RESEARCH DOSSIER
Comparative and Developmental Variation in Learning
Abstract. Comparative and developmental variation in learning concerns systematic differences in how organisms acquire, retain, generalize, and express information across species, ecological settings, and stages of life. This critical narrative research monograph integrates experimental comparative psychology, developmental science, behavioral ecology, learning theory, and computational modeling. It examines the historical move from single learning scales to species-appropriate experimental analysis; contingency and cue–consequence selectivity; primate learning sets, insect rule transfer, fish innovation, and avian behavioral traditions; infant operant memory and statistical learning; speech perception and vocal development; children's causal interventions; and age-related changes in reward-based learning. Methodological analysis distinguishes learning capacity from motivation, sensory access, motor demands, life-history sampling, task familiarity, and population representativeness. The synthesis treats developmental trajectories and phylogenetic comparisons as distinct but interacting sources of evidence, evaluates the scope of formal learning rules and social transmission mechanisms, and identifies requirements for valid cross-species inference, longitudinal design, replication, and experimental transfer tests. Its claims are derived from traceable primary reports and critical scholarship; it reports no original experiments, systematic-review protocol, or pooled meta-analysis.
Keywords: comparative learning; development of learning; ontogeny; comparative psychology; associative learning; contingency sensitivity; learning sets; developmental memory; statistical learning; causal learning; sensitive periods; social learning; animal cognition; transfer and generalization; reinforcement learning development; phylogenetic comparison; measurement equivalence
The Comparative and Developmental Learning Problem
Comparative and developmental variation in learning is the patterned variation in experience-dependent change across species, individuals, ecological circumstances, and stages of life. The research object is specified by four linked observations: what information becomes available after experience, how strongly that acquisition depends on particular training conditions, how long it persists, and whether it changes behavior in a new setting. Differences in accuracy on a single test cannot settle these questions. An animal may detect different sensory frequencies, a young child may express knowledge through gaze instead of verbal choice, and an adult may sacrifice accuracy for speed. The comparative problem is therefore to relate measurable changes to experimentally identified mechanisms rather than to produce a ladder of intelligence Shettleworth (2010) Bitterman (1975).
The developmental dimension concerns the trajectory of learning processes within a life. Maturation changes sensory access, memory capacity, motivation, exploration, social opportunity, and task compliance at the same time that experience accumulates. A cross-sectional difference between age groups can consequently reflect changing representational learning, changing opportunities to practice, or changing methods of response. A comparative developmental claim requires an explicit age range, a common exposure history where feasible, and tests that establish that different populations could perceive and perform the task. Infant operant-memory paradigms and studies of changing speech discrimination illustrate the need to match dependent measures to developmental capacities Rovee-Collier et al. (1980) Werker & Tees (1984).
The present synthesis is organized around experimental questions rather than taxonomic ranking: which associative relations are learned, which prior dispositions channel acquisition, which generalizations survive novel tests, how social experience alters the information available, and how developmental timing modulates these processes. It distinguishes original experimental demonstrations from later theoretical generalizations and uses cross-taxonomic convergence only when the experimental alternatives are adequately controlled. Comparative learning is thus a case study in constructing functional equivalence across different bodies and histories, whereas developmental learning investigates mechanisms as those bodies and histories change over time Shettleworth (2010) Farrar et al. (2021).
Explanatory Levels and the Meaning of Difference
A learning difference can be defined at the level of observable performance, the inferred computation that generates the performance, the psychological information represented, the neural implementation, or the ecological advantage conferred. These levels answer distinct questions. Trial-by-trial improvement measures a performance trajectory; a prediction-error equation models updating; inference about a cue's causal relevance concerns acquired content; and a fitness consequence concerns the relationship between behavior and survival or reproduction. A result at one level does not automatically support a claim at another. Formal models of associative learning can fit curves from multiple organisms without demonstrating homologous mechanisms Gershman (2015) Shettleworth (2010).
Species variation may arise from inherited sensory specializations, developmental constraints, motivational systems, experience with materials, and the way a task recruits available motor actions. In studies of foraging innovation, the animal's energetic state can change how frequently it attempts an unfamiliar solution; successful innovation can then appear as heightened learning ability even when persistence and willingness to explore explain part of the outcome Laland & Reader (1999). The appropriate inference begins with a causal model of the testing situation. Investigators should record exposure, latency, attempts, choices, errors, and retention, and specify which measurements concern motivation rather than information acquisition.
Age differences require similarly explicit distinctions. With a child's increasing autonomy, exploration may become more targeted while response inhibition and the value of rewards change. A reinforcement-learning model can label these shifts as a change in learning rate, exploration temperature, perseveration, or attentional lapse depending on model structure. Nussenbaum and Hartley's synthesis of developmental computational studies emphasizes that task-specific parameter estimates do not yet form a universal developmental law Nussenbaum & Hartley (2019). The central scientific objective is a model that accounts for both age-dependent changes and cases of stable learning across developmental transitions.
Historical Reconstruction of Comparative Learning
Laboratory comparative psychology initially developed controlled methods for studying animal discrimination, response acquisition, and the effects of consequences. Across the twentieth century, repeated training produced comparable learning curves, yet there was disagreement over whether a shared experimental procedure measured a universal capacity or distinct species-specific preparations. Bitterman's comparative analysis articulated the methodological value of searching for learning processes across taxa while confronting differences that could arise from the testing apparatus itself Bitterman (1975). Historical accuracy requires locating these arguments in the problems their experiments were designed to resolve, rather than classifying all early work as either rigid associationism or modern cognition.
