Constraints on Statistical Learning Across Species

Literacy in nonspeaking autistic people

LAY ABSTRACT

Many autistic people who do not talk cannot tell other people what they know or what they are thinking. As a result, they might not be able to go to the schools they want, share feelings with friends, or get jobs they like. It might be possible to teach them to type on a computer or tablet instead of talking. But first, they would have to know how to spell. Some people do not believe that nonspeaking autistic people can learn to spell. We did a study to see if they can. We tested 31 autistic teenagers and adults who do not talk much or at all. They played a game on an iPad where they had to tap flashing letters. After they played the game, we looked at how fast they tapped the letters. They did three things that people who know how to spell would do. First, they tapped flashing letters faster when the letters spelled out sentences than when the letters made no sense. Second, they tapped letters that usually go together faster than letters that do not usually go together. This shows that they knew some spelling rules. Third, they paused before tapping the first letter of a new word. This shows that they knew where one word ended and the next word began. These results suggest that many autistic people who do not talk can learn how to spell. If they are given appropriate opportunities, they might be able to learn to communicate by typing.

The developmental cognitive mechanism of learning algebraic rules from the dual-process theory perspective

Rule learning is an important ability that enables human beings to adapt to nature and develop civilizations. There have been many discussions on the mechanism and characteristics of algebraic rule learning, but there are still controversies due to the lack of theoretical guidance. Based on the dual-process theory, this study discussed the following arguments for algebraic rule learning across human and animal studies: whether algebraic rule learning is simply Type 1 processing, whether algebraic rule learning is a domain-general ability, whether algebraic rule learning is shared by humans and animals, and whether an algebraic rule is learned consciously. Moreover, we propose that algebraic rule learning is possibly a cognitive process that combines both Type 1 and Type 2 processing. Further exploration is required to establish the essence and neural basis of algebraic rule learning.

Visual linguistic statistical learning is traceable through neural entrainment

The human brain can detect statistical regularities in the environment across a wide variety of contexts. The importance of this process is well-established not just in language acquisition but across different modalities; in addition, several neural correlates of statistical learning have been identified. A current technique for tracking the emergence of regularity learning and localizing its neural background is frequency tagging (FT). FT can detect neural entrainment not only to the frequency of stimulus presentation but also to that of a hidden structure. Auditory learning paradigms with linguistic and nonlinguistic stimuli, along with a visual paradigm using nonlinguistic stimuli, have already been tested with FT. To complete the picture, we conducted an FT experiment using written syllables as stimuli and a hidden triplet structure. Both behavioral and neural entrainment data showed evidence of structure learning. In addition, we localized two electrode clusters related to the process, which spread across the frontal and parieto-occipital areas, similar to previous findings. Accordingly, we conclude that fast-paced visual linguistic regularities can be acquired and are traceable through neural entrainment. In comparison with the literature, our findings support the view that statistical learning involves a domain-general network.

Infants' top-down perceptual modulation is specific to own-race faces

Recent studies have revealed the influence of higher-level cognitive systems in modulating perceptual processing (top-down perceptual modulation) in infancy. However, more research is needed to understand how top-down processes in infant perception contribute to early perceptual development. To this end, this study examined infants' top-down perception of own- and other-race faces to reveal whether top-down modulation is linked to the emergence of perceptual specialization. Infants first learned an association between a sound and faces, with the race of the faces manipulated between groups (own race vs. other race). We then tested infants' face perception across various levels of perceptual difficulty (manipulated by presentation duration) and indexed top-down perception by the change in perception when infants heard the sound previously associated with the face (predictive sound) versus an irrelevant sound. Infants exhibited top-down face perception for own-race faces (Experiment 1). However, we present new evidence that infants did not show evidence of top-down modulation for other-race faces (Experiment 2), suggesting an experience-based specificity of this capacity with more effective top-down modulation in familiar perceptual contexts. In addition, we ruled out the possibility that this face race effect was due to differences in infants' associative learning of the sound and faces between the two groups. This work has important implications for understanding the mechanisms supporting perceptual development and how they relate to top-down perception in infancy.

Statistical word segmentation succeeds given the minimal amount of exposure

One of the first tasks in language acquisition is word segmentation, a process to extract word forms from continuous speech streams. Statistical approaches to word segmentation have been shown to be a powerful mechanism, in which word boundaries are inferred from sequence statistics. This approach requires the learner to represent the frequency of units from syllable sequences, though accounts differ on how much statistical exposure is required. In this study, we examined the computational limit with which words can be extracted from continuous sequences. First, we discussed why two occurrences of a word in a continuous sequence is the computational lower limit for this word to be statistically defined. Next, we created short syllable sequences that contained certain words either two or four times. Learners were presented with these syllable sequences one at a time, immediately followed by a test of the novel words from these sequences. We found that, with the computationally minimal amount of two exposures, words were successfully segmented from continuous sequences. Moreover, longer syllable sequences providing four exposures to words generated more robust learning results. The implications of these results are discussed in terms of how learners segment and store the word candidates from continuous sequences.

Playing hide and seek: Contextual regularity learning develops between 3 and 5 years of age

The ability to acquire contextual regularities is fundamental in everyday life because it helps us to navigate the environment, directing our attention where relevant events are more likely to occur. Sensitivity to spatial regularities has been largely reported from infancy. Nevertheless, it is currently unclear when children can use this rapidly acquired contextual knowledge to guide their behavior. Evidence of this ability is indeed mixed in school-aged children and, to date, it has never been explored in younger children and toddlers. The current study investigated the development of contextual regularity learning in children aged 3 to 5 years. To this aim, we designed a new contextual learning paradigm in which young children were presented with recurring configurations of bushes and were asked to guess behind which bush a cartoon monkey was hiding. In a series of two experiments, we manipulated the relevance of color and visuospatial cues for the underlying task goal and tested how this affected young children's behavior. Our results bridge the gap between the infant and adult literatures, showing that sensitivity to spatial configurations persists from infancy to childhood, but it is only around the fifth year of life that children naturally start to integrate multiple cues to guide their behavior.

