The use of an implicit grammar task for the study of the somatic marker hypothesis (Draft april 30, 2002)
Dick J. Bierman* and Axel Cleeremans**
- * Univ. of Amsterdam,
- ** Univ. Libre Brussels,.Uni. Colorado at Boulder,
Subjects were asked to select a ‘word’ from a pair of ‘words’. Both ‘words’ were constructed using two different set of rules (grammar ‘A and grammar ‘B’). A reward was given if the subject choose the ‘grammar A’ constructed ‘word’. This was considered to be the correct choice. A punishment followed upon a choice for the ‘grammar B’ constructed ‘word’. Skin conductance was measured during each of the 100 trials. After each 10 trials subjects were asked how they selected the ‘good word’.
Task performance increased long before the subjects could even formulate a single rule from the set. In this ‘pre-conceptual’ phase of the experiment, skin conductance was larger preceding incorrect than preceding correct choices. Thus the somatic marker hypothesis was confirmed using a more complex task and using improved methodology in comparison with the original gambling task published by the ‘Damasio group’.
Exploratory analyses show that the mean response time preceding a correct choice was smaller than the response time preceding incorrect choices. This effect was strongest in the pre-conceptual phase and can cautiously be interpreted as an indication that implicit knowledge is best used with a pop-up strategy, choosing the first ‘thing that comes to mind’.
Introduction
Damasio’s somatic marker hypothesis offers a concise scientific framework to study the phenomenon of intuition [Bechara et al, 1996, 1997]. Intuition can be conceptualized as a two step process in which first knowledge is acquired, generally in an implicit way. This knowledge is thought to be marked with a positive or negative valence depending on the outcome of previous decisions using the knowledge. Secondly the intuitive decision comes about when the subject uses this knowledge non consciously by using the somatic marker associated with the knowledge.
Thus these decisions can be made faster and in situations which exclude a full analytic evaluation of the problem at hand. Negative somatic markers are thought to repress possible solutions to even come to mind as a possible solution. This model could form a basis upon which to judge intuitive capacities of, for instance, CEO’s for large businesses. However the somatic marker research so far has dealt mainly with decisions in a gambling task [Bechara et al, 1996, 1997]. Such a task is not representative for tasks in daily life or in top management in which many interacting factors may play a role. Therefore we introduced a more complex task in the form of an implicit grammar task [Cleeremans et al, 1998, Reber, 1967]. The subjects task was to select from two ‘words’ the ‘correct’ word. That is the ‘word’ that was constructed according to a specific grammar. Here the correct decisions could only be made if some rules underlying the grammar were discovered implicitly or explicitly using the feedback from earlier decisions. This approach allowed also to more subtly enquire about the explicit knowledge of the subject. In the original gambling task the question posed to subjects to assess their explicit knowledge was directly asking for the only rule underlying the task, this could help the subjects ‘discover’ this rule early. Also in our pilot studies using the same gambling task we found that subjects were suspicious about this simple rule and would search for more complicated rules dealing with sequential effects.
The randomization of the 4 stacks used in the original gambling task and of the winning and loosing cards within the stack was, at least in the non-computerized studies, a fixed one for all subjects and also the sequences were far from truly random. Thus some of the reported outcomes might be attributed to sequential patterns specific for control and patient groups. And also the claim that the subject could “….by no means infer whether the next card was a winning or loosing card …” [Bechara et al, 1996] was not justified at least not in probabilistic terms. Thus, apart from introducing a more complex task to explore the somatic marker hypothesis, we used true randomization with replacement to select the position of the correct ‘word’.
Finally we would like to point out a paradox in the somatic marker hypothesis. The somatic marker is supposed to ‘help’ the subject to avoid the risky decision. Thus a subject who ‘listens’ to his/her somatic marker will make the correct decision. But in the experiments a somatic marker is found preceding the incorrect choices! This is relevant if we want to interpret the somatic marker as a crucial element in the intuitive decision process. An intuitive person would be a person that quickly develops a somatic marker discriminating between potential positive and negative decisions, and also USES the somatic marker to make the correct decision. We will return to this issue in the discussion.
Hypotheses
In this experiment we expected that, while performing this more complex decision task, subjects would
a)show implicit learning with above chance scoring before they could explicitly indicate any rule from the grammar underlying the words (implicit learning hypothesis)
b)show a larger Somatic marker before incorrect than before correct choices. Most notably in the phase preceding the explicit knowledge.(Somatic marker hypothesis)
Method
Experimenters
The experimenters in this study were 6 students who were doing this as an obligatory part of their training in experimental methods. The experimenters were not blind with respect to the main hypothesis.