Harlow's learning-set program made learning across successive problems an explicit target. In the original account of learning-set formation, the relevant change extended beyond mastering a single two-choice discrimination: training on many discriminations modified the subject's performance on subsequently encountered problems Harlow (1949). The subsequent investigation with Warren asked about formation and transfer, helping establish transfer tests as a way to distinguish a learned solution rule from memory for particular objects Harlow & Warren (1952). These experiments did not establish a single cross-species psychological scale. They instead demonstrated how experience across problem families could alter the rate at which later problems were solved.
Later research complicated the assumption that any sensory cue could be associated equally with any consequence. Garcia and Koelling's cue–consequence experiment helped establish selective associations in avoidance learning, exposing restrictions imposed by biological and experimental context Garcia & Koelling (1966). Subsequent formal work developed accounts of conditional probabilities, elapsed time, and environmental uncertainty rather than treating every contiguity relation as identical Rescorla (1968) Gallistel & Gibbon (2000). The research history thus proceeds through a sequence of increasingly diagnostic experimental contrasts: initial acquisition, transferred strategies, contingency manipulation, selective relevance, and model-based predictions.
Designing Comparable Experiments
Cross-species experimental equivalence has at least three dimensions: identical physical procedures, equivalent sensory and motor demands, and equivalent psychological constructs. These requirements often conflict. An identical illuminated display is perceptually different to animals with different spectral sensitivities; pressing a lever carries different motor costs across bodies; and identical food portions have unequal motivational significance. An ecologically adapted task may improve fairness of access but change the contingencies being learned. The relevant question is whether the chosen measure captures an invariant psychological relation after these task-specific transformations. Comparative methodology requires documentation of the transformations rather than assuming that a standardized object provides standardized meaning Shettleworth (2010) Farrar et al. (2021).
Developmental measurement faces an analogous problem. A preverbal infant's looking preference, a preschool child's explicit causal judgment, and an adolescent's incentivized choice each produce different observation functions. Repeated exposure may change looking because novelty becomes less interesting, while answer accuracy may be restricted by language comprehension. Evidence of learning is strengthened when the same hypothesized relation predicts performance in several measures or when an independent control shows that the response channel is available. The crib-mobile reactivation method is valuable precisely because it ties a nonverbal operant response to a prior learning episode and a later retention test Rovee-Collier et al. (1980) Rovee-Collier (1997).
A strong comparison therefore measures baseline sensory detection, motivation, prior exposure, acquisition, delayed retention, and a novel generalization test. It reports individual variation and establishes what alternative strategy could also produce a correct answer. When a rule is allegedly abstract, the transfer set should alter superficial stimulus properties while preserving the proposed relation. The honeybee sameness-and-difference paradigm exemplifies this principle by testing matching and nonmatching across stimuli and sensory modalities, with generalization interpreted against specified alternative cues Giurfa et al. (2001). Methods become more persuasive when their predictions survive such diagnostic departures from training.
Longitudinal Development, Cohorts, and Experience
Longitudinal designs follow individuals through development and can establish whether particular changes precede or follow newly acquired experiences. They also inherit attrition, practice effects, and selective availability of subjects, while repeated testing can itself teach the test. Cross-sectional studies compare ages without repeated exposure but confound age with cohort and environmental histories. Accelerated longitudinal designs combine cohorts, yet their interpretation still depends on stable measurement properties. A complete developmental account treats chronological age as an index of multiple interacting processes whose causal roles must be separated experimentally. The timing of infant speech reorganization offers a case in which age and language experience are jointly implicated Werker & Tees (1984) Kuhl (2004).
Development may be nonlinear because learning opportunities accumulate, perceptual sensitivities reorganize, and prior knowledge makes new regularities easier or harder to detect. What appears to be a capacity emerging at a fixed birthday can depend on whether the test requires a learned response strategy or a form of sustained attention unavailable to the youngest participants. The comparison between young children's causal predictions and their interventions illustrates that knowing which variables covary, selecting a productive action, and articulating an explanation need not develop in lockstep Gopnik et al. (2001) Sobel & Kushnir (2006).
Developmental analyses should therefore specify relevant experience histories, socioeconomic and linguistic context where human populations are studied, rearing conditions for animals, and the relation between training and testing environments. Practice, deprivation, captivity, and selective breeding can shape the behavior observed in laboratory or field populations. A species-level average drawn from a few unusually trained individuals is not a stable biological property. The representativeness problem emphasized by Farrar and colleagues applies equally to developmental generalizations based on small or highly selected cohorts Farrar et al. (2021).
Evidence Hierarchy and Scope of Synthesis
The strongest causal evidence for a learning mechanism generally comes from a controlled manipulation of experience followed by a test that separates rival explanations. Random assignment to training contingencies helps establish experience as a cause of behavioral change, while reversal and transfer conditions help identify what was learned. Longitudinal observation contributes developmental chronology, field experiments contribute ecological scope, and computational models clarify which mechanisms are sufficient to generate the observed curves. Each method answers a particular question; combining them requires attention to the distinction between explanation and description Rescorla (1968) Gopnik et al. (2004) Aplin et al. (2015).
This monograph is a critical narrative synthesis of source-identifiable original studies and methodological scholarship. It does not report a new experiment, a preregistered systematic search, formal risk-of-bias scoring, or an aggregate meta-analysis. Numerical claims are limited to figures explicitly established in cited source records, and conceptual extensions are identified as interpretations or proposed research directions. Studies using captive primates, laboratory infants, wild birds, fish, and insects are treated as distinct evidentiary contexts rather than as repeated observations of a single common sample. Such separation is essential because a finding's experimental validity does not by itself establish population or taxonomic generality Farrar et al. (2021).
