Neuroscience Letters 327 (2002) 66–70 www.elsevier.com/locate/neulet

Interactions between human explicit and implicit perceptual motor learning shown by kinematic variables Julien Lagarde a,*, Li Li b, Bernard Thon a, Richard Magill b, Emille Erbani a a

Acquisition et Transmission des Habilete´s Motrices (EA 2044), Universite´ Paul Sabatier, U.F.R.S.T.A.P.S. 118 Route de Narbonne, 31062 Toulouse Cedex 4, France b Department of Kinesiology, Motor Behavior Laboratory, Louisiana State University, Baton Rouge, LA, USA Received 26 July 2001; received in revised form 8 April 2002; accepted 8 April 2002

Abstract The goal of this study was to investigate in non-impaired humans the interaction between explicit and implicit learning in a catching task. The situation presented probabilistic contingencies between visual signals and various final pathways of the target. Subjects were asked to practice the interception task for 320 trials. Explicit instructions describing the probabilistic rules were given prior to (11 subjects), or in the middle of (11 subjects), physical practice. Eleven subjects not provided with verbal instructions served as control subjects. We measured the combination between explicit verbal instructions and implicit learning via kinematics of the end-effector and an outcome measure (i.e. spatial error). The time when explicit instructions were given resulted in systematic changes in the spatio-temporal ordering of the action. These data suggested that analyzing the way the task is executed with scrutiny allows a new understanding of how the aforementioned learning systems interact. q 2002 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Explicit learning; Implicit learning; Catching task; Kinematics

Major contributions to the understanding of human behavior have resulted in differentiating independent structures and processes for visual perception [5], memory [15], and learning [8,12]. Though providing a general picture of human functions, these theoretical and empirical constructs have to be enriched by exploring the boundaries of these independences [2,7,16]. A separation was established between explicit and implicit motor skill learning. The independent operation of explicit and implicit learning was introduced by Reber [11], stating that implicit learning consisted of the abstraction of information out of the environment without the recourse to explicit strategies for responding (p. 863). In the context of a fine-motor task, this separation was invoked to interpret a somewhat counterintuitive result, which was that subjects who attempted to apply explicit instructions during learning performed more poorly than uninstructed subjects [6]. However, a considerable debate focused on both the abstract and the unconscious nature of implicit learning. In fact, a close behavioral analysis demonstrated the acquisition, instead

* Corresponding author. Tel.: 133-06-27-67-22; fax: 133-05-6155-88-65. E-mail address: [email protected] (J. Lagarde).

of abstract rules, of partial regularities that were correlated with the sequential structure of the stimuli [4,10]. Recently this separation has received renewed interest, relying on the findings that explicit instructions can increase performances or attenuate motor deficit in implicit motor learning [1,3], and that a functional relationship does exist between explicit and implicit learning [17]. A further step can be made by investigating empirically whether the relations between explicit and implicit perceptual motor learning exceed the scope of a dichotomous theory. To bridge this gap, it appears profitable to consider the organization of the action in space and time rather than exclusively the outcome measures. This last point renders motor tasks highly attractive, offering through behavioral variables such as kinematics, a way to capture most of the features of the explicit–implicit interactions. In this letter, we report two combinations between explicit and implicit learning that show that these processes are integrated. With a view to manipulating the link between explicit and implicit learning, we communicated to two different groups of subjects exactly the same set of explicit instructions but at two different times: prior to, or in the middle of, the physical practice. A group without any explicit instructions was used as a control condition. As laws of learning describe it, during

0304-3940/02/$ - see front matter q 2002 Elsevier Science Ireland Ltd. All rights reserved. PII: S0 30 4- 39 40 ( 02) 0 03 80- 4