Subjects
39 subjects, 13 male and 26 female, entered this study. Their age ranged from 18 to 51 (mean=21.8, sd=6.4). The subjects were either friends of the experimenters, or freshman psychology students who have to participate in a number of experiments as an obligatory part of their training. Participants were compensated with 7 Euro, plus a bonus dependent on their performance in the task.
Materials
The stimuli consisted of pairs of ‘words’, series of 6 symbols constructed according to two sets of rules also called grammars. The symbols [, #, * and + were used. For each stimulus exposure the two words were displayed on the screen. The location of the two words, one constructed according to grammar A, the second according to grammar B was randomized (left and right). The construction of the words was also completley randomized. The 4 possible symbols for the two grammars were identical, only the transition probabilities differed (see fig. 1)
Figure 1. Transition probabilities for grammar A and B. The transition of a symbol to itself was only allowed once.
The subject started the trial by hitting any button on the keyboard. This resulted in a display of the two ‘words’. The location of the correct word (left or right) was truly random. The subject could take as much time as (s)he needed to determine which of the two words was ‘from planet A’. After the choice was entered by a single key press the computer waited 3 seconds and then marked the correct word in ‘green’ color and generated a visual and auditory feedback in (pseudo Euros).
Feedback
For incorrect choices there was a truly random ‘punishment’ of –10 or –100 Euros. These Euro’s were also taken actually away by the experimenter while the display showed the cumulative score. For correct choices there was a reward of +10 or +100 Euro’s. Feedback remained on the screen for 10 seconds after which a message appeared that the subject was allowed to initiate the next trial. (see fig. 2)
Fig 2. Timing of a single trial. Data are stored from 4 seconds before till 13 seconds after the choice between the two words.
Equipment
Skin conductance measurement
Two Ag-AgCl electrodes were attached to the middle and index finger of the non preferred hand. Isotonic paste was used. The skin conductance was measured using the Orion 4AD22 which measures skin conductance using a constant AC current (10 microamps, 100 Hz). Epochs were stored from 4 seconds before the choice till 13 seconds after a choice was made (see fig.2). The data were sampled on interupt basis with a sample frequency of 5 samples/sec.
Stimuli were presented in Fonttype / Fontsize (18) on the iMac built in screen.
Elicitation of explicit knowledge
After each 10 trials the computer generated the question: How do you come to a choice between the two words? The responses were entered by the experimenter on a standardized score form. Knowledge of the grammar was scored to have become explicit when the subject correctly formulated at least one rule and did not relapse to a state where the rule disappeared. However, rather than using this figure in the further analysis we conservatively subtracted another 5 trials because the knowledge formulated at for instance trial 50 could have become explicit anywhere between 41 and 50, so 45 was taken as the best estimate.
Procedure
Upon arrival the subject received a written instruction describing the goal of the experiment as a learning task and providing information about the possibility to earn money. After attaching the electrodes to the non-preferred hand, this hand was positioned on a small pillow and the response of skin conductance on a deep breath was measured. Subsequently a demo trial was started which familiarized the subject with the type of ‘words’ (s)he had to judge. If no question remained, the experimenter started the experiment. The experimenter stayed in the experimental room without having a view on the display. Upon each trial the experimenter adjusted the pile of (fake) money in front of the subject according to the auditory feedback. After the experiment the subject had to answer a few questions primarily dealing with the interpretation of the task.
Results
Subjects
Due to inresponsiveness with skin conductance variance typically smaller than 10 microMho three subjects were eliminated. Due to malfunctioning of the equipment 4 further subjects were removed. These decisions were made before analysis of the remaining 32 subjects started.
Data-reduction
Baseline of the skin conductance was set to the first sample taken (4 seconds before the choice of the subject). The skin conductance samples after subtraction from baseline before feedback were averaged for each trial, resulting in a variable correlating with ‘arousal’ before feedback, i.e. during the decision and anticipation phase. This variable represents Damasio’s somatic marker (SM). Subsequently these ‘arousal’ values were separately averaged for the correct and incorrect choices per subject. Only trials were used for which it was assessed that the subject had no explicit knowledge with regard to the grammar rules. This resulted in the dependent variables SM_correct and SM_incorrect.
Implicit learning hypothesis
For each subject the start of the conceptual phase (explicit knowledge phase) was determined using the method described earlier. This was compared with their performance curve. For most subjects the performance started to increase far before they entered the conceptual phase. A typical example is given in fig. 3.
Fig.3 Example of a performance curve. For each correct response the score is incremented while it is decremented for an incorrect response. The vertical dashed line gives the trial where the subject was able to formulate at least one rule.
The average performance curve with the average trial number where explicit knowledge was reported are given in fig. 4. For all subjects there was a discrepancy between the performance increase onset and the moment that they formulated the first rule they had discovered. Thus it can be concluded that implicit learning occurred in this experiment.