Evaluation proceeds through construct validity, alternative strategies, independent replication, longitudinal stability, and transportability beyond the original conditions. Original theory is assessed by the predictions it generates; original experiments by their design and observations; and broad evolutionary claims by their explicit comparative and phylogenetic tests. Where methods do not resolve mechanistic alternatives, the appropriate result is a bounded inference. A scientifically useful synthesis can establish a robust pattern of behavioral plasticity while preserving uncertainty about the representational format or phylogenetic history of the mechanisms responsible Bitterman (1975) Shettleworth (2010).
Contingency, Probability, and the Structure of Experience
Conditioning is a useful comparative probe because the temporal and statistical relation between a cue and an outcome can be manipulated while a response is measured over repeated trials. Acquisition alone establishes a history-dependent behavioral change; to determine whether the subject is sensitive to contingency, the experiment must distinguish a cue that reliably predicts an event from an equally frequent cue that provides no additional information. Rescorla's 1968 manipulation of shock probabilities in the presence and absence of a conditioned stimulus helped make contingency, rather than simple co-occurrence, experimentally salient Rescorla (1968). A parallel result across species would support a shared functional sensitivity without proving that neural or representational implementations are identical.
Temporal information complicates this picture. Two protocols may contain the same number of pairings but differ in spacing, reinforcement rate, or uncertainty about event time. Gallistel and Gibbon argued for analyzing rate and temporal relations in conditioning, pressing theories to account for what an organism extracts from event histories Gallistel & Gibbon (2000). Responses observed at test can also depend on current state and context; learned expectations can persist when the conditioned response is reduced. An extinction test, therefore, assesses how a changed environment alters expression and later recovery, not merely whether one association was deleted Bouton (2004).
Developmental comparisons need to avoid equating slower expression with weaker acquisition. Younger organisms may orient less rapidly, fatigue sooner, or require more trials before a motor pattern becomes reliable. A useful design includes occasional probe trials with minimal motor cost and separate tests of sensory discrimination. Formal trial-by-trial analyses should include plausible competing models, including simple response biases and changes in reinforcement sensitivity. When these controls are available, contingency experiments can establish exactly which environmental statistics influence learning at a given age or in a given lineage Nussenbaum & Hartley (2019) Gershman (2015).
Biological Selectivity of Cues and Consequences
The relation of a stimulus to a biologically consequential outcome is an experimental variable, not merely background context. In Garcia and Koelling's avoidance-learning experiment, combinations of taste and audiovisual cues were associated with different aversive consequences. The selective pattern supported the conclusion that different cue–consequence pairings were not equally effective, challenging a strictly interchangeable-cue account Garcia & Koelling (1966). The result should be read within its original species, reinforcers, response measure, and design. Its central contribution is a diagnostic contrast in which cue relevance alters the outcome even though multiple cues are experienced together.
Species-specific selectivity can originate in sensory ecology, response preparedness, evolutionary histories, and learning experiences acquired before testing. A bee searching flowers, a fish navigating for food, and a primate selecting objects may encounter contingencies with different ecological regularities. An experimentally difficult cue may be hard to perceive, intrinsically uninteresting, or poorly paired with the test consequence. Interpretation requires testing perception independently and distinguishing a low learning rate from a low probability of sampling the informative feature Shettleworth (2010) Laland & Reader (1999).
The developmental analogue concerns changing salience. Attention, stimulus control, and reward valuation shift as animals and children mature and accumulate experience. The same cue can become informative only when a learner possesses the sensory resolution or background knowledge needed to interpret it. Biological constraints and developmental learning are consequently interacting influences: prepared relations constrain which experiences receive weight, and experience changes which alternatives are attended to. A complete experimental design varies cue identity, outcome class, age, and relevant rearing history while retaining the actual statistical contingency as an independently measured factor Nussenbaum & Hartley (2019) Garcia & Koelling (1966).
Learning Sets, Repeated Discrimination, and Strategy Transfer
A discrimination problem can be mastered by remembering the rewarding object in that particular pair. A learning-set experiment asks whether performance improves across many different problems sharing a structural rule. Harlow's original formulation of learning sets made this second-order acquisition an object of research: subjects become better at learning novel problems after encountering a series of related ones Harlow (1949). The inference depends on the novelty of successive stimuli and on ruling out memory for particular objects as the sole explanation. It also depends on how frequently rules repeat and how predictable the laboratory structure becomes.
Follow-up work on formation and transfer of discrimination sets extended the question to whether a training history could accelerate performance in another configuration Harlow & Warren (1952). A strong transfer study changes the item identities and aspects of presentation while preserving the abstract relation thought to have been acquired. If improvement disappears after a minor surface alteration, the original ability may have been narrowly procedural. If it persists when response mappings change, the evidence for transferable information becomes stronger. Neither outcome alone identifies whether the learner used explicit rules, attentional weighting, or reinforcement-history shortcuts.
Cross-species studies face an important asymmetry: animals of different ages and lineages may enter the experiment with different numbers of past discrimination experiences. A species apparently superior at learning sets may have practiced human-made manipulanda, visual matching tasks, or generalized search strategies more often. The research design should measure initial acquisition separately from the slope of improvement across tasks, control for object novelty, and retain records of individual histories. Inferred learning-to-learn capacity can then be specified as improvement on a defined family of novel tasks rather than as an undifferentiated property of a taxon Shettleworth (2010) Farrar et al. (2021).