J. Lagarde et al. / Neuroscience Letters 327 (2002) 66–70

the first trials, a perceptual motor pattern develops to approximate the task goal, which is later refined with practice [9]. Therefore, the application of the instructions given in the middle of the practice will interact with the pattern already established and will lead to specific action kinematics. We hypothesized that the time when explicit instructions are communicated during the learning process engenders action kinematics that reflect the complexity of the explicit–implicit interactions. In order to test this hypothesis, we designed a task similar to the probabilistic catching task elaborated by Green and Flowers [6]. The interest of this task was three-fold: firstly, the probabilistic contingencies enabled the assessment of learning; secondly, these contingencies could be meaningfully described in verbal instructions; and finally, the catching response allowed us to analyze the action kinematics. This analysis of kinematics parameters was performed at a level relevant to the goal task: the peak velocity and the time to peak velocity of the end-effector [13,14]. Thirty-three adults, having normal or corrected-to-normal vision, participated in the experiment. The experimental setup consisted of a mouse connected to a personal computer. Subjects sat at a table, elbow and forearm positioned on the table, the mouse being placed between them and the monitor, at a distance of 20 cm from the monitor. Stimuli were displayed on the computer screen. The task was to intercept a descending circular target (1 cm diameter), when it intersected an interception line, with a rectangular cursor (0.5 cm £ 0.2 cm). Subjects had to move the mouse laterally with their dominant hand to control the cursor. The cursor moved horizontally at the bottom of the screen along the interception line. The motion of the target lasted 1230 ms. Firstly, the target moved 16.8 cm down from the top and center of the window, along a vertical path (920 ms). Colored circular signals (1 cm diameter; blue, red, yellow, or green) were displayed during the last 230 ms of this first part, each signal being associated with two specific trajectories of the target belonging to the second part of its motion. This association had a probability value of 0.25 with one trajectory, and of 0.75 with the other. For each signal, length and direction of the two trajectories were distinct. These signals descended on the right side of the target, and at the same speed. In the second part, the target took one of four possible straight paths (300 ms), that intersected the cursor axis at a distance from the center of 15.1 (speed, 46.4 cm/s) or 7.5 cm (speed, 28.4 cm/s), on the right or the left side. The distance was controlled as a within subject factor. Subjects performed 320 trials, consisting of 20 blocks of 16 trials each. To start a new trial, the subject had to place the cursor at the center of the screen. After each trial, the measured distance between the cursor and the target, when passing in the interception line, was provided to the subject. Cursor positions (accuracy: 0.5 mm) were recorded at a frequency of 80 Hz. Before the numerical differentiation of the position to calculate the velocity, position time series were filtered with a bi-directional, fourth order low-pass Butterworth filter (cut-off frequency: 9

67

Hz). Three experimental conditions were introduced: (1), subjects were given explicit instructions describing the probabilistic contingencies prior to practice (henceforth, group P); (2), subjects were given exactly the same explicit instructions after 160 trials at the middle of the practice (henceforth, group M); (3), control subjects were not provided with any explicit instructions (henceforth, group C). We worded for each colored signal the instructions as follows: “a colored dot appears in the first part of the display, when a red dot appears, the target goes far left in 25% of the cases and near right in 75% of the cases (...)”. In conditions P and M, the tester provided orally as many repetitions of the instructions as necessary to allow the subject to perform two consecutive complete verbal recalls of the probabilistic rules. The recall was judged complete when it included a set of four ‘if–then’ rules, each containing the true associations between one specific color dot, the two distances, the two directions and the two probability values. It is noteworthy that implicit learning can lead to the use of probabilistic contingencies or can be restricted to the refining of the visuo-motor control of the interception. In order to interpret the results, a key step was to scrutinize the possible redundancy between prior learning and the content of the middle instructions. Guided by this tenet, we investigated kinematics parameters before the communication of middle instructions, and after it. The statistical analysis of the kinematics variables was split into two parts: one that included the blocks 1–10; and the other that included the blocks 11–20. Moreover, a global statistical analysis was carried out on the outcome measure, that included the 20 blocks of practice. We took the absolute error as an outcome measure (i.e. the mean absolute difference between positions of the center of the cursor and of the center of the target when passing in the interception line). Kinematics variables consisted of the peak velocity and the time to peak velocity of the cursor movement. Subjects’ averages of these measures were computed for blocks of 32 consecutive trials. To accomplish the global analysis of absolute errors, the data were entered into a 3 (group) £ 2 (probability) £ 20 (blocks of practice) analysis of variance (ANOVA) with repeated measures on the two last factors. A significant effect of group was found (Fð2;30Þ ¼ 5:62, P , 0:01), and also of probability (Fð1;30Þ ¼ 6:31, P , 0:05). Post-hoc Newman–Keuls tests revealed that absolute errors were significantly higher in group P (37.46 mm) than in group M (19.43 mm) and group C (18.2 mm). The analysis showed an effect of block, that indicated effective learning (Fð19;570Þ ¼ 4:72, P , 0:001). The interaction between group and probability was significant (Fð2;30Þ ¼ 4:32, P , 0:05). Fig. 1 illustrates that learning in group P resulted in very large spatial errors when the probability was 0.25, and no reduction of spatial errors when the probability was 0.75. Such differences in accuracy as a function of the probabilistic contingencies were not seen for the other two groups.