What's Next? Advances and Challenges in Understanding How Environmental Predictability Shapes the Development of Cognitive Control

Forming predictions about what will happen next in the world happens early in development, without instruction, and across species. Some environments support more accurate predictions. These more predictable environments also support what appear to be positive developmental trajectories, including increases in cognitive control over thoughts and actions. Such consequences of predictable environments have broad-reaching implications for society and have been explained across ecological, psychological, computational, and neural frameworks. However, many challenges remain in understanding the effects of environmental predictability, including adaptive responses to unpredictable environments and the mechanisms underlying the effects of predictable environments on developmental trajectories. Future work addressing different dimensions of predictability -- across time scales, locations, actions, people, and outcomes -- and their interactions will advance the ability to understand, predict, and support developmental trajectories.

Gamma oscillations in visual statistical learning correlate with individual behavioral differences

Statistical learning is assumed to be a fundamentally general sensory process across modalities, age, other cognitive functions, and even species. Despite this general role, behavioral testing on regularity acquisition shows great variance among individuals. The current study aimed to find neural correlates of visual statistical learning showing a correlation with behavioral results. Based on a pilot study, we conducted an EEG study where participants were exposed to associated stimulus pairs; the acquisition was tested through a familiarity test. We identified an oscillation in the gamma range (40-70 Hz, 0.5-0.75 s post-stimulus), which showed a positive correlation with the behavioral results. This change in activity was located in a left frontoparietal cluster. Based on its latency and location, this difference was identified as a late gamma activity, a correlate of model-based learning. Such learning is a summary of several top-down mechanisms that modulate the recollection of statistical relationships such as the capacity of working memory or attention. These results suggest that, during acquisition, individual behavioral variance is influenced by dominant learning processes which affect the recall of previously gained information.

Implicit auditory memory in older listeners: From encoding to 6-month retention

Any listening task, from sound recognition to sound-based communication, rests on auditory memory which is known to decline in healthy ageing. However, how this decline maps onto multiple components and stages of auditory memory remains poorly characterised. In an online unsupervised longitudinal study, we tested ageing effects on implicit auditory memory for rapid tone patterns. The test required participants (younger, aged 20-30, and older adults aged 60-70) to quickly respond to rapid regularly repeating patterns emerging from random sequences. Patterns were novel in most trials (REGn), but unbeknownst to the participants, a few distinct patterns reoccurred identically throughout the sessions (REGr). After correcting for processing speed, the response times (RT) to REGn should reflect the information held in echoic and short-term memory before detecting the pattern; long-term memory formation and retention should be reflected by the RT advantage (RTA) to REGr vs REGn which is expected to grow with exposure. Older participants were slower than younger adults in detecting REGn and exhibited a smaller RTA to REGr. Computational simulations using a model of auditory sequence memory indicated that these effects reflect age-related limitations both in early and long-term memory stages. In contrast to ageing-related accelerated forgetting of verbal material, here older adults maintained stable memory traces for REGr patterns up to 6 months after the first exposure. The results demonstrate that ageing is associated with reduced short-term memory and long-term memory formation for tone patterns, but not with forgetting, even over surprisingly long timescales.

Brain-imaging evidence for compression of binary sound sequences in human memory

According to the language-of-thought hypothesis, regular sequences are compressed in human memory using recursive loops akin to a mental program that predicts future items. We tested this theory by probing memory for 16-item sequences made of two sounds. We recorded brain activity with functional MRI and magneto-encephalography (MEG) while participants listened to a hierarchy of sequences of variable complexity, whose minimal description required transition probabilities, chunking, or nested structures. Occasional deviant sounds probed the participants' knowledge of the sequence. We predicted that task difficulty and brain activity would be proportional to the complexity derived from the minimal description length in our formal language. Furthermore, activity should increase with complexity for learned sequences, and decrease with complexity for deviants. These predictions were upheld in both fMRI and MEG, indicating that sequence predictions are highly dependent on sequence structure and become weaker and delayed as complexity increases. The proposed language recruited bilateral superior temporal, precentral, anterior intraparietal, and cerebellar cortices. These regions overlapped extensively with a localizer for mathematical calculation, and much less with spoken or written language processing. We propose that these areas collectively encode regular sequences as repetitions with variations and their recursive composition into nested structures.

Do they know or just do it? Investigating implicit and explicit sequence learning by capuchin monkeys, human adults and children

In humans, it is now established that sequential regularities can be learned implicitly (i.e. without acquiring conscious knowledge) or explicitly (with acquisition of conscious knowledge). Is this dual-processing capability also the case for non-human primates? In this study, we designed a non-verbal task to probe implicit and explicit sequence learning in capuchin monkeys (Sapajus sp., n = 12), human adults (n = 12), and children from 5 to 10 years old (n = 64). After learning spatial sequences on a touchscreen, participants' conscious access to the sequences was probed with a forced choice sequence completion test. All performed above chance level in this test, without being instructed or trained to do so. However, only human adults who reported the presence of regularities performed at ceiling level. We suggest future directions that could build on our findings to disentangle implicit and explicit learning in monkeys and children.

Cognitive Mechanisms Underlying Recursive Pattern Processing in Human Adults

The capacity to generate recursive sequences is a marker of rich, algorithmic cognition, and perhaps unique to humans. Yet, the precise processes driving recursive sequence generation remain mysterious. We investigated three potential cognitive mechanisms underlying recursive pattern processing: hierarchical reasoning, ordinal reasoning, and associative chaining. We developed a Bayesian mixture model to quantify the extent to which these three cognitive mechanisms contribute to adult humans' performance in a sequence generation task. We further tested whether recursive rule discovery depends upon relational information, either perceptual or semantic. We found that the presence of relational information facilitates hierarchical reasoning and drives the generation of recursive sequences across novel depths of center embedding. In the absence of relational information, the use of ordinal reasoning predominates. Our results suggest that hierarchical reasoning is an important cognitive mechanism underlying recursive pattern processing and can be deployed across embedding depths and relational domains.