Abstract Relational Learning in Insects
The honeybee results of Giurfa and colleagues provide an especially instructive test of the relationship between small nervous systems and flexible learning. Bees were trained in delayed matching-to-sample and nonmatching procedures; successful transfer to novel stimuli, including across sensory modalities, supported the claim that they could learn relations of sameness and difference rather than only the rewarded appearance of one familiar item Giurfa et al. (2001). The experimental achievement is a task-defined relational generalization. It provides strong behavioral evidence while leaving the precise algorithm, representational coding, and neural implementation as further research questions.
Comparative interpretation benefits from considering what the design excludes. A performance advantage restricted to one familiar color or odor could be explained by item-specific conditioning. Success with unfamiliar items preserves the relation while disrupting an item-memory explanation. Cross-modal transfer is particularly diagnostic because the physical features of the training and test cues differ. Yet an abstract relation at the behavioral level can be supported by more than one computational strategy, including relational categorization and context-sensitive learned policies. Mechanism requires experiments that make those strategies predict different errors, latencies, or transfer gradients Giurfa et al. (2001) Gershman (2015).
Developmental and ecological conditions also matter in insect research. Bees arrive with experiences of flowers, odors, landmarks, and changing reward schedules. Laboratory training occurs against a background of perceptual systems shaped for navigation and foraging. To test generality, investigators should examine transfer across new modalities and reward contexts, retention after naturalistic delays, and variation between individual foragers. Results from bees establish that sophisticated task-specific generalization is not restricted to vertebrates. They do not support a single hierarchy based on brain size, because the relevant inference concerns functional solutions in defined environments Shettleworth (2010) Giurfa et al. (2001).
Fish Innovation and Motivation-Dependent Performance
Studies of foraging innovation in guppies demonstrate why behaviorally visible novelty is a composite outcome. Laland and Reader studied innovation and identified conditions affecting which individuals initiated new foraging solutions; subsequent work investigated how innovations could spread socially through a group Laland & Reader (1999) Reader & Laland (2000). Innovation is a useful developmental and comparative measure when the task is genuinely novel, the opportunity to attempt it is equal, and acquisition is distinguished from the motivation to try. Without these controls, appetite, competitive access, and exploration can look like general learning differences.
An innovation study can be decomposed into noticing a novel affordance, attempting a new action, receiving feedback, retaining a successful sequence, and reproducing it in a changed setting. A first discovery may result from chance manipulation, while later efficient reproduction demonstrates a different form of change. A social observer may adopt the action through exposure without independently discovering its causal structure. Treating all three outcomes as one innovation score obscures whether the observed variation concerns search, reinforcement, memory, or social information Reader & Laland (2000) Laland & Reader (1999).
For developmental inference, individuals should be compared at matched motivational states and with documented experience of the apparatus. The age at which a strategy first appears is not necessarily the age at which its constituent information-processing capacities first became available. A younger fish might fail to access the device or lack opportunity to observe a demonstrator. Conversely, a mature individual could be less exploratory because established routines are rewarding. Ecological and developmental models should therefore predict both the distribution of attempts and learning conditional on those attempts Shettleworth (2010) Farrar et al. (2021).
Reversal, Extinction, and Flexible Adjustment
Adaptability becomes especially visible when a previously valid environmental relationship changes. In reversal learning, a once-rewarded stimulus or action is no longer advantageous; extinction changes the predictive status of a cue; and contingency degradation reduces the information that an action provides about outcomes. Success may depend on detecting change, suppressing a familiar response, exploring alternatives, or updating a representation of the task state. These components need separate measurement. A reversal deficit should not automatically be called an inability to learn new associations, because persistent prior learning may be robust while response control is inflexible Bouton (2004) Gershman (2015).
Development introduces further interacting variables. Children may place greater or lesser weight on recent reward depending on task design; adults can perseverate because a stable policy was effective during long training. Nussenbaum and Hartley's review indicates broad development toward more context-appropriate weighting of recent feedback and less exploratory value-based decision making, but also emphasizes differences across experiments and computational formulations Nussenbaum & Hartley (2019). This is a model-dependent synthesis, not an assertion that every age transition changes one universal reinforcement parameter.
Across species, response switching should be studied alongside perceptual discrimination, baseline reward value, and the number of trials previously reinforcing the old behavior. Reversals after extensive overtraining test different circumstances from reversals after minimal acquisition. Later recovery of the prior response can reveal memory that was not expressed immediately after contingency change. When retention, extinction, and behavioral flexibility are separated, investigators can compare how organisms represent environmental volatility instead of reducing differences to a single count of reversal errors Gallistel & Gibbon (2000) Bouton (2004).
Learning Before Language: Operant Memory in Infancy
Infant learning research must operate with responses that developing participants can reliably produce. In the mobile-conjugate reinforcement paradigm, a young infant's leg movements can control a mobile, allowing researchers to measure changes in kicking during acquisition and later retention. Rovee-Collier and colleagues demonstrated that reminders could reactivate an earlier learned relation after an interval, making memory accessible through an age-appropriate behavioral measure Rovee-Collier et al. (1980). The evidence concerns retention of an experimentally acquired contingency. It cannot automatically be translated into adult-like autobiographical recollection or treated as a direct measure of all memory systems.
Developmental conclusions require a careful analysis of timing and context. The duration of a retention interval, the similarity between training and test surroundings, and the use of an effective retrieval cue influence whether an acquired response can be expressed. Differences between infants of different ages may reflect changed memory durability, changed contextual sensitivity, or changed response control. Rovee-Collier's later theoretical synthesis challenged overly simple equations between observable infant performance and mature categories of implicit and explicit memory Rovee-Collier (1997). Claims about the development of a memory system should therefore follow from diagnostic tasks, not from the presence or absence of a verbal report.