[insert fig.4 here]
The Somatic Marker hypothesis
Figure 5 shows the time course of the average skin conductance over all subjects using only the trials where no explicit knowledge was formulated.
Fig. 5. The skin conductance preceding, during and after feedback of incorrect and correct decisions for all subjects averaged over their pre-conceptual trials.
It can be seen that the skin conductance preceding feedback for incorrect choices is larger than the skin conductance in case of a correct choice. In table 1 the results are given in numerical form. The hypothesis was tested statistically by a binomial test comparing the number of subjects that had a larger SM before the incorrect trials (SM_incorrect) than SM_correct with the number of subject that did not behave according to this prediction. Twenty subjects (69%) did behave according to the prediction while nine did not (binomial p= 0.0315), For x subjects the preconceptual period had no incorrect or incorrect choices so that a comparison was impossible.
Exploratory analyses
SM over the whole experiment
The above analyses was also done over all 100 trials in order to explore the effect of explicit knowledge on the somatic marker hypothesis.. Interestingly the effect slightly increased. The number of subjects behaving according to the SM hypothesis raised to twenty-two while only eight behaved not consistent with the hypothesis. The mean effectsize increased with about 25%. The binomial test resulted in a p of 0.009.
Although the increase is not significant it suggests that the somatic marker might be stronger in the conceptual period. This could be explained by assuming that subjects get bored and voluntarily start to select incorrect choices knowing that this will lead to a loss.
Correct versus incorrect decisions and response times.
In most decision task there is a trade off between the response time and the performance on the decision task., However in a number of tasks that involve non-conscious processes in the realm of perception it has been found that using a pop-up strategy will improve performance [references]. This pop-up strategy consists basically in the first alternative that comes to mind preventing any analytical analysis.
We compared the mean response times for correct and incorrect decisions in the preconceptual and explicit phase of the experiment (table 1).
Table 1: Response times for incorrect and correct decisions in preconceptual and explicit phase of the experiment (units of 16.7 ms)
Mean / SD / N / Minimum / MaximumRT incorr preconceptua / 254.3 / 115.5 / 28 / 113.4 / 678.0
RT corr preconceptual / 204.8 / 98.7 / 28 / 101.7 / 526.0
RT incorr explicit / 321.4 / 158 / 16 / 96.0 / 618
RT corr explicit / 165.5 / 55.7 / 17 / 80.6 / 293.0
RT = reactietijd in units van 70 per seconde; n-expl = niet-expliciete fase; expl. = expliciete fase.
It can be seen that the response times before correct decisions is significantly smaller than before the incorrect ones. It is difficult however to interpret this result. This is because the correlation between response times and performance can be attributed either to a causal factor originating in the speed of the response (resulting in a pop-up strategy with better performance) or in the difficulty of the specific item (resulting a larger response time). Especially because the correlation is also present and even more outspoken in the explicit phase. In that phase the correct decisions are taken in only 2.76 sec but the incorrect ones take about twice that long suggesting that these concerned trials where the subject’s explicit knowledge was insufficient to solve the problem.
Discussion
The major finding in this experiment is that the Somatic Marker hypothesis was supported in a conceptual replication which has more ecological validity than the original gambling task. This lends support to the suggestion that this somatic marker process is important in every-day complex decisions in problems which are under-specified or where not enough time is available for a complete analytical solution. However, as noted in the introduction, this finding is also confusing if one wants to conclude that the somatic marker is helpful in avoiding incorrect decisions. The Somatic marker is strongest preceding incorrect decisions. The most parsimonious conclusion is that the somatic marker ‘causes’ the incorrect decision. This is completely in conflict with the interpretation that seems to be favored by the Damasio group. In order to solve this dilemma a new type of experiment is needed where we can separate the early speculations of the subject from the final decisions. Such a separation could also shed some light on the proper interpretations with regard to response times. If one could disentangle this process the hypothesized three components of the intuitive decision process, a) implicit learning, b) somatic marking, and c) acting according to the somatic marker, can be studied separately and thus would allow for the development of an intuition measuring instrument. This is the goal of our future work.
References
Bechara, A., Tranel, D., Damasio, H., & Damasio, A. R. (1996). Failure to respond to anticipated future outcomes following damage to prefrontal cortex. Cerebral cortex, Vol. 6 No. 2, 215-225.
Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (1997). Deciding advantageously before knowing the advantageous strategy. Science, 275, 1293-1295.
Cleeremans, A., Destrebecqz, A., & Boyer, M. (1998). Implicit learning: news from the front. Trends in cognitive sciences, Vol.2 No.10, 406-416.
Reber, A.S. (1967) Implicit learning of artificial grammars. Journal of verbal learning and verbal behavior, 6, 855-863.