漢字記憶検索過程を支える類似性空間の計算モデル : IsomapとMDSの比較

概要

論文の詳細を見る
The aim of the present study was to investigate the role of similarity in memory retrieval processes by employing similarity-judgment tasks and computational models. Five conditions of similarity-judgment tasks of Kanji characters were first conducted to 195 university students. Correlational analyses indicated that the similarity among the targets was primarily dependent on their figural feature rather than the semantic or phonetic feature, and moreover, the similarity was significantly correlated with the retrieval order of the targets during a memory retrieval task (Jimura et al., 1999). Then, the similarity space constituted by the shape of the targets was applied to two computational models in subsymbolic level: the multi-dimensional scaling (MDS) and the Isometric feature mapping (Isomap). Dimensional reduction of the shape-based similarity space indicated that the Isomap more accurately represented the similarity derived from the judgment-task and the retrieval pattern than the MDS. These results suggest the importance of the similarity in memory retrieval processes and the superiority of Isomap as a subsymbolic model of similarity.
日本行動計量学会の論文
2004-01-30