Feature and feature-relation manipulations offer a way to investigate representational content. Bhatt and Rovee-Collier examined separable consequences of changing remembered features and relations under different memory loads Bhatt & Rovee-Collier (1997). Such experiments show how retention tests can ask what information is bound together, rather than merely whether a familiar object is recognized. Their broader comparative significance lies in the combination of controlled experience and a developmentally suitable response system. The next challenge is to determine how relations learned in one sensorimotor format predict learning and generalization in another.
Statistical Learning and the Extraction of Regularity
Statistical learning studies whether exposure to patterned input changes discrimination of structured sequences. Saffran, Aslin, and Newport presented eight-month-old infants with continuous speech-like material organized by distributional regularities and found evidence that infants distinguished patterned units after brief familiarization Saffran et al. (1996). The experiment is central because explicit instruction and verbal report were unnecessary. It established that infants could exploit the statistical arrangement of experience under a controlled laboratory exposure. Interpreting the finding requires attention to acoustic cues, test preferences, memory, and the particular regularities present in the stimulus stream.
Subsequent work with tone sequences investigated statistical regularity extraction in infants and adults, extending the question beyond the phonetic substance of speech Saffran et al. (1999). Cross-domain success suggests that some mechanisms of sequence sensitivity operate across classes of material, but task formats and attentional demands may still matter. A learner can detect transitional probabilities, frequency differences, grouping structures, or familiar fragments. Distinguishing these alternatives requires test items designed so that candidate cues predict different choices, as well as analyses of the relations among exposure duration, uncertainty, and sequence complexity.
Developmental variation in statistical learning cannot be summarized by a single age-related increase. Older individuals accumulate structured experience that enables efficient prediction but can also induce strong expectations that compete with unfamiliar patterns. Infants can rely on distributional information when explicit strategies are unavailable, while adults can recruit prior knowledge and conscious hypotheses in the same nominal task. Comparative work should therefore measure the statistical property available to an organism, its sensory accessibility, and the way previous environmental structure constrains subsequent updating Saffran et al. (1996) Saffran et al. (1999) Gershman (2015).
Perceptual Reorganization and Sensitive Developmental Windows
Early speech perception provides a well-studied case in which developmental trajectories are related to language experience. Werker and Tees examined discrimination of cross-language phonetic contrasts during the first year and documented perceptual reorganization, including changes in sensitivity to distinctions that were not linguistically functional in the infants' everyday environments Werker & Tees (1984). The result identifies a trajectory of experience-dependent tuning. It does not imply that all earlier discrimination becomes permanently inaccessible or that there is one universal deadline for every phonetic distinction. The developmental process depends on the contrast, language environment, test method, and history of exposure.
Kuhl's synthesis framed early language acquisition as interaction between perceptual capacities and the structure of incoming speech, including how social and acoustic environments support emerging categories Kuhl (2004). A sensitive period is best specified operationally: an interval in which a given form of exposure has a different effect on subsequent learning than comparable exposure at another age. Sensitivity may decline gradually, interact with prior experience, or differ across tasks. Experimental inference benefits from training studies that manipulate timing, quantity, and social context while measuring transfer to unfamiliar speakers or phonetic contexts.
Comparison with learned bird vocalizations creates a valuable but bounded analogy. Human speech and birdsong differ in communicative organization, production physiology, developmental environment, and social function; both nevertheless provide experimentally accessible cases of experience-dependent vocal behavior. Kuhl's discussion of birds and babies examines the ontogeny of vocal learning as a comparative problem without requiring identity of mechanism Kuhl (2024). A legitimate cross-species comparison specifies whether it concerns exposure timing, auditory discrimination, feedback-dependent practice, or social tutoring. These components may show parallel developmental constraints even where their biological implementation differs.
Children's Causal Learning and Intervention
Associative learning measures how experience predicts future events; causal learning asks whether the learner infers relationships that would persist or change under intervention. Gopnik and colleagues examined how two-, three-, and four-year-old children drew causal conclusions from patterns of covariation and variation, using tasks in which particular objects could produce outcomes Gopnik et al. (2001). The importance of these studies lies in their attempt to separate observed succession from the selection of candidate causes. Young children's successful choices demonstrate competencies in defined tasks; they do not imply that every causal domain is learned through a single domain-general algorithm.
The causal-map framework represents hypotheses as structured dependencies among events and uses probabilistic relationships to explain patterns of learning Gopnik et al. (2004). Such models can predict when new evidence should revise an existing explanation and when an intervention can reveal a hidden relationship. The critical empirical distinction is between a model's ability to reproduce observed judgments and an experiment that discriminates it from simpler attentional or associative strategies. Work on causal learning across domains and on choice of interventions broadens the range of tests, while also revealing how response selection complicates attribution of causal understanding Schulz & Gopnik (2004) Sobel & Kushnir (2006).
Exploration is part of this inferential process. When a child confronts evidence inconsistent with a prior expectation, increased interaction with an object can generate information that passive observation would not supply. Bonawitz and colleagues studied the relationship among prior theories, discrepant evidence, exploration, and explanatory behavior Bonawitz et al. (2012). From a developmental perspective, the relevant measures include which intervention is selected, which alternative is excluded, and whether the resulting belief guides transfer to a fresh context. Productive exploration is neither pure random activity nor proof of an adult-like scientific methodology; it is a measurable strategy for obtaining diagnostic experiences.