68

J. Lagarde et al. / Neuroscience Letters 327 (2002) 66–70

Fig. 1. Mean absolute errors of interception for 320 trials of practice, for groups C, P, and M, for the low (0.25) and high (0.75) probability values of contingencies between visual signal and final pathway of the target.

In order to investigate the first part of practice, data were entered into 3 (group) £ 2 (probability) £ 10 (blocks of practice) ANOVAs. For the peak velocity, group and probability interacted (Fð2;30Þ ¼ 6:48, P , 0:005). The threeway interaction between group, probability and blocks was also significant (Fð18;270Þ ¼ 1:18, P , 0:05). Separated analysis of data of groups P and M showed a significant

interaction between group and probability (Fð1;20Þ ¼ 5:89, P , 0:05). The comparison of groups P and C showed an interaction of group and probability (Fð1;20Þ ¼ 8:27, P , 0:01), and an interaction of group, probability, and blocks (Fð9;180Þ ¼ 2:17, P , 0:05). No significant results were found when groups M and C were compared. The results for the time to peak velocity were quite similar and revealed an effect of group (Fð2;30Þ ¼ 5:87, P , 0:01), and that group, probability and blocks interacted (Fð2;270Þ ¼ 2:49, P , 0:005). Post-hoc Newman–Keuls tests showed that group P differed from group C and group M. Separated analysis of data of groups P and M showed an effect of group (Fð1;20Þ ¼ 6:22, P , 0:05), an effect of block (Fð9;180Þ ¼ 2:36, P , 0:05), and an effect of the interaction between probability and blocks (Fð9;180Þ ¼ 2:57, P , 0:01). Comparison of groups P and C showed an effect of group (Fð1;20Þ ¼ 7, P , 0:05) and that group, probability and blocks interacted (Fð9;180Þ ¼ 4:22, P , 0:0001). Separated analysis of groups M and C showed no significant results. The statistical analysis revealed that subjects in conditions M and C did not learn the probabilistic contingencies during the first part of the practice. Thus, in the current experiment, there was clear evidence that the middle instructions were not redundant vis-a`-vis the learning that preceded their communication. For the sake of clarity, we represented in Figs. 2 and 3 averages of kinematics variables calculated for 64 trials, therefore the total number of blocks was reduced

Fig. 2. Mean peak velocity plotted across blocks of practice, for groups C and P (left panel) and for group M (right panel), for the low and high probability values of contingencies (0.25 vs. 0.75). Means represented were calculated for 64 consecutive trials, therefore in the right panel, the middle instructions were inserted between block 5 and block 6.

