言語記銘過程における系列位置効果の分析

抄録

A model of the learning process having a serial position effect was constructed. By this model we can carry out two analyses, one being an analysis of the learning process of each item in a list in terms of the learning speed and the initial learning probability, the other being an analysis of the serial position effect appearing in the initial learning probabilities in terms of pro- and retro-active inhibitions and of forgetting.
A free-recall experiment and a serialanticipation experiment were carried out in order to verify this model. Almost random numbers of two to six digits and nonsense syllables of two Japanese letters were used as material. Each list is composed of 5, 10 or 20 items of these numbers or syllables, and each subject participated in only one experiment per day.
The learning speed and the initial learning probability based on this model were obtained from the results of each experiment for each subject. The learning speed was found to be independent of position in the list, as shown in Fig. 1, 4 and 5. Moreover, by the statistical test of significance, it was also found that the learning speed is independent of the difficulty of the items, the length of lists and the kind of material. On the other hand, the initial learning probability was the Ushaped curve with respect to the position in the list, as shown in Fig. 1, 4 and 5. This tendency was analyzed by our model and we arrived at the following conclusions:
1, Pro- and retro-active inhibitions remain constant throughout all the experiments for each subject.
2. The forgetting rate does not depend on the difficulty of the item but decreases as the list lengthens.
3. The amount of immediate memory does not depend on the length of the list but decreases as the items increase in difficulty.
The curves in Fig. 2, 3, 4 and 5 were obtained by fitting our model to the results of the above experiments. By plotting the amount of immediate memory in logarithmic scale against the number of digits, it was found that the relation is linear, as shown in Fig. 6. The same relation is found between the forgetting rate in logarithmic scale and the logarithm of the length of lists, as shown in Fig. 7. By considering those relations, our model was extended so as to cover every case which ranges, over 2 to 6 digits per item and 5 to 20 items per list, and it was found that after two experiments with differing numbers of digits in the items and lists of different lengths are carried out, all the cases mentioned above can be predicted by our model.
Finally, it was observed that random numbers are better than nonsense syllables as learning material.