Statistical learning in a serial reaction time task: access to separable statistical cues by individual learners

Cue validity effects in response preparation: a pupillometric study

This study examined the effects of cue validity and cue difficulty on response preparation to provide a test of the Grouping Model [Adam, J.J., Hommel, B. and Umiltà, C., 2003. Preparing for perception and action (I): the role of grouping in the response-cuing paradigm. Cognit. Psychol. 46(3), 302-58, Adam, J.J., Hommel, B. and Umiltà, C., 2005. Preparing for perception and action (II) automatic and effortful processes in response cuing. Vis. Cogn. 12(8), 1444-1473.]. We used the pupillary response to index the cognitive processing load during and after the preparatory interval (2 s). Twenty-two participants performed the finger-cuing tasks with valid (75%) and invalid (25%) cues. Results showed longer reaction times, more errors, and larger pupil dilations for invalid than valid cues. During the preparation interval, pupil dilation varied systematically with cue difficulty, with easy cues (specifying 2 fingers on 1 hand) showing less pupil dilation than difficult cues (specifying 2 fingers on 2 hands). After the preparation interval, this pattern of differential pupil dilation as a function of cue difficulty reversed for invalid cues, suggesting that cues which incorrectly specified fingers on one hand required more effortful reprogramming operations than cues which incorrectly specified fingers on two hands. These outcomes were consistent with predictions derived from the Grouping Model. Finally, all participants exhibited two distinct pupil dilation strategies: an "early" strategy in which the onset of the main pupil dilation was tied to onset of the cue, and a "late" strategy in which the onset of the main pupil dilation was tied to the onset of the target. Thus, whereas the early pupil dilation strategy showed a strong dilation during the preparation interval, the late pupil strategy showed a strong constriction. Interestingly, only the late onset pupil dilation strategy revealed the above reported sensitivity to cue difficulty, showing for the first time that the well-known pupil's sensitivity to task difficulty can also emerge when the pupil is constricting instead of dilating.

Segmenting dynamic human action via statistical structure

Human social, cognitive, and linguistic functioning depends on skills for rapidly processing action. Identifying distinct acts within the dynamic motion flow is one basic component of action processing; for example, skill at segmenting action is foundational to action categorization, verb learning, and comprehension of novel action sequences. Yet little is currently known about mechanisms that may subserve action segmentation. The present research documents that adults can register statistical regularities providing clues to action segmentation. This finding provides new evidence that structural knowledge gained by mechanisms such as statistical learning can play a role in action segmentation, and highlights a striking parallel between processing of action and processing in other domains, such as language.

Is the mind inherently forward looking? Comparing prediction and retrodiction

It has been suggested that prediction may be an organizing principle of the mind and/or the neocortex, with cognitive machinery specifically engineered to detect forward-looking temporal relationships, rather than merely associating temporally contiguous events. There is a remarkable absence of behavioral tests of this idea, however. To address this gap, we showed subjects sequences of shapes governed by stochastic Markov processes, and then asked them to choose which shape reliably came after a probe shape (prediction test) or before a probe shape (retrodiction test). Prediction was never superior to retrodiction, even when subjects were forewarned of a forward-directional test.

Strategic Modulation of Cognitive Control

The neural substrate of cognitive control is thought to comprise an evaluative component located in the anterior cingulate cortex (ACC) and an executive component in the prefrontal cortex (PFC). The control mechanism itself is mainly local, triggered by response conflict (monitored by the ACC) and involving the allocation of executive resources (recruited by the PFC) in a trial-to-trial fashion. However, another way to achieve control would be to use a strategic mechanism based on long-term prediction of upcoming events and on a chronic response strategy that ignores local features of the task. In the current study, we showed that such a strategic control mechanism was based on a functional dissociation or complementary relationship between the ACC and the PFC. When information in the environment was available to make predictions about upcoming stimuli, local task features (e.g., response conflict) were no longer used as a control signal. We suggest that having separate control mechanisms based on local or global task features allows humans to be persistent in pursuing their goals, yet flexible enough to adapt to changes in the environment.

Implicit learning and statistical learning: one phenomenon, two approaches

The domain-general learning mechanisms elicited in incidental learning situations are of potential interest in many research fields, including language acquisition, object knowledge formation and motor learning. They have been the focus of studies on implicit learning for nearly 40 years. Stemming from a different research tradition, studies on statistical learning carried out in the past 10 years after the seminal studies by Saffran and collaborators, appear to be closely related, and the similarity between the two approaches is strengthened further by their recent evolution. However, implicit learning and statistical learning research favor different interpretations, focusing on the formation of chunks and statistical computations, respectively. We examine these differing approaches and suggest that this divergence opens up a major theoretical challenge for future studies.