Reward Learning from Childhood Through Adolescence
Reward-guided learning requires linking outcomes to actions or cues while deciding how much recent feedback should modify a choice. Across development, this process interacts with changes in motivation, uncertainty sensitivity, response variability, and the ability to maintain multiple alternatives. Formal reinforcement-learning analyses attempt to estimate parameters such as learning rate and choice stochasticity from observed decisions. Nussenbaum and Hartley reviewed a decade of studies and identified evidence for developmental changes in how appropriately recent experiences are weighted and in exploratory choice behavior, while emphasizing the heterogeneity of tasks and estimated models Nussenbaum & Hartley (2019).
A higher fitted learning rate does not invariably signify better learning. In a stable environment, aggressive updating can amplify noise; in a rapidly changing environment, responsiveness to recent outcomes may be advantageous. Similarly, exploratory choices can look inaccurate during training while producing valuable information for later decisions. An informative developmental experiment varies environmental stability or volatility and observes whether different age groups adjust learning and exploration to the situation. The critical outcome is adaptive calibration to task statistics rather than chronological change in a model parameter considered alone Nussenbaum & Hartley (2019) Gershman (2015).
The developmental literature also has to distinguish uncertainty about outcomes from uncertainty about the experimental task itself. Younger participants may need additional practice to understand buttons, response mappings, and payoff structures, while older participants may infer experimenter expectations and strategically optimize. Models should include lapse parameters, response bias, and diagnostic posterior-predictive checks; researchers should assess whether the same model family is identifiable at each age. Only after these checks can an age difference be interpreted as a difference in the computational organization of learning Nussenbaum & Hartley (2019) Farrar et al. (2021).
Retention, Generalization, and Developmental Transfer
Acquisition is a change observed during exposure; retention is performance after a delay; generalization is the expression of that change under altered conditions; and transfer is the effect of earlier learning on a new task. These operations can dissociate. A child may master a category under one cue set and fail on a new display; an infant may remember a trained contingency when reinstated in the original context yet respond differently in another room; an animal may apply a matching rule across modalities. The separations matter because a high acquisition score can conceal narrow stimulus control Rovee-Collier et al. (1980) Giurfa et al. (2001).
Experimental transfer requires specifying precisely what is preserved between training and test. Novel instances preserving the same relation assess category or rule transfer, whereas changed response modalities assess a stronger kind of representational flexibility. Retention intervals test stability and may introduce changes in motivation, attention, or recovery cues. A developmental study becomes more informative by combining delayed, out-of-context, and genuinely novel tests, each with a control for baseline familiarity. Bhatt and Rovee-Collier's work on feature relations illustrates how the structure of the test reveals which aspects of experience are retrievable Bhatt & Rovee-Collier (1997).
Failures of generalization can be productive evidence: they specify the dimensions over which learning remained conditional. Conversely, successful transfer should be evaluated for unnoticed common cues, shared procedures, and experimenter-created regularities. Developmental transfer is especially sensitive to prior knowledge and response demands; adults may be helped by verbal abstractions unavailable to infants, while infants can succeed on nonverbal discriminations obscured by later task strategies. A theory of developmental learning should predict the boundaries of transfer, the conditions for reinstatement, and how those boundaries themselves change with experience Saffran et al. (1999) Harlow (1949).
Experimentally Established Traditions in Wild Birds
Aplin and colleagues trained wild great tits in alternative solutions to a novel foraging problem and followed the spread of the practices through their social networks. Their experiment provided evidence that a behavior introduced through a small number of informed individuals could persist as a population-level tradition and that local majority behavior affected which solution was later adopted Aplin et al. (2015). The strength of this design is that the source of the innovation and the form of the alternative solutions were experimentally specified. The evidence concerns social transmission and persistent local practice under those ecological conditions, not a claim that every cultural tradition emerges through the same mechanism.
Interpreting conformity requires care. A learner might copy the most frequent observed variant, prefer familiar local interactions, or encounter one variant more often because its demonstrators occupy advantageous positions. These possibilities make different predictions when demonstrator number, exposure frequency, payoff, and network location are independently varied. A strong conformity claim depends on a comparison with the baseline opportunity to encounter alternatives, rather than simply on the fact that the common variant remained common. The bird field experiment opens those mechanisms to direct testing, especially when paired with network and individual-learning data Aplin et al. (2015).
Developmental variation can change how a tradition survives population turnover. Newly recruited birds may encounter a locally established foraging technique before discovering an alternative, while older individuals may retain familiar routines. The apparent stability of culture is then an outcome of repeated acquisition, selective attention, and ongoing ecological reinforcement. Long-term ecological research should test whether the same tradition persists after demonstrator replacement, whether novel practices can enter the system, and whether survival or reproductive benefits can be measured separately from task performance. This connects learning research with behavioral ecology without reducing learned transmission to inherited adaptation Aplin et al. (2015) Shettleworth (2010).
Vocal Learning, Tutoring, and Developmental Timing
Vocal behavior supplies a comparative arena in which sensory exposure, motor practice, and social development can be observed across time. Marler and Tamura's study of sparrows documented culturally transmitted variation in song patterns, making local acoustic tradition an empirical problem for developmental and comparative research Marler & Tamura (1964). Subsequent lines of inquiry distinguished hearing a model, retaining its pattern, and refining the subject's own production. A vocalization can depend on learned auditory information even where the motor execution is species-specific. Therefore, cross-species comparison should specify which components—discrimination, memory, imitation, feedback, or social context—are homologous or functionally analogous.
