4. Follow up simulations
4.1 Corbin and Macquar’s finding
國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
4 . Follow up simulations
Previous chapter inspects the possible choise of extension and figures out that C-SOB model with linear criterion is the best model among the other three. Thus this model is tested with other phenomena. There are plenty of discovers in short-term recognition task besides the basis phenomena. An accurate model should be able to explain as much phenomenon as possible. Thus in this chapter, experiment findings are simulated: Corbin and Marquer (2008, 2009)’s finding, McElree and Dosher (1989)’s recent/distant negatives, and Mewhort and Johns (2000)’s extralist-feature effect.
4.1 Corbin and Macquar’s finding
Corbin and Marquer (2008, 2009) proposed an experiment which might go against my assumption. In their studies, participants are separated into two groups. One group of participants are asked to perform the regular recognition task. The other group have an additional serial recall task following the recognition task. The result shows that the recognition RT in the additional serial recall condition is longer than without serial recall and increases along with setsize as Figure 4.1.
Corbin and Marquer explains their finding with that additional serial recall task required additional attention to order, and order information is not necessary in recognition task. This additional information increases the capacity loading and results in longer RT . Also, the load of order information increases with setsize.
Hence the RT difference between two conditions increases along with setsize. Since recognition and recall require different information, they should not share the same encoding process.
In C-SOB model, serial order is coded with a position marker, as Corbin sug-gested. Positional information is essential in memory representation and a part of network structure in serial recall. I assume that serial recall and recognition share
‧
Figure 4.1: The RT of recognition in additional serial recall/recognition only con-dition. Figure is adapted from Corbin and Macquar (2008)’s figure 2.
the same encoding process and memory trace. Thus recognition also requires order information. The question that rises here is how to explain Corbin’s finding of the increasing RT in additional serial recall condition.
The RT of recognition in those two conditions should not be different if recogni-tion and recall both require the same order informarecogni-tion. However, in the addirecogni-tional recall condition, participants have to bear in mind that there is another task after recognition. The ”there is another task“ thought might reserve a certain amount of resources while participants are doing the recognition task, and therefore the performance of recognition is reduced. Though capacity is not modeled in original SOB papers (Lewandowsky & Farrell, 2008), there is a parameter which directly influence the encoding strength, ϕe. According to Equation 2.12, ϕe decreasing re-sults in smaller encoding strength (ηe), as Figure 4.2. Poor encoding should be one of the main influences in lower capacity. Thus the decreasing in capacity could be simulated with lowering the ϕe and keeping remain parameters unchanged.
There is another possible explanation for the difference between conditions. In order to handle the additional serial recall task, participants have to remember each item individually and retain more ordering information. The difference between the context of items should be large enough to distinguish each other. In recognition
‧
Figure 4.2: The encoding strength (ηe) from different ϕe.
only condition, participants could group all the items in the trial together with more similar context. Reducing the similarity between context (tc) increases the performance of serial recall task, as shown in Figure 4.3. This method is also used to simulate the grouping effect in serial recall (Lewandowsky & Farrell, 2008).
1 2 3 4 5 6
Figure 4.3: The probability of correct (P C) in serial recall task from different tc.
‧
4.1.1 Simulations of Corbin and Macquar’s finding
Two simulations are conducted to simulate Corbin and Marquer’s finding. One keeps all the parameters unchanged except ηe. This simulation represents the capacity reduction caused by a following serial recall task. Another simulation changes tc for simulating the grouping in recognition only condition. Both simulation uses liner criteria model and the same parameter values as previous simulation except ηe and tc. The ηe in additional serial recall condition is set to 200, and ηe in recognition only condition is 240. In grouping/distinguishing simulation, tc is set to .75 and .8 for additional serial recall condition and recognition only condition, respectively.
The simulation results are shown in Figure 4.4.
3 4 5 6
Figure 4.4: The response tendency of changing ϕe or tc among different setsize.
Upper figure shows the response tendency of different ϕe. Lower one shows the response tendency of different tc.
The capacity reduction simulation shows that different ϕe do change the re-sponse tendency. However, the rere-sponse tendency in different ϕe cross with each other. The change of capacity benefits in certain setsize but the others. This trend is different from Corbin and Macquar’s finding. In grouping/distinguishing simu-lation, the response tendency is also influenced by the change of tc. By grouping
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
the items together, the response tendency is higher than distinguishing them. More importantly, the difference between two conditions gradually increases along with setsize increases, and both lines do not cross. This is similar to Corbin and Mac-quar’s finding.
The simulations shows that the difference between additional serial recall con-dition and recognition only concon-dition is because different level of context distinc-tiveness. In order to achieve better performance on serial recall task, participants have to link items with more distinctive context. In recognition only condition, since there is no serial recall requirement, participants could group items together and link them with more similar context. In other hand, the capacity reduction simulation does not show the same pattern as real data. It shall not be the reason of performance difference in recognition.