Origins of Hierarchical Logical Reasoning

Hierarchical cognitive mechanisms underlie sophisticated behaviors, including language, music, mathematics, tool-use, and theory of mind. The origins of hierarchical logical reasoning have long been, and continue to be, an important puzzle for cognitive science. Prior approaches to hierarchical logical reasoning have often failed to distinguish between observable hierarchical behavior and unobservable hierarchical cognitive mechanisms. Furthermore, past research has been largely methodologically restricted to passive recognition tasks as compared to active generation tasks that are stronger tests of hierarchical rules. We argue that it is necessary to implement learning studies in humans, non-human species, and machines that are analyzed with formal models comparing the contribution of different cognitive mechanisms implicated in the generation of hierarchical behavior. These studies are critical to advance theories in the domains of recursion, rule-learning, symbolic reasoning, and the potentially uniquely human cognitive origins of hierarchical logical reasoning.

Novel phonotactic learning by children and infants: Generalizing syllable-position but not co-occurrence regularities

Restrictions in the sequencing of sounds (phonotactic constraints) can be represented at the level of sound co-occurrences (e.g., in baF.Pev, F and P co-occur) and at the level of the syllable (e.g., F is syllable-coda/end, P is syllable-onset/start). Can children (5-year-olds) and infants (11-month-olds) represent constraints as sound co-occurrences and/or relative to syllable positions? Participants listened to artificial languages displaying both word-medial consonant restrictions in co-occurrence pairs (e.g., FP or DZ but not FZ) and in the position of consonants within syllables (e.g., P/Z onsets and D/F codas) in words like baF.Pev and tiD.Zek. Children responded similarly to novel words with the same (e.g., FP) versus different (e.g., FZ) co-occurrence pairs, but they were more misled (i.e., responded "heard it before") by novel words with consonants in the same (e.g., onset-P) versus different (e.g., coda-P) syllable positions (Experiment 1). With the same training stimuli, infants had similar orientation times for novel words with the same versus different co-occurrence pairs, but they had longer orientation times for novel words with consonants in the same versus different syllable positions (Experiment 2). Thus, across different methods and ages, syllable-position information was more readily available for generalization than consonant co-occurrence information. The results suggest that when multiple regularities are present simultaneously, some phonotactic constraints (e.g., consonants in particular syllable positions) may be spontaneously represented and generalized by children and infants, whereas others (e.g., consonant co-occurrences) might not be available. The results contribute toward understanding how children and infants represent sound sequences.

Computational and neural mechanisms of statistical pain learning

Pain invariably changes over time. These fluctuations contain statistical regularities which, in theory, could be learned by the brain to generate expectations and control responses. We demonstrate that humans learn to extract these regularities and explicitly predict the likelihood of forthcoming pain intensities in a manner consistent with optimal Bayesian inference with dynamic update of beliefs. Healthy participants received probabilistic, volatile sequences of low and high-intensity electrical stimuli to the hand during brain fMRI. The inferred frequency of pain correlated with activity in sensorimotor cortical regions and dorsal striatum, whereas the uncertainty of these inferences was encoded in the right superior parietal cortex. Unexpected changes in stimulus frequencies drove the update of internal models by engaging premotor, prefrontal and posterior parietal regions. This study extends our understanding of sensory processing of pain to include the generation of Bayesian internal models of the temporal statistics of pain.

Emotional Faces Facilitate Statistical Learning

Detecting regularities and extracting patterns is a vital skill to organize complex information in our environments. Statistical learning, a process where we detect regularities by attending to relationships between cues in our environment, contributes to knowledge acquisition across myriad domains. However, less is known about how emotional cues-specifically facial configurations of emotion-influence statistical learning. Here, we tested two pre-registered aims to advance knowledge about emotional signals and statistical learning: (1) we examined statistical learning in the context of emotional compared to non-emotional information, and (2) we assessed how emotional congruency (i.e., whether facial stimuli conveyed the same, or different emotions) influenced regularity extraction. We demonstrated statistical learning in the context of emotional signals. Further, we showed that statistical learning occurs more efficiently in the context of emotional faces. We also established that congruent cues benefited an online measure of statistical learning, but had varied effects when statistical learning was assessed via post-exposure recognition test. The results shed light on how affective signals influence well-studied cognitive skills and address a knowledge gap about how cue congruency impacts statistical learning, including how emotional cues might guide predictions in our social world.

Supplementary Information

The online version contains supplementary material available at 10.1007/s42761-022-00130-9.

Acquiring Complex Communicative Systems: Statistical Learning of Language and Emotion

During the early postnatal years, most infants rapidly learn to understand two naturally evolved communication systems: language and emotion. While these two domains include different types of content knowledge, it is possible that similar learning processes subserve their acquisition. In this review, we compare the learnable statistical regularities in language and emotion input. We then consider how domain-general learning abilities may underly the acquisition of language and emotion, and how this process may be constrained in each domain. This comparative developmental approach can advance our understanding of how humans learn to communicate with others.

Knowledge of Statistics or Statistical Learning? Readers Prioritize the Statistics of their Native Language Over the Learning of Local Regularities

A large body of evidence suggests that people spontaneously and implicitly learn about regularities present in the visual input. Although theorized as critical for reading, this ability has been demonstrated mostly with pseudo-fonts or highly atypical artificial words. We tested whether local statistical regularities are extracted from materials that more closely resemble one’s native language. In two experiments, Italian speakers saw a set of letter strings modelled on the Italian lexicon and guessed which of these strings were words in a fictitious language and which were foils. Unknown to participants, words could be distinguished from foils based on their average bigram frequency. Surprisingly, in both experiments, we found no evidence that participants relied on this regularity. Instead, lexical decisions were guided by minimal bigram frequency, a cue rooted in participants’ native language. We discuss the implications of these findings for accounts of statistical learning and visual word processing.

