MAINTENANCE OF CROSS-DOMAIN ASSOCIATIONS IN WORKING MEMORY
6.2. The involvement of domain-specific resources
The involvement of domain-specific verbal or visuo-spatial resources in the
maintenance of cross-domain associations was tested in this fourth experiment by making use of a selective interference paradigm. In a complex span paradigm, letters in locations were to be maintained in combination with a neutral, a verbal or a spatial processing task. These processing tasks corresponded to the tone discrimination task as used in Experiment 1, 2 and 3, a semantic decision task and a spatial fit task. In order to confirm the results of Experiment 1 and 3, and verify a possible interplay between domain-general and domain-specific
resources, we also manipulated the cognitive load of these processing tasks similarly. We explained in chapter four the rationale for our predictions, which are as follows. We predict that the maintenance of cross-domain associations depends on domain-general attentional resources and on domain-specific verbal resources but not on domain-specific visuo-spatial resources. This should result in a decrease in span scores as the cognitive load increases for all of the processing tasks. We should also observe that the verbal processing task results in lower span scores than the neutral and the spatial processing task, which should in turn result in similar span scores.
Participants and Design
Ninety-seven students of the University of Geneva took part in this experiment. They were paid or given course credit for their participation. Participants were randomly attributed to one of the three processing task conditions (neural, verbal or spatial). Cognitive load (low, medium or high) was manipulated within subjects. Two and four participants in the verbal and spatial processing condition respectively did not reach the predetermined criterion of 80
% correct on the processing task in the training phase and the experiment was stopped for them after this training phase. Ninety-one participants (mean age = 21.40 years, SD = 3.65, 80 female) accomplished the actual experiment.
Materials and Procedure
Cross-domain maintenance (letters in locations) was combined with three different processing tasks according to a complex span paradigm. Series of letters in locations were created in the same way as in Experiment 1 and 3, ranging from one to five cross-domain items. Nine series were created per list length. The attribution to a cognitive load condition, its counterbalancing and the order of displaying the series was as in Experiment 1 and 3.
The three processing tasks concerned a neutral, a verbal and a spatial task. The neutral task was the same tone discrimination task as used in Experiments 1, 2 and 3. The verbal task was a semantic decision task (Vergauwe et al., 2010, 2012). Participants had to judge whether a noun corresponded to an animal or not. Twenty-four five-letter words were selected, of which 12 were animals and 12 were not. The two types of nouns were matched for frequency in French. Participants heard the word through a headphone and had to decide as fast as possible whether it corresponded to an animal or not by pressing appropriate keys (“S” or
“L”). The duration of the word presentation ranged from 440 to 600 ms. As for the tone discrimination task, one processing item was presented every 2000, 1293 or 1000 ms in the low, medium and high cognitive load condition respectively and participants had to make their choice reaction within that time interval. The spatial processing task was a spatial fit task (Vergauwe et al., 2010, 2012). Participants were presented with a box presenting two dots and a line (see Figure 6.1). They had to decide by pressing keys (“S” or “L”) whether the line could fit between the two dots or not. As in the neutral and verbal processing tasks,
participants had to decide about one spatial configuration every 2000, 1293 or 1000 ms in the low, medium and high cognitive load condition respectively. The configuration remained on screen for 1333, 862 and 667 ms in the respective cognitive load conditions, but participants had until the appearance of the next configuration to respond.
Figure 6.1: Examples of the spatial fit task.
The training and experimental procedure were the same as for the cross-domain maintenance condition in Experiment 1 with two minor changes. The presentation time of the letters in locations was reduced to 1000 ms. This was done to reduce the total time of the experiment. There were no move forward or stop rules implemented, so all participants performed the nine trials per list length (three per cognitive load) and performed all list lengths ranging from one to five. Cross-domain span scores were calculated according to the all-or-nothing scoring method, per processing task and per cognitive load.