J. Lagarde et al. / Neuroscience Letters 327 (2002) 66–70

69

Fig. 3. Mean time to peak velocity plotted across blocks of practice, for groups C and P (left panel) and for group M (right panel), for the low and high probability values of contingencies (0.25 vs. 0.75). Means represented were calculated for 64 consecutive trials, therefore in the right panel, the middle instructions were inserted between block 5 and block 6.

from 20 to 10, and consider that in the right panels the communication of the middle instructions was inserted between block 5 and block 6. As left panels of these figures illustrate, the probabilistic contingencies did influence the kinematics of the interception for subjects who had received the prior explicit instructions. In this last condition, the peak velocity occurred relatively early and was reduced when the probability of the contingency was 0.75. The analysis of the second part of the practice showed that the middle instructions induced the use of the probabilistic contingencies. This can be seen from blocks 6–10 in Figs. 2 and 3, right panels. However, and this was the most important result of the present study, the pattern of movement adopted in group M was not identical to the one adopted with prior instructions in group P. For the peak velocity, a 3 (group) £ 2 (probability) £ 10 (blocks of practice) ANOVA showed an effect of group (Fð2;30Þ ¼ 5:015, P , 0:05). Post-hoc Newman–Keuls tests showed that group P differed from group C and group M. Separated analysis comparing groups P and M demonstrated an effect of group (Fð1;20Þ ¼ 7:45, P , 0:05) and of probability (Fð1;20Þ ¼ 10:5, P , 0:005). When groups P and C were compared, the group was significant (Fð1;20Þ ¼ 4:57, P , 0:005) and so was the probability (Fð1;20Þ ¼ 4:43, P , 0:05), and the interaction between group and probability approached significance (Fð1;20Þ ¼ 4:31, P ¼ 0:0508). The comparison of groups M and C showed an effect of

probability (Fð1;20Þ ¼ 9:05, P , 0:01) and that group and probability interacted (Fð1;20Þ ¼ 8:8, P , 0:01). For the time to peak velocity, the analysis revealed an effect of group (Fð2;30Þ ¼ 4:34, P , 0:05) and of probability (Fð1;30Þ ¼ 15:8, P , 0:0005), and that group and probability interacted (Fð2;30Þ ¼ 3:32, P , 0:05). Post-hoc Newman– Keuls tests showed that group P differed from groups M and C. Separated analysis comparing groups P and M showed an effect of group (Fð1;20Þ ¼ 4:35, P , 0:05) and of probability (Fð1;20Þ ¼ 13:17, P , 0:005). Separated analysis of groups P and C showed an effect of group (Fð1;20Þ ¼ 6:5, P , 0:05), an effect of probability (Fð1;20Þ ¼ 5:54, P , 0:05), and of the interaction between group and probability (Fð1;20Þ ¼ 4:36, P , 0:05). The analysis comparing groups M and C showed an effect of probability (Fð1;20Þ ¼ 14:89, P , 0:001) and an interaction between group and probability (Fð1;20Þ ¼ 11:88, P , 0:005). The statistical analysis of kinematics parameters in the second part of the practice revealed that middle instructions induced a new adaptation to probabilistic contingencies. Note that there was still no effect of probabilistic contingencies for control subjects, which proved that learning in group M was not the result of extended practice. More importantly, the way the interception was executed differed between groups M and P. In the case of middle instructions, the peak velocity occurred early and was reduced when the probability of the contingency