Assessing the Sensitivity of EEG-Based Frequency-Tagging as a Metric for Statistical Learning

Statistical learning (SL) is hypothesized to play an important role in language development. However, the measures typically used to assess SL, particularly at the level of individual participants, are largely indirect and have low sensitivity. Recently, a neural metric based on frequency-tagging has been proposed as an alternative measure for studying SL. We tested the sensitivity of frequency-tagging measures for studying SL in individual participants in an artificial language paradigm, using non-invasive electroencephalograph (EEG) recordings of neural activity in humans. Importantly, we used carefully constructed controls to address potential acoustic confounds of the frequency-tagging approach, and compared the sensitivity of EEG-based metrics to both explicit and implicit behavioral tests of SL. Group-level results confirm that frequency-tagging can provide a robust indication of SL for an artificial language, above and beyond potential acoustic confounds. However, this metric had very low sensitivity at the level of individual participants, with significant effects found only in 30% of participants. Comparison of the neural metric to previously established behavioral measures for assessing SL showed a significant yet weak correspondence with performance on an implicit task, which was above-chance in 70% of participants, but no correspondence with the more common explicit 2-alternative forced-choice task, where performance did not exceed chance-level. Given the proposed ubiquitous nature of SL, our results highlight some of the operational and methodological challenges of obtaining robust metrics for assessing SL, as well as the potential confounds that should be taken into account when using the frequency-tagging approach in EEG studies.

Speech segmentation: New dogs, old tricks?

A new study using electroencephalography and functional magnetic resonance imaging suggests that dogs and humans may segment speech in similar ways.

Voice modulatory cues to structure across languages and species

Voice modulatory cues such as variations in fundamental frequency, duration and pauses are key factors for structuring vocal signals in human speech and vocal communication in other tetrapods. Voice modulation physiology is highly similar in humans and other tetrapods due to shared ancestry and shared functional pressures for efficient communication. This has led to similarly structured vocalizations across humans and other tetrapods. Nonetheless, in their details, structural characteristics may vary across species and languages. Because data concerning voice modulation in non-human tetrapod vocal production and especially perception are relatively scarce compared to human vocal production and perception, this review focuses on voice modulatory cues used for speech segmentation across human languages, highlighting comparative data where available. Cues that are used similarly across many languages may help indicate which cues may result from physiological or basic cognitive constraints, and which cues may be employed more flexibly and are shaped by cultural evolution. This suggests promising candidates for future investigation of cues to structure in non-human tetrapod vocalizations. This article is part of the theme issue 'Voice modulation: from origin and mechanism to social impact (Part I)'.

Domestic dogs (Canis lupus familiaris) are sensitive to the correlation between pitch and timbre in human speech

The perceived pitch of human voices is highly correlated with the fundamental frequency (f0) of the laryngeal source, which is determined largely by the length and mass of the vocal folds. The vocal folds are larger in adult males than in adult females, and men’s voices consequently have a lower pitch than women’s. The length of the supralaryngeal vocal tract (vocal-tract length; VTL) affects the resonant frequencies (formants) of speech which characterize the timbre of the voice. Men’s longer vocal tracts produce lower frequency, and less dispersed, formants than women’s shorter vocal tracts. Pitch and timbre combine to influence the perception of speaker characteristics such as size and age. Together, they can be used to categorize speaker sex with almost perfect accuracy. While it is known that domestic dogs can match a voice to a person of the same sex, there has been no investigation into whether dogs are sensitive to the correlation between pitch and timbre. We recorded a female voice giving three commands (‘Sit’, ‘Lay down’, ‘Come here’), and manipulated the recordings to lower the fundamental frequency (thus lowering pitch), increase simulated VTL (hence affecting timbre), or both (synthesized adult male voice). Dogs responded to the original adult female and synthesized adult male voices equivalently. Their tendency to obey the commands was, however, reduced when either pitch or timbre was manipulated alone. These results suggest that dogs are sensitive to both the pitch and timbre of human voices, and that they learn about the natural covariation of these perceptual attributes.

Neural processes underlying statistical learning for speech segmentation in dogs

To learn words, humans extract statistical regularities from speech. Multiple species use statistical learning also to process speech, but the neural underpinnings of speech segmentation in non-humans remain largely unknown. Here, we investigated computational and neural markers of speech segmentation in dogs, a phylogenetically distant mammal that efficiently navigates humans' social and linguistic environment. Using electroencephalography (EEG), we compared event-related responses (ERPs) for artificial words previously presented in a continuous speech stream with different distributional statistics. Results revealed an early effect (220-470 ms) of transitional probability and a late component (590-790 ms) modulated by both word frequency and transitional probability. Using fMRI, we searched for brain regions sensitive to statistical regularities in speech. Structured speech elicited lower activity in the basal ganglia, a region involved in sequence learning, and repetition enhancement in the auditory cortex. Speech segmentation in dogs, similar to that of humans, involves complex computations, engaging both domain-general and modality-specific brain areas. VIDEO ABSTRACT.

Spanish input accelerates bilingual infants' segmentation of English words

We asked whether increased exposure to iambs, two-syllable words with stress on the second syllable (e.g., guitar), by way of another language - Spanish - facilitates English learning infants' segmentation of iambs. Spanish has twice as many iambic words (40%) compared to English (20%). Using the Headturn Preference Procedure we tested bilingual Spanish and English learning 8-month-olds' ability to segment English iambs. Monolingual English learning infants succeed at this task only by 11 months. We showed that at 8 months, bilingual Spanish and English learning infants successfully segmented English iambs, and not simply the stressed syllable, unlike their monolingual English learning peers. At the same age, bilingual infants failed to segment Spanish iambs, just like their monolingual Spanish peers. These results cannot be explained by bilingual infants' reliance on transitional probability cues to segment words in both their native languages because statistical cues were comparable in the two languages. Instead, based on their accelerated development, we argue for autonomous but interdependent development of the two languages of bilingual infants.