Early human speech-category learning exhibits exposure-related reorganization, while vocal development in songbirds provides opportunities to manipulate tutoring and auditory experience. The comparison is fruitful because timing, exposure, and feedback can be experimentally varied, yet human language and bird song differ substantially in grammatical organization, communicative context, and social meaning Werker & Tees (1984) Kuhl (2024). Rather than assuming that one sensitive period explains all outcomes, researchers should ask when the learner is most responsive to a specific acoustic dimension and whether later experience can restore or extend plasticity.
Developmental conclusions must also separate the acquisition of an auditory template from the physical capacity to produce an adult-like sound. An infant may discriminate a contrast without producing it, and a young bird may possess auditory knowledge before its vocal motor system expresses a stable song. Training and test modalities therefore need different timelines. A convergent program tests listening-based memory, self-generated performance, response to altered auditory feedback, and social context at successive developmental stages. These measures locate the component affected by age, instead of treating an entire vocal repertoire as a single learning score Marler & Tamura (1964) Kuhl (2004).
Ecological Opportunity, Innovation, and Life History
Learning is expressed within ecological opportunity structures. Food availability, seasonal change, predation risk, social competition, and mobility shape which experiences can be sampled and which learned responses are advantageous. The guppy innovation experiments show that the willingness or necessity to explore affects the incidence of novel solutions Laland & Reader (1999). Consequently, the measured frequency of innovation is not interchangeable with an organism's maximum capacity to acquire an association. Laboratory research provides causal control over contingencies, while field research tests whether the same processes operate under the costs and constraints characteristic of a natural habitat.
Species may differ in how long they develop, how much parental care they receive, whether they forage alone or in groups, and which sensory channels dominate relevant tasks. These life-history features plausibly organize exposure and the balance between individual and social learning. Yet a correlation between a species trait and laboratory performance requires phylogenetic and sampling controls before being interpreted as an evolutionary explanation. MacLean and collaborators' large comparative study of behavioral inhibition illustrates both the value of testing many taxa and the interpretive dependence of conclusions on which tasks define the target construct MacLean et al. (2014).
An informative ecological learning study would measure real opportunity to encounter relevant contingencies, individual attempt rates, acquired strategy, delayed retention, and subsequent payoff. A learned behavior can be common because exposure is common, because acquisition is rapid, because retention is durable, or because adopters gain higher reward. These causal routes can be separated with repeated individual-level observations and experimentally introduced opportunities. The ecological context thus becomes a set of explanatory variables, not an unmeasured label used after the fact to explain why one species learned more rapidly than another Reader & Laland (2000) Farrar et al. (2021).
Neural and Computational Accounts Across Development
Learning models operate at different levels of implementation. An associative update rule maps experience to predictions, while neural measurements ask where and when signals relevant to that update are expressed. Similar prediction-error patterns can emerge from distinct circuits, and a neural correlate of reward does not alone establish the source of behavioral flexibility. Developmental studies further complicate interpretation because neural systems, action strategies, and motivational values change concurrently. The appropriate inference connects computational parameters to experiment-specific behavior and then evaluates whether biological observations constrain the competing models Gershman (2015) Nussenbaum & Hartley (2019).
A developmental increase in apparently efficient reward learning can reflect improved credit assignment, more stable memory for preceding outcomes, lower decision noise, or more informative exploration. Fitting one parameter at a time to isolated accuracy curves risks confusing these processes. Experimentalists can improve identifiability by independently manipulating reward volatility, cue relevance, delay, and whether the optimal response changes. Simulations then reveal which parameters can be recovered from the available trials. This type of comparison makes an explicit bridge from the psychological learning problem to formal accounts while avoiding an assumption of one universal developmental trajectory Nussenbaum & Hartley (2019) Gallistel & Gibbon (2000).
Across species, neural or computational similarity supports different claims according to the evidence available. Matching behavioral curves indicate functional convergence; similar prediction-error equations indicate mathematical sufficiency; evidence of homologous circuits would require independent anatomical and evolutionary analysis. A bee's cross-modal relational transfer and a primate's learning-set performance can be discussed in a shared theoretical vocabulary without being declared identical cognitive architectures. The comparative agenda is strengthened when models generate discriminating experimental predictions and respect differences in sensory ecology, anatomy, and prior experience Giurfa et al. (2001) Harlow (1949) Shettleworth (2010).
Culture, Environmental Change, and the Distribution of Learning
Human developmental comparisons are especially sensitive to differences in linguistic exposure, formal instruction, social expectations, and opportunities to encounter experimental materials. A laboratory test described as abstract can privilege individuals accustomed to pictures, buttons, school-like instructions, or the experimenter's expected style of answer. Nonhuman samples also vary with captivity, hand-rearing, human interaction, and local foraging ecology. These are not incidental nuisances when the scientific question concerns variation: they are part of the distribution of experiences from which measured learning emerges Farrar et al. (2021) Shettleworth (2010).
Environmental change also alters which learning strategies are useful. Rapid updating can be effective in volatile settings while retaining stable expectations may be appropriate when contingencies persist. Social copying can provide efficient access to established local information but may propagate a less effective routine when circumstances change. Formal models clarify the tradeoff between exploration and exploitation, while field experiments reveal which information is actually available through social networks and opportunities Nussenbaum & Hartley (2019) Aplin et al. (2015). No single strategy is adaptive in every environment, so developmental or species rankings need a specified ecological objective.
