / Patterson, Karalyn / Tatsumi, Itaru F. (2000): "Normal and impaired reading of Japanese kanji and kana", In ICSLP-2000, vol.2, 26-31.