Vocal learning as a preadaptation for the evolution of human beat perception and synchronization

The human capacity to synchronize movements to an auditory beat is central to musical behaviour and to debates over the evolution of human musicality. Have humans evolved any neural specializations for music processing, or does music rely entirely on brain circuits that evolved for other reasons? The vocal learning and rhythmic synchronization hypothesis proposes that our ability to move in time with an auditory beat in a precise, predictive and tempo-flexible manner originated in the neural circuitry for complex vocal learning. In the 15 years, since the hypothesis was proposed a variety of studies have supported it. However, one study has provided a significant challenge to the hypothesis. Furthermore, it is increasingly clear that vocal learning is not a binary trait animals have or lack, but varies more continuously across species. In the light of these developments and of recent progress in the neurobiology of beat processing and of vocal learning, the current paper revises the vocal learning hypothesis. It argues that an advanced form of vocal learning acts as a preadaptation for sporadic beat perception and synchronization (BPS), providing intrinsic rewards for predicting the temporal structure of complex acoustic sequences. It further proposes that in humans, mechanisms of gene-culture coevolution transformed this preadaptation into a genuine neural adaptation for sustained BPS. The larger significance of this proposal is that it outlines a hypothesis of cognitive gene-culture coevolution which makes testable predictions for neuroscience, cross-species studies and genetics. This article is part of the theme issue 'Synchrony and rhythm interaction: from the brain to behavioural ecology'.

Nonhuman primates learn adjacent dependencies but fail to learn nonadjacent dependencies in a statistical learning task with a salient cue

There is ample evidence that humans and nonhuman animals can learn complex statistical regularities presented within various types of input. However, humans outperform their nonhuman primate counterparts when it comes to recognizing relationships that exist across one or several intervening stimuli (nonadjacent dependencies). This is especially true when the two elements in the dependency do not share any perceptual similarity (arbitrary associations). In the present study, we investigated whether manipulating the saliency of the predictive stimulus would enhance nonadjacent dependency learning in nonhuman primates. Rhesus macaques and tufted capuchins engaged in a computerized signal detection task that included sequences that were random in nature, included an adjacent dependency, or included a nonadjacent dependency. We manipulated the saliency of the predictive stimulus, such that the predictor jittered in place on the screen in some grammar blocks, as well as the transitional probability (the likelihood of the stimulus preceding the target to accurately predict the target's appearance) from block to block. Some monkeys evidenced learning of adjacent dependencies by faster response times to targets that followed a predictive stimulus compared to targets that were not preceded by a predictor. However, consistent with the body of evidence that indicates that nonhuman animals' statistical learning mechanisms are not at the same level of sophistication as humans', there was no evidence that monkeys learned nonadjacent dependencies of arbitrary associations, even when the salient cue was present.

The Origins of Social Knowledge in Altricial Species

Human infants are altricial, born relatively helpless and dependent on parental care for an extended period of time. This protracted time to maturity is typically regarded as a necessary epiphenomenon of evolving and developing large brains. We argue that extended altriciality is itself adaptive, as a prolonged necessity for parental care allows extensive social learning to take place. Human adults possess a suite of complex social skills, such as language, empathy, morality, and theory of mind. Rather than requiring hardwired, innate knowledge of social abilities, evolution has outsourced the necessary information to parents. Critical information for species-typical development, such as species recognition, may originate from adults rather than from genes, aided by underlying perceptual biases for attending to social stimuli and capacities for statistical learning of social actions. We draw on extensive comparative findings to illustrate that, across species, altriciality functions as an adaptation for social learning from caregivers.

What happened? Do preschool children and capuchin monkeys spontaneously use visual traces to locate a reward?

The ability to infer unseen causes from evidence is argued to emerge early in development and to be uniquely human. We explored whether preschoolers and capuchin monkeys could locate a reward based on the physical traces left following a hidden event. Preschoolers and capuchin monkeys were presented with two cups covered with foil. Behind a barrier, an experimenter (E) punctured the foil coverings one at a time, revealing the cups with one cover broken after the first event and both covers broken after the second. One event involved hiding a reward, the other event was performed with a stick (order counterbalanced). Preschoolers and, with additional experience, monkeys could connect the traces to the objects used in the puncturing events to find the reward. Reversing the order of events perturbed the performance of 3-year olds and capuchins, while 4-year-old children performed above chance when the order of events was reversed from the first trial. Capuchins performed significantly better on the ripped foil task than they did on an arbitrary test in which the covers were not ripped but rather replaced with a differently patterned cover. We conclude that by 4 years of age children spontaneously reason backwards from evidence to deduce its cause.

Evidence of ordinal position encoding of sequences extracted from continuous speech

Infants' capacity to extract statistical regularities from sequential information is impressive and well documented. However, statistical learning's underlying mechanism remains mostly unknown, and its role in language acquisition is still under debate. To shed light on these issues, here we address the question of which information human subjects extract and encode after familiarisation with a continuous sequence of stimuli and its dependence on the type of segmentation cues and on the stimuli modality. Specifically, we investigate whether adults and 5-month-old infants learn the syllables' co-occurrence in the stream or generate a representation of the Words that include syllables' ordinal position. We test if subtle pauses signalling word boundaries change the encoding and, in adults, if it varies across modalities. In six behavioural experiments, we show that: (i) Adults and infants learn the streams' statistical structure. (ii) Ordinal encoding emerges in the auditory modality, and pauses enhanced it. However, (iii) ordinal encoding seems to depend on the learning stage and not on pauses marking Words' edges. Interestingly, (iv) for visual presentation of orthographic syllables, we do not find evidence of ordinal encoding in adults. Our results support the emergence, in the auditory modality, of a Word representation where its constituents are associated with an ordinal position at play already early in life, bringing new insights into speech processing and language acquisition. Additionally, we successfully use for the first time pupillometry in an infant segmentation task.

