Axel Cleeremans

Experience-dependent changes in cerebral activation during human Rem sleep

The function of rapid-eye-movement (REM) sleep is still unknown. One prevailing hypothesis suggests that REM sleep is important in processing memory traces. Here, using positron emission tomography (PET) and regional cerebral blood flow measurements, we show that waking experience influences regional brain activity during subsequent sleep. Several brain areas activated during the execution of a serial reaction time task during wakefulness were significantly more active during REM sleep in subjects previously trained on the task than in non-trained subjects. These results support the hypothesis that memory traces are processed during REM sleep in humans.

Can sequence learning be implicit?: New evidence with the process dissociation procedure

Can we learn without awareness? Although this issue has been extensively explored through studies of implicit learning, there is currently no agreement about the extent to which knowledge can be acquired and projected onto performance in an unconscious way. The controversy, like that surrounding implicit memory, seems to be at least in part attributable to unquestioned acceptance of the unrealistic assumption that tasks are process-pure—that is, that a given task exclusively involves either implicit or explicit knowledge. Methods such as the process dissociation procedure (PDP, Jacoby, 1991) have been developed to overcome the conceptual limitations of the process purity assumption but have seldom been used in the context of implicit learning research. In this paper, we show how the PDP can be applied to a free generation task so as to disentangle explicit and implicit sequence learning. Our results indicate that subjects who are denied preparation to the next stimulus nevertheless exhibit knowledge of the sequence through their reaction time performance despite remaining unable (1) to project this knowledge in a recognition task and (2) to refrain from expressing their knowledge when specifically instructed to do so. These findings provide strong evidence that sequence learning can be unconscious.

Finite state automata and simple recurrent networks

We explore a network architecture introduced by Elman (1988) for predicting successive elements of a sequence. The network uses the pattern of activation over a set of hidden units from time-step t1, together with element t, to predict element t 1. When the network is trained with strings from a particular finite-state grammar, it can learn to be a perfect finite-state recognizer for the grammar. When the network has a minimal number of hidden units, patterns on the hidden units come to correspond to the nodes of the grammar, although this correspondence is not necessary for the network to act as a perfect finite-state recognizer. We explore the conditions under which the network can carry information about distant sequential contingencies across intervening elements. Such information is maintained with relative ease if it is relevant at each intermediate step; it tends to be lost when intervening elements do not depend on it. At first glance this may suggest that such networks are not relevant to natural language, in which dependencies may span indefinite distances. However, embeddings in natural language are not completely independent of earlier information. The final simulation shows that long distance sequential contingencies can be encoded by the network even if only subtle statistical properties of embedded strings depend on the early information.

Learned material content and acquisition level modulate cerebral reactivation during posttraining rapid-eye-movements sleep

We have previously shown that several brain areas are activated both during sequence learning at wake and during subsequent rapid-eye-movements (REM) sleep (Nat. Neurosci. 3 (2000) 831-836), suggesting that REM sleep participates in the reprocessing of recent memory traces in humans. However, the nature of the reprocessed information remains open. Here, we show that regional cerebral reactivation during posttraining REM sleep is not merely related to the acquisition of basic visuomotor skills during prior practice of the serial reaction time task, but rather to the implicit acquisition of the probabilistic rules that defined stimulus sequences. Moreover, functional connections between the reactivated cuneus and the striatum--the latter being critical for implicit sequence learning--are reinforced during REM sleep after practice on a probabilistic rather than on a random sequence of stimuli. Our results therefore support the hypothesis that REM sleep is deeply involved in the reprocessing and optimization of the high-order information contained in the material to be learned. In addition, we show that the level of acquisition of probabilistic rules attained prior to sleep is correlated to the increase in regional cerebral blood flow during subsequent REM sleep. This suggests that posttraining cerebral reactivation is modulated by the strength of the memory traces developed during the learning episode. Our data provide the first experimental evidence for a link between behavioral performance and cerebral reactivation during REM sleep.

Implicit learning out of the lab: the case of orthographic regularities

Childrens (Grades 1 to 5) implicit learning of French orthographic regularities was investigated through nonword judgment (Experiments 1 and 2) and completion (Experiments 3a and 3b) tasks. Children were increasingly sensitive to (a) the frequency of double consonants (Experiments 1, 2, and 3a), (b) the fact that vowels can never be doubled (Experiment 2), and (c) the legal position of double consonants (Experiments 2 and 3b). The latter effect transferred to never doubled consonants but with a decrement in performance. Moreover, this decrement persisted without any trend toward fading, even after the massive amounts of experience provided by years of practice. This result runs against the idea that transfer to novel material is indicative of abstract rule-based knowledge and suggests instead the action of mechanisms sensitive to the statistical properties of the material. A connectionist model is proposed as an instantiation of such mechanisms.

Measuring consciousness: relating behavioural and neurophysiological approaches

The resurgent science of consciousness has been accompanied by a recent emphasis on the problem of measurement. Having dependable measures of consciousness is essential both for mapping experimental evidence to theory and for designing perspicuous experiments. Here, we review a series of behavioural and brain-based measures, assessing their ability to track graded consciousness and clarifying how they relate to each other by showing what theories are presupposed by each. We identify possible and actual conflicts among measures that can stimulate new experiments, and we conclude that measures must prove themselves by iteratively building knowledge in the context of theoretical frameworks. Advances in measuring consciousness have implications for basic cognitive neuroscience, for comparative studies of consciousness and for clinical applications.

