Where do statistically-derived indicators and human strategies meet when identifying On- and Kun-reading of Japanese kanji
Ninchikagaku_10_4_441.pdf 4.08 MB
On- and Kun-readings
The present study investigated attributes of kanji On- and Kun-readings from the perspectives of both statistical prediction and human strategy. In Study 1, discriminant analysis using the stepwise method revealed four significant indicators out of ten kanji characteristics for distinguishing On- and Kun-readings. These indicators are semantic concreteness, naming latency, special sounds and number of strokes. In Study 2, an On- or Kun-reading test is given to 30 native Japanese speakers. The result showed tendencies similar to the accuracy rates of discriminant analysis. After the test, a questionaire revealed that 6 out of 10 strategies were employed by more than 6 out of the 30 participants. Three of these were congruent with significant indicators specified by discriminant analysis, namely, semantic concreteness, naming latency and special sounds. Despite the significant indicator in Study 1, particular strategies concerning kanji strokes and radical frequency were not used by humans. Native Japanese speakers are likely to use kanji neighborhood, kanji homophones and number of morae. The results between indicators and strategies illustrate a more general point: On- and Kun-readings can be effectively predicted by discriminant analysis on the basis of various kanji characteristics; however, due to a lack of consistency in On- and Kun-readings attached to each kanji, humans can flexibly incorporate a wider variety of strategies when making their determinations.
Copyright (c) 2003 日本認知科学会