The automaticity of visual statistical learning

The visual environment contains massive amounts of information involving the relations between objects in space and time, and recent studies of visual statistical learning (VSL) have suggested that this information can be automatically extracted by the visual system. The experiments reported in this article explore the automaticity of VSL in several ways, using both explicit familiarity and implicit response-time measures. The results demonstrate that (a) the input to VSL is gated by selective attention, (b) VSL is nevertheless an implicit process because it operates during a cover task and without awareness of the underlying statistical patterns, and (c) VSL constructs abstracted representations that are then invariant to changes in extraneous surface features. These results fuel the conclusion that VSL both is and is not automatic: It requires attention to select the relevant population of stimuli, but the resulting learning then occurs without intent or awareness.

Can a Preschooler’s Mistaken Belief Benefit Learning?

Young children erroneously believe that differences either in mass alone or in volume alone can predict differences in sinking speed. The current study was an attempt to teach preschool children that neither mass nor volume alone is predictive for sinking speed. Instead, it is the average density of an object that can predict differences in sinking speed. Twenty-four 4-to 6-year-olds participated. In an initial phase, children’s mistaken beliefs about the effects of mass and volume on sinking speed were called to their minds. Then they were presented with demonstrations of sinking objects that disconfirmed these mistaken beliefs. The findings show that preschool children can replace mistaken beliefs and learn that two dimensions, originally thought of as being relevant, are indeed irrelevant. Children who did not perform correctly demonstrated a mass bias. The results also shed light on the origins of this bias.

Statistical approaches to language acquisition and the self-organizing consciousness: a reversal of perspective

Recent years have seen the upsurge of a new approach to language that moves away from the rule-based conventional framework. In this approach, mostly supported by the success of connectionist models, children learn language by exploiting the distributional properties of the input. It is argued in this paper that, in the same way as conforming to rules does not imply the existence of mental rules, conforming to statistical regularities does not imply that statistical computations are performed mentally. Sensitivity to statistical regularities can alternatively be conceived of as a by-product of the recurrent interplay between the properties of the current conscious content and the properties of the linguistic and extralinguistic environment. The validity of including the content of conscious experiences in an otherwise standard dynamical approach rooted in the notion of self-organization is discussed.

Learning at a distance II. Statistical learning of non-adjacent dependencies in a non-human primate

In earlier work we have shown that adults, infants, and cotton-top tamarin monkeys are capable of computing the probability with which syllables occur in particular orders in rapidly presented streams of human speech, and of using these probabilities to group adjacent syllables into word-like units. We have also investigated adults' learning of regularities among elements that are not adjacent, and have found strong selectivities in their ability to learn various kinds of non-adjacent regularities. In the present paper we investigate the learning of these same non-adjacent regularities in tamarin monkeys, using the same materials and familiarization methods. Three types of languages were constructed. In one, words were formed by statistical regularities between non-adjacent syllables. Words contained predictable relations between syllables 1 and 3; syllable 2 varied. In a second type of language, words were formed by statistical regularities between non-adjacent segments. Words contained predictable relations between consonants; the vowels varied. In a third type of language, also formed by regularities between non-adjacent segments, words contained predictable relations between vowels; the consonants varied. Tamarin monkeys were exposed to these languages in the same fashion as adults (21 min of exposure to a continuous speech stream) and were then tested in a playback paradigm measuring spontaneous looking (no reinforcement). Adult subjects learned the second and third types of language easily, but failed to learn the first. However, tamarin monkeys showed a different pattern, learning the first and third type of languages but not the second. These differences held up over multiple replications, using different sounds instantiating each of the patterns. These results suggest differences among learners in the elementary units perceived in speech (syllables, consonants, and vowels) and/or the distance over which such units can be related, and therefore differences among learners in the types of patterned regularities they can acquire. Such studies with tamarins open interesting questions about the perceptual and computational capacities of human learners that may be essential for language acquisition, and how they may differ from those of non-human primates.

Probability detection mechanisms and motor learning

The automatic detection of patterns or regularities in the environment is central to certain forms of motor learning, which are largely procedural and implicit. The rules underlying the detection and use of probabilistic information in the perceptual-motor domain are largely unknown. We conducted two experiments involving a motor learning task with direct and crossed mapping of motor responses in which probabilities were present at the stimulus set level, the response set level, and at the level of stimulus-response (S-R) mapping. We manipulated only one level at a time, while controlling for the other two. The results show that probabilities were detected only when present at the S-R mapping and motor levels, but not at the perceptual one (experiment 1), unless the perceptual features have a dimensional overlap with the S-R mapping rule (experiment 2). The effects of probability detection were mostly facilitatory at the S-R mapping, both facilitatory and inhibitory at the perceptual level, and predominantly inhibitory at the response-set level. The facilitatory effects were based on learning the absolute frequencies first and transitional probabilities later (for the S-R mapping rule) or both types of information at the same time (for perceptual level), whereas the inhibitory effects were based on learning first the transitional probabilities. Our data suggest that both absolute frequencies and transitional probabilities are used in motor learning, but in different temporal orders, according to the probabilistic properties of the environment. The results support the idea that separate neural circuits may be involved in detecting absolute frequencies as compared to transitional probabilities.