The ontogeny of early language discrimination: Beyond rhythm

Infants can discriminate languages that belong to different rhythmic classes at birth. The ability to perform within-class discrimination emerges around the fifth month of life. The cues that infants use to discriminate between prosodically close languages remain elusive. Segmental information could be a potential cue, since infants notice vowel mispronunciations of their names, show the first signs of word recognition and the first signs of perceptual narrowing for vowels around 6 months of age. If infants have in place some proto-segmental information, most likely it is about vowels. Another potential cue infants may use to discriminate languages is intonation. We tested participants using sentences in Eastern Catalan, Western Catalan and Spanish. The two Catalan dialects and Spanish belong to the same rhythmic class, they are syllable-timed, but they differ in terms of vowel distribution, given that only Eastern Catalan has vocalic reduction. The vowel distributions of Western Catalan and Spanish are more comparable. However, they differ in terms of their intonational patterns. In Experiment 1, we tested the ability of 4.5-month-old infants learning Eastern Catalan and/or Spanish to discriminate between sentences in Eastern and Western Catalan and in Experiment 2 their ability to discriminate between sentences in Western Catalan and Spanish. In order to disentangle the contribution of segmental and suprasegmental information, we also tested infants using low-pass filtered sentences in the two dialects (Experiment 3) and low-pass filtered sentences in Western Catalan and Spanish (Experiment 4). Infants discriminated the two Catalan dialects only when the stimuli were natural sentences, whereas they were able to discriminate between Western Catalan and Spanish when the stimuli were either natural or low-pass filtered sentences. The research also provides evidence of equivalent language discrimination abilities in infants growing up in monolingual and bilingual environments.

Experience with research paradigms relates to infants' direction of preference

Interpreting and predicting direction of preference in infant research has been a thorny issue for decades. Several factors have been proposed to account for familiarity versus novelty preferences, including age, length of exposure, and task complexity. The current study explores an additional dimension: experience with the experimental paradigm. We reanalyzed the data from 4 experiments on artificial grammar learning in 12-month-old infants run using the head-turn preference procedure (HPP). Participants in these studies varied substantially in their number of laboratory visits. Results show that the number of HPP studies is related to direction of preference: Infants with limited experience with the HPP setting were more likely to show familiarity preferences than infants who had amassed more experience with this paradigm. This evidence has important implications for the interpretation of experimental results: Experience with a given method or, more broadly, with the laboratory environment may affect infants' patterns of preferences.

Statistical Learning Model of the Sense of Agency

A sense of agency (SoA) is the experience of subjective awareness regarding the control of one's actions. Humans have a natural tendency to generate prediction models of the environment and adapt their models according to changes in the environment. The SoA is associated with the degree of the adaptation of the prediction models, e.g., insufficient adaptation causes low predictability and lowers the SoA over the environment. Thus, identifying the mechanisms behind the adaptation process of a prediction model related to the SoA is essential for understanding the generative process of the SoA. In the first half of the current study, we constructed a mathematical model in which the SoA represents a likelihood value for a given observation (sensory feedback) in a prediction model of the environment and in which the prediction model is updated according to the likelihood value. From our mathematical model, we theoretically derived a testable hypothesis that the prediction model is updated according to a Bayesian rule or a stochastic gradient. In the second half of our study, we focused on the experimental examination of this hypothesis. In our experiment, human subjects were repeatedly asked to observe a moving square on a computer screen and press a button after a beep sound. The button press resulted in an abrupt jump of the moving square on the screen. Experiencing the various stochastic time intervals between the action execution (button-press) and the consequent event (square jumping) caused gradual changes in the subjects' degree of their SoA. By comparing the above theoretical hypothesis with the experimental results, we concluded that the update (adaptation) rule of the prediction model based on the SoA is better described by a Bayesian update than by a stochastic gradient descent.

Different mechanisms underlie implicit visual statistical learning in honey bees and humans

The ability of developing complex internal representations of the environment is considered a crucial antecedent to the emergence of humans' higher cognitive functions. Yet it is an open question whether there is any fundamental difference in how humans and other good visual learner species naturally encode aspects of novel visual scenes. Using the same modified visual statistical learning paradigm and multielement stimuli, we investigated how human adults and honey bees (Apis mellifera) encode spontaneously, without dedicated training, various statistical properties of novel visual scenes. We found that, similarly to humans, honey bees automatically develop a complex internal representation of their visual environment that evolves with accumulation of new evidence even without a targeted reinforcement. In particular, with more experience, they shift from being sensitive to statistics of only elemental features of the scenes to relying on co-occurrence frequencies of elements while losing their sensitivity to elemental frequencies, but they never encode automatically the predictivity of elements. In contrast, humans involuntarily develop an internal representation that includes single-element and co-occurrence statistics, as well as information about the predictivity between elements. Importantly, capturing human visual learning results requires a probabilistic chunk-learning model, whereas a simple fragment-based memory-trace model that counts occurrence summary statistics is sufficient to replicate honey bees' learning behavior. Thus, humans' sophisticated encoding of sensory stimuli that provides intrinsic sensitivity to predictive information might be one of the fundamental prerequisites of developing higher cognitive abilities.