One, eight and two participants in the neutral, verbal and spatial processing condition respectively were excluded from further analysis as they had not reached the 75 % accuracy criterion for the processing task11. The mean cross-domain span scores of the eighty
remaining participants were submitted to a 3 (Processing task: neutral, verbal or spatial) X 3 (Cognitive load: low, medium or high) repeated measure ANOVA with processing task as between subject and cognitive load as within subject factor. There was a significant effect of cognitive load, F(2, 154) = 46.33 , p < .001, η2 = .38. Span scores were higher for the low (M
= 3.12) than for the medium cognitive load condition (M = 2.80), F(1, 77) = 19.16, p < .001, η2 = .20, which were in turn higher than for the high cognitive load condition (M = 2.40), F(1, 77) = 31.41, p < .001, η2 = .29. The linear trend was also highly significant, F(1, 77) = 80.87, p < .001 and accounted for 99 % of the variance. There was no significant effect of the
11 The higher number of participants not reaching the 75 % accuracy criterion in the verbal processing condition is due to a certain number of participants not being native French speakers, which slowed down their response times.
processing task, F(2, 77) < 1, and the interaction between processing domain and cognitive load was not significant either, F(4, 154) < 1. These results were supported by the Bayesian analysis that favored the null hypothesis in both cases, pBIC (H0|D) = .97 and pBIC (H0|D) = .99 respectively. The ensemble of these results is shown in Figure 6.212. Even though Figure 6.2 seems to suggest a difference between the neutral processing condition and the verbal or spatial processing condition under high cognitive load, this was not the case (p >.10).
Figure 6.2: Span scores for cross-domain associations as a function of processing task and cognitive load in Experiment 4. Error bars represent standard errors.
The obtained results do not indicate any involvement of domain-specific resources in the maintenance of cross-domain associations. First of all, it was observed that the span scores for cross-domain associations were not lower when combined with a verbal or spatial
processing task then with a neutral processing task. Secondly, no interaction between
12 The cross-domain maintenance condition in Experiment 1 and the neutral processing condition in Experiment 4 used the same task, material and procedure except for the presentation time (1500 ms and 1000 ms for Experiment 1 and 4 respectively) and the use of move forward and stop rules in Experiment 1 but not in Experiment 4. By curiosity, we compared the cross-domain recall scores of both experiments to see whether these minimal changes would have an effect on the results. The data of the twenty-seven participants in each experiment were analyzed using a 2 (Experiment: 1 or 4) X 3 (Cognitive load: low, medium or high) repeated measure design with experiment as between subject and cognitive load as within subject factor. There was no significant difference in the cross-domain span scores between the two experiments, F(1, 52) < 1. A Bayesian analysis confirmed this result, with the posterior probability favoring the absence of any difference being pBIC (H0|D) = .86. The effect of cognitive load reported for both experiments was confirmed, F(2, 104) = 18.02 , p <
.001, η2 = .26 and no interaction was observed between experiment and cognitive load F(2, 104) = 1.06 , p = .35.
The absence of interaction was confirmed by the Bayesian analysis, pBIC (H0|D) = .97.
processing domain and cognitive load was observed. If an interaction had been present, we could have imagined that domain-specific resources take over the maintenance activity if attentional resources lack. This was not even the case. The involvement of domain-specific resources seems thus strongly denied. The involvement of domain-general attentional
resources was however confirmed. Span scores decreased as cognitive load increased and this was the case for all three processing tasks. The pattern of results suggests hence that domain-general resources are strongly implied in the maintenance of cross-domain associations, while domain-specific resources do not seem to contribute strongly to this maintenance.
We had predicted the involvement of domain-general attentional resources in the maintenance of cross-domain resources, and this appeared indeed to be the case. We had also predicted the involvement of domain-specific verbal resources, but this did not seem to be the case. Could it have been possible that the processing tasks we had used were not efficient enough to cause domain-specific interference? This was tested in Experiment 5.