70

J. Lagarde et al. / Neuroscience Letters 327 (2002) 66–70

was 0.75, reflecting an advanced preparation of the interception. This slowing down of the speed of the movement for high probable trials in group M was significantly less pronounced than the one observed in group P. We found also exclusively in group M, when compared with control subjects, an increase of the peak velocity when the probability associated with the advanced signal was 0.25. These adaptations to the probabilistic advanced cues could explain that the spatial errors for group M were reduced compared with group P. It seemed likely that prior instructions conducted subjects to intercept essentially targets that took the more frequent final pathway associated with each signal. The way these subjects used the advanced signals to prepare the movement did not allow them to make on-line corrections when the pathway was the low frequent one. Differently, as revealed by delayed and increased peak velocity, subjects receiving middle instructions corrected the movement when the final pathway was the low frequent one. To sum up, it was shown that behavioral adaptation to the same probabilistic contingencies differed when prior and middle instructions were compared, even though these explicit instructions were identical. The goal of this study was to investigate whether explicit learning and implicit perceptual motor learning interact. Firstly, we observed that the explicit learning of probabilistic contingencies with prior instructions led to large spatial errors and secondly that implicit learning was restricted to the development and refinement of visuo-motor processes. Measures revealing the underlying movement demonstrated that explicit instructions given prior to, or in the middle of practice, both conducted to the use of probabilistic contingencies. The most important result was that the way probabilistic advanced signals were integrated into the movement of interception differed according to the time when the instructions were given. This result provides new evidence for complex interactions between explicit and implicit learning, that were shown by the analysis of the action kinematics. The two interactions seen in the current study cannot be fully explained by a dichotomous theory. As a first approximation, the explicit–implicit integration is more likely to be rooted within a subtle, timedependent process. Indeed, further research is necessary to formulate properly a theory that accounts for the present findings, and that makes new predictions about how explicit

and implicit learning are integrated in the acquisition of a motor task. [1] Boyd, L.A. and Winstein, C.J., Implicit motor-sequence learning in humans following unilateral stroke: the impact of practice and explicit knowledge, Neurosci. Lett., 298 (2001) 65–69. [2] Creem, S.H. and Proffitt, D.R., Grasping objects by their handles: a necessary interaction between cognition and action, J. Exp. Psychol. Hum. Percept. Perform., 27 (2001) 218–228. [3] Curran, T. and Keele, S.W., Attentional and nonattentional forms of sequence learning, J. Exp. Psychol. Learn. Mem. Cogn., 19 (1993) 189–202. [4] Dulany, D.E., Carlson, R.C. and Dewey, G.I., A case of syntactical learning and judgement: how conscious and how abstract? J. Exp. Psychol. Gen., 113 (1984) 541–555. [5] Goodale, M.A. and Milner, A.D., Separate visual pathways for perception and action, Trends Neurosci., 67 (1992) 20–25. [6] Green, T.D. and Flowers, J.H., Implicit versus explicit learning processes in a probabilistic, continuous fine-motor catching task, J. Mot. Behav., 23 (1991) 293–300. [7] Jervis, C., Bennett, K., Thomas, J., Lim, S. and Castiello, U., Semantic category interference effects upon the reach-tograsp movement, Neuropsychologia, 37 (1999) 857–868. [8] Knowlton, B.J., Mangles, J.A. and Squire, L.R., A neostriatal habit learning system in humans, Science, 273 (1996) 1399– 1402. [9] Newell, K.M., Motor skill acquisition, Annu. Rev. Psychol., 42 (1991) 213–237. [10] Perruchet, P. and Pacteau, C., Synthetic grammar learning: implicit rule acquisition or explicit fragmentary knowledge? J. Exp. Psychol. Gen., 119 (1990) 264–275. [11] Reber, A.S., Implicit learning of artificial grammar, J. Verb. Learn. Verb. Behav., 5 (1967) 855–863. [12] Reber, P.J. and Squire, L.R., Encapsulation of implicit and explicit memory in sequence learning, J. Cogn. Neurosci., 10 (1998) 248–263. [13] Scho¨ ner, G., Dynamic theory of action-perception patterns: the time-to-contact paradigm, Hum. Mov. Sci., 13 (1994) 415–439. [14] Sorensen, V., Ingvalsen, R.P. and Whiting, H.T.A., The application of co-ordination dynamics to the analysis of discrete movements using table-tennis as a paradigm skill, Biol. Cybern., 85 (2001) 27–38. [15] Squire, L.R., Memory and Brain, Oxford University Press, New York, 1987. [16] Willingham, D.B., A neuropsychological theory of motor skill learning, Psychol. Rev., 105 (1998) 558–584. [17] Willingham, D.B. and Goedert-Eschmann, K., The relation between implicit and explicit learning: evidence for parallel development, Psychol. Sci., 10 (1999) 531–534.