Measuring consciousness: Is one measure better than the other?

What is the best way of assessing the extent to which people are aware of a stimulus? Here, using a masked visual identification task, we compared three measures of subjective awareness: The Perceptual Awareness Scale (PAS), through which participants are asked to rate the clarity of their visual experience; confidence ratings (CR), through which participants express their confidence in their identification decisions, and Post-decision wagering (PDW), in which participants place a monetary wager on their decisions. We conducted detailed explorations of the relationships between awareness and identification performance, looking to determine (1) which scale best correlates with performance, and (2) whether we can detect performance in the absence of awareness and how the scales differ from each other in terms of revealing such unconscious processing. Based on these findings we discuss whether perceptual awareness should be considered graded or dichotomous. Results showed that PAS showed a much stronger performance-awareness correlation than either CR or PDW, particularly for low stimulus intensities. In general, all scales indicated above-chance performance when participants claimed not to have seen anything. However, such above-chance performance only showed when we also observed a correlation between awareness and performance. Thus (1) PAS seems to be the most exhaustive measure of awareness, and (2) we find support for above-chance performance in the absence of subjective awareness, but such unconscious knowledge only contributes to performance when we observe conscious knowledge as well. Similarities and differences between scales are discussed in the light of consciousness theories and response strategies.

Graded State Machines: The Representation of Temporal Contingencies in Simple Recurrent Networks

We explore a network architecture introduced by Elman (1990) for predicting successive elements of a sequence. The network uses the pattern of activation over a set of hidden units from time-step t-1, together with element t, to predict element t+1. When the network is trained with strings from a particular finite-state grammar, it can learn to be a perfect finite-state recognizer for the grammar. When the net has a minimal number of hidden units, patterns on the hidden units come to correspond to the nodes of the grammar, however, this correspondence is not necessary for the network to act as a perfect finite-state recognizer. Next, we provide a detailed analysis of how the network acquires its internal representations. We show that the network progressively encodes more and more temporal context by means of a probability analysis. Finally, we explore the conditions under which the network can carry information about distant sequential contingencies across intervening elements to distant elements. Such information is maintained with relative ease if it is relevant at each intermediate step, it tends to be lost when intervening elements do not depend on it. At first glance this may suggest that such networks are not relevant to natural language, in which dependencies may span indefinite distances. However, embed dings in natural language are not completely independent of earlier information. The final simulation shows that long distance sequential contingencies can be encoded by the network even if only subtle statistical properties of embedded strings depend on the early information. The network encodes long-distance dependencies by shading internal representations that are responsible for processing common embeddings in otherwise different sequences. This ability to represent simultaneously similarities and differences between several sequences relies on the graded nature of representations used by the network, which contrast with the finite states of traditional automata. For this reason, the network and other similar architectures may be called Graded State Machines.

Striatum forever, despite sequence learning variability : A random effect analysis of PET data

This PET study is concerned with the what, where, and how of implicit sequence learning. In contrast with previous studies imaging the serial reaction time (SRT) task, the sequence of successive locations was determined by a probabilistic finite‐state grammar. The implicit acquisition of statistical relationships between serially ordered elements (i.e., what) was studied scan by scan, aiming to evidence the brain areas (i.e., where) specifically involved in the implicit processing of this core component of sequential higher‐order knowledge. As behavioural results demonstrate between‐ and within‐subjects variability in the implicit acquisition of sequential knowledge through practice, functional PET data were modelled using a random‐effect model analysis (i.e., how) to account for both sources of behavioural variability. First, two mean condition images were created per subject depending on the presence or not of implicit sequential knowledge at the time of each of the 12 scans. Next, direct comparison of these mean condition images provided the brain areas involved in sequential knowledge processing. Using this approach, we have shown that the striatum is involved in more than simple pairwise associations and that it has the capacity to process higher‐order knowledge. We suggest that the striatum is not only involved in the implicit automatization of serial information through prefrontal cortex‐caudate nucleus networks, but also that it plays a significant role for the selection of the most appropriate responses in the context created by both the current and previous stimuli, thus contributing to better efficiency and faster response preparation in the SRT task. Hum. Brain Mapping 10:179–194, 2000.

Comparing direct and indirect measures of sequence learning.

Comparing the sensitivity of similar direct and indirect measures is proposed as the best way to provide evidence for unconscious learning. The authors apply this approach, first proposed by E. M. Reingold and P. M. Merikle (1988), to a choice reaction-time task in which the material is generated probabilistically on the basis of a finite-state grammar (A. Cleeremans, 1993). The data show that participants can learn about the structure of the stimulus material over training with the choice reaction-time task, but only to a limited extent - a result that is well predicted by the simple recurrent network model of A. Cleeremans and J. L. McClelland (1991). Participants can also use some of this knowledge to perform a subsequent generation task. However, detailed partial correlational analyses that control for knowledge as assessed by the generation task show that large effects of sequence learning are exclusively expressed through reaction time. This result suggests that at least some of this learning cannot be characterized as conscious.