Statistical learning in domestic chicks is modulated by strain and sex

Statistical learning is a key mechanism for detecting regularities from a variety of sensory inputs. Precocial newborn domestic chicks provide an excellent model for (1) exploring unsupervised forms of statistical learning in a comparative perspective, and (2) elucidating the ecological function of statistical learning using imprinting procedures. Here we investigated the role of the sex of the chicks in modulating the direction of preference (for familiarity or novelty) in a visual statistical learning task already employed with chicks and human infants. Using both automated tracking and direct human coding, we confirmed chicks' capacity to recognize the presence of a statistically defined structure underlying a continuous stream of shapes. Using a different chicken strain than previous studies, we were also able to highlight sex differences in chicks' propensity to approach the familiar or novel sequence. This could also explain a previous failure to reveal statistical learning in chicks which sex was however not determined. Our study confirms chicks' ability to track visual statistics. The pivotal role of sex in determining familiarity or novelty preferences in this species and the interaction with the animals' strain highlight the importance to contextualize comparative research within the ecology of each species.

Social learning exploits the available auditory or visual cues

The ability to acquire a behavior can be facilitated by exposure to a conspecific demonstrator. Such social learning occurs under a range of conditions in nature. Here, we tested the idea that social learning can benefit from any available sensory cue, thereby permitting learning under different natural conditions. The ability of naïve gerbils to learn a sound discrimination task following 5 days of exposure adjacent to a demonstrator gerbil was tested in the presence or absence of visual cues. Naïve gerbils acquired the task significantly faster in either condition, as compared to controls. We also found that exposure to a demonstrator was more potent in facilitating learning, as compared to exposure to the sounds used to perform the discrimination task. Therefore, social learning was found to be flexible and equally efficient in the auditory or visual domains.

Non-adjacent Dependency Learning in Humans and Other Animals

Learning and processing natural language requires the ability to track syntactic relationships between words and phrases in a sentence, which are often separated by intervening material. These nonadjacent dependencies can be studied using artificial grammar learning paradigms and structured sequence processing tasks. These approaches have been used to demonstrate that human adults, infants and some nonhuman animals are able to detect and learn dependencies between nonadjacent elements within a sequence. However, learning nonadjacent dependencies appears to be more cognitively demanding than detecting dependencies between adjacent elements, and only occurs in certain circumstances. In this review, we discuss different types of nonadjacent dependencies in language and in artificial grammar learning experiments, and how these differences might impact learning. We summarize different types of perceptual cues that facilitate learning, by highlighting the relationship between dependent elements bringing them closer together either physically, attentionally, or perceptually. Finally, we review artificial grammar learning experiments in human adults, infants, and nonhuman animals, and discuss how similarities and differences observed across these groups can provide insights into how language is learned across development and how these language-related abilities might have evolved.

Structured Sequence Learning: Animal Abilities, Cognitive Operations, and Language Evolution

Human language is a salient example of a neurocognitive system that is specialized to process complex dependencies between sensory events distributed in time, yet how this system evolved and specialized remains unclear. Artificial Grammar Learning (AGL) studies have generated a wealth of insights into how human adults and infants process different types of sequencing dependencies of varying complexity. The AGL paradigm has also been adopted to examine the sequence processing abilities of nonhuman animals. We critically evaluate this growing literature in species ranging from mammals (primates and rats) to birds (pigeons, songbirds, and parrots) considering also cross-species comparisons. The findings are contrasted with seminal studies in human infants that motivated the work in nonhuman animals. This synopsis identifies advances in knowledge and where uncertainty remains regarding the various strategies that nonhuman animals can adopt for processing sequencing dependencies. The paucity of evidence in the few species studied to date and the need for follow-up experiments indicate that we do not yet understand the limits of animal sequence processing capacities and thereby the evolutionary pattern. This vibrant, yet still budding, field of research carries substantial promise for advancing knowledge on animal abilities, cognitive substrates, and language evolution.

Language can shape the perception of oriented objects

Seeing an object is a natural source for learning about the object's configuration. We show that language can also shape our knowledge about visual objects. We investigated sign language that enables deaf individuals to communicate through hand movements with as much expressive power as any other natural language. A few signs represent objects in a specific orientation. Sign-language users (signers) recognized visual objects faster when oriented as in the sign, and this match in orientation elicited specific brain responses in signers, as measured by event-related potentials (ERPs). Further analyses suggested that signers' responsiveness to object orientation derived from changes in the visual object representations induced by the signs. Our results also show that language facilitates discrimination between objects of the same kind (e.g., different cars), an effect never reported before with spoken languages. By focusing on sign language we could better characterize the impact of language (a uniquely human ability) on object visual processing.

Long-term implicit memory for sequential auditory patterns in humans

Memory, on multiple timescales, is critical to our ability to discover the structure of our surroundings, and efficiently interact with the environment. We combined behavioural manipulation and modelling to investigate the dynamics of memory formation for rarely reoccurring acoustic patterns. In a series of experiments, participants detected the emergence of regularly repeating patterns within rapid tone-pip sequences. Unbeknownst to them, a few patterns reoccurred every ~3 min. All sequences consisted of the same 20 frequencies and were distinguishable only by the order of tone-pips. Despite this, reoccurring patterns were associated with a rapidly growing detection-time advantage over novel patterns. This effect was implicit, robust to interference, and persisted for 7 weeks. The results implicate an interplay between short (a few seconds) and long-term (over many minutes) integration in memory formation and demonstrate the remarkable sensitivity of the human auditory system to sporadically reoccurring structure within the acoustic environment.

How does the brain learn environmental structure? Ten core principles for understanding the neurocognitive mechanisms of statistical learning

Despite a growing body of research devoted to the study of how humans encode environmental patterns, there is still no clear consensus about the nature of the neurocognitive mechanisms underpinning statistical learning nor what factors constrain or promote its emergence across individuals, species, and learning situations. Based on a review of research examining the roles of input modality and domain, input structure and complexity, attention, neuroanatomical bases, ontogeny, and phylogeny, ten core principles are proposed. Specifically, there exist two sets of neurocognitive mechanisms underlying statistical learning. First, a "suite" of associative-based, automatic, modality-specific learning mechanisms are mediated by the general principle of cortical plasticity, which results in improved processing and perceptual facilitation of encountered stimuli. Second, an attention-dependent system, mediated by the prefrontal cortex and related attentional and working memory networks, can modulate or gate learning and is necessary in order to learn nonadjacent dependencies and to integrate global patterns across time. This theoretical framework helps clarify conflicting research findings and provides the basis for future empirical and theoretical endeavors.

