文字系列の学習と反応時間

抄録

In many investigations of reaction time reported so far, results have been fitted by an equation of the form
t_i=-alogp_i+b (1)
where t_i is the disjunctive reaction time for the stimulus alternative i, a and b are constants and p_i is the probability of occurrence of the alternative i. It appears that, however, the reaction time is determined by the subject's expectancy for the toccurrence of the stimulus alternative, so we assume that the above relation is better represented by the equation
t_i=-alogp′_i+b (2)
where p′_i is the subjective probability for the occurrence of the alternative i. The characteristics of p′_i have not yet been fully investigated.
It is known that the reaction time for the stimulus sequence is shorter when
(1) the subjective probability approaches to the corresponding actual probability in the sequence, and
(2) the stimulus sequence is statistically redundant.
Eased on the previous studies, we assume that the subjective probability for each stimulus in the sequence consists of several weighted terms; i.e, a term which is equally weighted for all stimuli, a term related to the frequency of each stimulus, a term related to the conditional probability of the stimulus when we regard the sequence as a simple Markoff process, a term related to the conditional probability of the stimulus when we regard the sequence as a double Markoff process, and so on, Thus a formula was given as follows;
p″_{i, m}=w₀⋅1/n+^mΣ_k=0w_k+1⋅p_{i, k}
where, p″_{i, m} means the subjective probability for the stimulus i when we take into account up to the m-th conditional probabilities, and w's mean weights for the terms mentioned above.
Using this subjective probability a new formula representing the mean reaction time was obtained. We called this new formula a new model, and compared it with formula (1), the old model.
Under several experimental conditions, the weights of subjective probability were examined. Ss were asked to press the corresponding key to each letter in the given sequence.
Experiment 1. A random sequence of letter was used as stimuli. Parameters of formula (1), a and b, were determined for each subject.
Experiment 2. A sequence was used in which letters occurred independently but their frequency was unbalanced. Weights in formula (3), w₀ and w_1-, were determined for each subject.
Experiment 3. The sequence was a simple Markoff process. The weight w_2- in formula (3) was determined for each subject.
Experiment 4. The sequence was a double Markoff process. The weight w_3- in formula (3) was determined for each subject.
Experiment 5. The same procedure as in Exp. 1 was used, and parameters a and b were determined again.
All the results obtained for different subjects under different experimental conditions showed better fit to the new model than to the old model. In Exp. 4, the weights w_3- of the double Markoff process were 0. It seems that the subject found it difficult to reach the state of perfect learning of objective probabilities, when the order of statistical dependency in the stimulus sequence was high. In other words, there was a limit in the learning of redundancy.
We have introduced the concept of the differential contribution of various orders of sequential dependency of stimuli to the construction of the subjective probability in relation to the problem of reaction time. There are still many problems left to