Learning at a distance I. Statistical learning of non-adjacent dependencies

In earlier work we have shown that adults, young children, and infants are capable of computing transitional probabilities among adjacent syllables in rapidly presented streams of speech, and of using these statistics to group adjacent syllables into word-like units. In the present experiments we ask whether adult learners are also capable of such computations when the only available patterns occur in non-adjacent elements. In the first experiment, we present streams of speech in which precisely the same kinds of syllable regularities occur as in our previous studies, except that the patterned relations among syllables occur between non-adjacent syllables (with an intervening syllable that is unrelated). Under these circumstances we do not obtain our previous results: learners are quite poor at acquiring regular relations among non-adjacent syllables, even when the patterns are objectively quite simple. In subsequent experiments we show that learners are, in contrast, quite capable of acquiring patterned relations among non-adjacent segments-both non-adjacent consonants (with an intervening vocalic segment that is unrelated) and non-adjacent vowels (with an intervening consonantal segment that is unrelated). Finally, we discuss why human learners display these strong differences in learning differing types of non-adjacent regularities, and we conclude by suggesting that these contrasts in learnability may account for why human languages display non-adjacent regularities of one type much more widely than non-adjacent regularities of the other type.

Anticipating by Pigeons Depends on Local Statistical Information in a Serial Response Time Task

Pigeons responded in a serial response time task patterned after that of M. J. Nissen and P. Bullemer (1987) with humans. Experiment 1 produced global facilitation: Response times in repeating lists of locations were faster than when locations were random. Response time to a spatial location was also a function of both that location's 1st- and 2nd-order local predictability, in rough agreement with the Hick-Hyman law, according to which response time is a linear function of amount of information. Experiment 2 showed that both local and global facilitation is limited to moderate response-to-stimulus intervals of about 0.50 to 2.00 s. Experiment 3 showed that response time did not depend on global statistical information. Overall, local and global performances depended on local statistical information, but global performance did not depend on global information. Local facilitation was interpreted in plain English as anticipating.

Distant Melodies: Statistical Learning of Nonadjacent Dependencies in Tone Sequences

Human listeners can keep track of statistical regularities among temporally adjacent elements in both speech and musical streams. However, for speech streams, when statistical regularities occur among nonadjacent elements, only certain types of patterns are acquired. Here, using musical tone sequences, the authors investigate nonadjacent learning. When the elements were all similar in pitch range and timbre, learners acquired moderate regularities among adjacent tones but did not acquire highly consistent regularities among nonadjacent tones. However, when elements differed in pitch range or timbre, learners acquired statistical regularities among the similar, but temporally nonadjacent, elements. Finally, with a moderate grouping cue, both adjacent and nonadjacent statistics were learned, indicating that statistical learning is governed not only by temporal adjacency but also by Gestalt principles of similarity.

Musical Learning and Language Development

How do infant learners acquire structure, given complex environments? In this chapter, we consider the role played by statistical learning-tracking patterns in the environment-in the acquisition of language and music. The results from a series of experiments suggest that similar learning mechanisms may operate in both domains, but that these mechanisms are also influenced by domain-specific perceptual biases and input structure.

The influence of response--effect compatibility in a serial reaction time task

Participants performed a serial reaction time task, responding to either asterisks presented at varying screen locations or centrally presented letters. Stimulus presentation followed a fixed second-order conditional sequence. Each keypress in the experimental groups produced a contingent, key-specific tone effect. The critical variation concerned the mapping of tones to keys. In Experiment 1, keypresses in one control condition produced noncontingent tone effects, while in another control condition there were no tone effects. In Experiment 2, three different keytone mappings were compared to a control condition without tone effects. The results show that tone effects improve serial learning when they are mapped to the response keys contingently and in a highly compatible manner. The results are discussed with reference to an ideomotor mechanism of motor sequence acquisition.

Statistical learning of higher-order temporal structure from visual shape sequences

In 3 experiments, the authors investigated the ability of observers to extract the probabilities of successive shape co-occurrences during passive viewing. Participants became sensitive to several temporal-order statistics, both rapidly and with no overt task or explicit instructions. Sequences of shapes presented during familiarization were distinguished from novel sequences of familiar shapes, as well as from shape sequences that were seen during familiarization but less frequently than other shape sequences, demonstrating at least the extraction of joint probabilities of 2 consecutive shapes. When joint probabilities did not differ, another higher-order statistic (conditional probability) was automatically computed, thereby allowing participants to predict the temporal order of shapes. Results of a single-shape test documented that lower-order statistics were retained during the extraction of higher-order statistics. These results suggest that observers automatically extract multiple statistics of temporal events that are suitable for efficient associative learning of new temporal features.