Network architectures supporting learnability

Human learners acquire complex interconnected networks of relational knowledge. The capacity for such learning naturally depends on two factors: the architecture (or informational structure) of the knowledge network itself and the architecture of the computational unit-the brain-that encodes and processes the information. That is, learning is reliant on integrated network architectures at two levels: the epistemic and the computational, or the conceptual and the neural. Motivated by a wish to understand conventional human knowledge, here, we discuss emerging work assessing network constraints on the learnability of relational knowledge, and theories from statistical physics that instantiate the principles of thermodynamics and information theory to offer an explanatory model for such constraints. We then highlight similarities between those constraints on the learnability of relational networks, at one level, and the physical constraints on the development of interconnected patterns in neural systems, at another level, both leading to hierarchically modular networks. To support our discussion of these similarities, we employ an operational distinction between the modeller (e.g. the human brain), the model (e.g. a single human's knowledge) and the modelled (e.g. the information present in our experiences). We then turn to a philosophical discussion of whether and how we can extend our observations to a claim regarding explanation and mechanism for knowledge acquisition. What relation between hierarchical networks, at the conceptual and neural levels, best facilitate learning? Are the architectures of optimally learnable networks a topological reflection of the architectures of comparably developed neural networks? Finally, we contribute to a unified approach to hierarchies and levels in biological networks by proposing several epistemological norms for analysing the computational brain and social epistemes, and for developing pedagogical principles conducive to curious thought. This article is part of the theme issue 'Unifying the essential concepts of biological networks: biological insights and philosophical foundations'.

Deep learning can be used to train naïve, nonprofessional observers to detect diagnostic visual patterns of certain cancers in mammograms: a proof-of-principle study

The scientific, clinical, and pedagogical significance of devising methodologies to train nonprofessional subjects to recognize diagnostic visual patterns in medical images has been broadly recognized. However, systematic approaches to doing so remain poorly established. Using mammography as an exemplar case, we use a series of experiments to demonstrate that deep learning (DL) techniques can, in principle, be used to train naïve subjects to reliably detect certain diagnostic visual patterns of cancer in medical images. In the main experiment, subjects were required to learn to detect statistical visual patterns diagnostic of cancer in mammograms using only the mammograms and feedback provided following the subjects' response. We found not only that the subjects learned to perform the task at statistically significant levels, but also that their eye movements related to image scrutiny changed in a learning-dependent fashion. Two additional, smaller exploratory experiments suggested that allowing subjects to re-examine the mammogram in light of various items of diagnostic information may help further improve DL of the diagnostic patterns. Finally, a fourth small, exploratory experiment suggested that the image information learned was similar across subjects. Together, these results prove the principle that DL methodologies can be used to train nonprofessional subjects to reliably perform those aspects of medical image perception tasks that depend on visual pattern recognition expertise.

Statistical learning for vocal sequence acquisition in a songbird

Birdsong is a learned communicative behavior that consists of discrete acoustic elements (“syllables”) that are sequenced in a controlled manner. While the learning of the acoustic structure of syllables has been extensively studied, relatively little is known about sequence learning in songbirds. Statistical learning could contribute to the acquisition of vocal sequences, and we investigated the nature and extent of sequence learning at various levels of song organization in the Bengalese finch, Lonchura striata var. domestica. We found that, under semi-natural conditions, pupils (sons) significantly reproduced the sequence statistics of their tutor’s (father’s) songs at multiple levels of organization (e.g., syllable repertoire, prevalence, and transitions). For example, the probability of syllable transitions at “branch points” (relatively complex sequences that are followed by multiple types of transitions) were significantly correlated between the songs of tutors and pupils. We confirmed the contribution of learning to sequence similarities between fathers and sons by experimentally tutoring juvenile Bengalese finches with the songs of unrelated tutors. We also discovered that the extent and fidelity of sequence similarities between tutors and pupils were significantly predicted by the prevalence of sequences in the tutor’s song and that distinct types of sequence modifications (e.g., syllable additions or deletions) followed distinct patterns. Taken together, these data provide compelling support for the role of statistical learning in vocal production learning and identify factors that could modulate the extent of vocal sequence learning.

Statistical language learning in infancy

Research to date suggests that infants exploit statistical regularities in linguistic input to identify and learn a range of linguistic structures, ranging from the sounds of language (e.g., native-language speech sounds, word boundaries in continuous speech) to aspects of grammatical structure (e.g., lexical categories like nouns and verbs, basic aspects of syntax). This article presents a brief review of the infant statistical language learning literature, and raises broader questions concerning why infants are sensitive to statistical regularities. In doing so, we consider the relationship between statistical learning, prediction, and uncertainty reduction in infancy.

Automated Study Challenges the Existence of a Foundational Statistical-Learning Ability in Newborn Chicks

What mechanisms underlie learning in newborn brains? Recently, researchers reported that newborn chicks use unsupervised statistical learning to encode the transitional probabilities (TPs) of shapes in a sequence, suggesting that TP-based statistical learning can be present in newborn brains. Using a preregistered design, we attempted to reproduce this finding with an automated method that eliminated experimenter bias and allowed more than 250 times more data to be collected per chick. With precise measurements of each chick's behavior, we were able to perform individual-level analyses and substantially reduce measurement error for the group-level analyses. We found no evidence that newborn chicks encode the TPs between sequentially presented shapes. None of the chicks showed evidence for this ability. Conversely, we obtained strong evidence that newborn chicks encode the shapes of individual objects, showing that this automated method can produce robust results. These findings challenge the claim that TP-based statistical learning is present in newborn brains.