対連合学習における般化と分化の機制

抄録

The purpose of this report was to clarify the concepts of “generalization” and “differentiation” in verbal learning from the view of mediate process.
The experimental design used was as follows: Each of the four conditions, ht, lt, hs and ls was learned to a criterion of 5 errorless trials [ht: each of two stimulus items (adjectives) which were high similar each other was paired with a common response item (nonsense syllable), hs: each of high-similar stimulus items was paired with each different response item, lt: each of two stimulus items which were dissimilar each other was paired with a common response item, ls: each of dissimilar stimulus items was paired with each different response item.]
After the original learning (OL), each condition was immediately learned under 2 types (t′ and s′) of transfer learning (TL). (e.g., htt′ and hts′ was learned for ht in OL, in t′, two stimulus items were paired with a common response item (a letter of alphabet) and in s′, each stimulus item was paired with a different response item in paired associate learning.
Experimental hypotheses: A) According to traditional uiew of “generalization” and “differentiation”, in OL, learning is easier in ht than in lt and also in ls than in hs. In TL, in the comparison of the number of correct response on the first trial, the difference between two conditions htt′ and ltt′, hst′ and lst′, hts′ and lts′, and hss′ and lss′ should not be expected. Because it may be generally considered that the performance by the criterion of 5 errorless trials brings out perfect learning. B) According to the view of mediate process, in OL, the same expectations as hypothesis A are made. But hypothesis B should be expected on the basis of mediate process on each stimulus item, that is, learning is more difficult in lt than in ht, because in lt the differential response (dr) on each stimulus item disturbs the establishment of new experimental response integration and also more difficult in hs than in ls, because the common dr on stimulus item disturbs the establishment of new differential response integration. In TL, according to hypothesis B, the transfer effects should be expected on the basis of the degree of response integration established in OL, therefore, transfer learning is easier in condition htt′ than in ltt′ and also easier in lss′ than in hss′.
The main results were as follows: In OL, the results supported the expectations, and moreover a new remarkable finding was obtained, that is, learning by lt was not easier than ls, though ls had two times responses to be learned. It seems to me that traditional view of “generalization” and “differentiation” could not give the proper explanation about this finding. The most suitable explanation of the probable ones may agree with the view of mediate process. In TL, 1) In comparison of the numbers of correct responses on the first trial the results were not consistent with hypothesis A but were clearly consistent with hypothesis B. 2) In comparison of the saving score by [(OL-TL)/(OL+TL)]×100 we could obtain the remarkable finding that we did not find from the number of correct response on the first trial in TL, that is, the effect of OL in lt interfered transfer learning. This finding could not be explained from view of hypothesis A, even if we expand the concept to “semantic generalization”. The view of hypothesis B explained this finding in terms of the interference resulting from the weak response integration in OL, for in OL,