Purpose and experimental procedureThe purpose of the present study was to see how well the stochastic model for free recall verbal learning, devised by Bush-Mosteller and Miller-McGill, fits to data obtained by group experiments. Twenty-five items of meaningful or nonsense syllables were presented to subjects twenty-five times in randam order. On every trial, subject were required to recall them as many as possible in 70sec.
Analysis of results: 1
Using Bush and Mosteller's avoidance model (
cf. their book: chap. 11) the recall parameter (α
1) and non recall parameter (α
2) of specific syllables (data from 20 Ss.) were estimated (Fig. 1). The significance of these values is as follows; in case of meaningful syllables, not only on recall trials, but also on non-recall trials, there occured some“conditioning”or memorization, but in case of nonsense syllables, only the recall trial had the significant effect. The empirical values were put into the theoretical equation (1, 9) of recall scores, and statman data were computed (Fig. 3). The results show in general a rather good fit, but in some of the nonsense syllables fitting was not very good. A tentative explanation might be as follows; although in estimating the recall parameter, subjects were considered to attain perfect learning, they in fact did not reach the perfect level in 25 trials.
Analysis of results: 2
Following Miller-McGill's two-parameter case, the initial recall probability (
p0) and the recall parameter (α
1) for 10 subjects were estimated (nonrecall parameter (α
2) was proved to be identity for all Ss.) (Fig. 4).
Putting these values into the theoretical equation for recall scores (2, 11) and probabilities of recalling just
k times up to the
n th trial
P(Ak, n) (2, 3) the former fits pretty well as in the case of the stat-data (Fig. 5), but the latter does not fit so well (Fig. 6).
Lastly, transitional probability (τ
k)
i.e. probability of recalling on the next trial when previous recall had taken place
k times, was computed from the data (Fig. 7). Two types of subjects can be categorized, of which the one showes continuous or gradual change of τ
k and the other showes rather non-continuous change or a qualitative gap. Using Spearmann's rank correlation between recall order on successive trials, it was found that in the former type (Fig. 9) items were recalled nearly in random order, while in the latter type items were framed in a certain order (not necessarily a meaningful one), which may be called“spontaneous liaison”(Fig. 8) on the part of subjects. Stochastic models for verbal learning were fairly good for the former type, but not necessarily so for the latter type.
In this experiment, learning was stopped before subjects attained perfect level, chiefly because of time limit to administer the experiment. If we reduce the amount of material, arrange the similarity between items, or lengthen the memorization trials, will the goodness of fit be improved? For the answer we must wait for further research.
抄録全体を表示