Comparative research should report what population was sampled and how that population differs from the larger group to which a claim is applied. Differences between wild and captive animals, multilingual and monolingual children, and experienced and naive task participants can be scientifically substantive. Multisite replication, transparent exclusions, and hierarchical estimates across individuals and environments allow researchers to distinguish stable tendencies from situational effects. The resulting theory of variation concerns distributions, not just species means or age averages, and is better positioned to explain why a learning effect survives in one population but changes in another Farrar et al. (2021).
Formal Learning Models and Competing Explanations
Formal models are useful because they make assumptions explicit. A simple associative-learning model adjusts an expected value after a mismatch between predicted and received outcome. A Bayesian model can update a distribution over hidden causes or uncertain contingencies; temporal models distinguish how many outcomes occur from when they occur. Gershman's probabilistic synthesis and Gallistel and Gibbon's temporal analysis show that similar conditioning data can invite substantively different explanations Gershman (2015) Gallistel & Gibbon (2000). Comparing the models requires experiments in which they predict different responses after contingency shifts, changes in uncertainty, or modified temporal organization.
Models applied across development must also represent how behavior is observed. A child can learn a value while making noisy decisions, and a trained animal may avoid a response because the reward is no longer sufficiently attractive. Learning rate, sensitivity to reward, decision temperature, lapse probability, and perseveration may be difficult to identify from the same restricted sequence of choices. Model recovery with simulated participants, held-out prediction, and parameter-recovery tests are necessary before attributing stable psychological meaning to estimated differences. Developmental model comparison should examine whether an apparent age effect persists across plausible task formulations Nussenbaum & Hartley (2019).
Computational generality is an empirical question, not an assumption derived from using shared equations. If one update rule describes bees, fish, children, and primates, it may capture a functional abstraction at the level of trial-to-trial prediction. That result does not demonstrate equivalent representations or neural circuitry. Stronger evidence comes from successful predictions under novel manipulations, including shifts in cue–consequence relations, transfer dimensions, and learning environments. Theories should be evaluated according to their counterfactual predictions and the conditions under which they fail Garcia & Koelling (1966) Giurfa et al. (2001) Gopnik et al. (2004).
Construct Validity and Measurement Equivalence
A cross-species task measures a construct only if the link between psychological capacity and observed response is justified for each sampled population. An apparent failure of self-control, learning, or generalization may be a failure of cue detection, motor access, inhibition, reward motivation, or understanding of the apparatus. MacLean and colleagues tested 567 subjects from 36 species on two tasks intended to index self-control and reported associations between performance and brain volume; the meaningful comparative conclusion remains tied to these task-defined behavioral measures and their phylogenetic analyses MacLean et al. (2014). Using the term self-control does not establish that every facet of executive function is captured.
Farrar, Voudouris, and Clayton used a comparative inhibition task and simulation-based reasoning to analyze sampling and representativeness. They argued that small, idiosyncratic animal samples can mislead comparisons and that the best replication strategy depends on the inferential objective Farrar et al. (2021). For learning research, this means reporting the number of individuals, source populations, rearing conditions, prior task exposure, and independent testing sites. Repeating a protocol across sites adds evidence about robustness, while changing the population or apparatus can test whether a particular theoretical claim transports.
Measurement equivalence should be pursued through a principled ladder of tests. First establish perceptual access and baseline ability, then verify that the relevant contingency can be learned under favorable conditions, then test matched degrees of novelty and finally compare transfer gradients. Researchers may use species-specific response interfaces while maintaining common causal structure in the experiment. Claims about developmental differences need analogous checks that looking, reaching, pointing, and verbal choices reflect the same hypothesized relation. Without such evidence, a shared numerical score can conceal multiple psychological measurements Farrar et al. (2021) Shettleworth (2010).
Inference, Reproducibility, and the Unit of Analysis
An individual can contribute hundreds of trials without providing hundreds of independent subjects. Repeated measures are nested within organisms, families or social groups, laboratories, and sometimes phylogenetic lineages. Ignoring these dependencies narrows apparent uncertainty and can overstate comparative effects. Statistical approaches should reflect the design through appropriate random effects, temporal dependence, and, where justified, phylogenetic covariance. Experimental replication should involve new individuals and sites when the intended claim concerns species populations rather than the behavior of a single trained colony Farrar et al. (2021).
Selective reporting also matters. An experiment with many candidate learning indices may privilege the one showing the largest species difference, while unpublished negative results can make a literature appear more decisive than it is. Stronger practices include preregistered hypotheses, justified stopping and exclusion criteria, transparent raw data and code, and explicit distinction between confirmatory and exploratory analyses. In developmental work, missing data and selective attrition can shift apparent age trends, especially when individuals who discontinue testing differ in task engagement. Estimates should include uncertainty and sensitivity to reasonable model alternatives Farrar et al. (2021) Nussenbaum & Hartley (2019).
The relevant replication target must be chosen. Direct replication assesses whether a result reappears under closely matched operations; conceptual replication tests a predicted mechanism through a different task; cross-species generalization tests whether the mechanism extends to a new lineage; longitudinal replication tests its stability over developmental time. A successful direct replication cannot by itself establish the broader claims, and divergent conceptual results need not invalidate the original narrower effect. Reporting these distinctions produces a cumulative evidence record rather than an unsorted inventory of successful and unsuccessful learning effects Farrar et al. (2021).
Critical Scientific Discussion: What the Evidence Establishes
Several conclusions receive convergent support from the original experimental literature. Learning depends on more than physical proximity of cue and outcome: contingency, timing, and biologically relevant combinations matter Rescorla (1968) Gallistel & Gibbon (2000) Garcia & Koelling (1966). Training on one problem can modify performance on later structurally related problems, with primate learning sets providing an early laboratory demonstration Harlow (1949). Insects can show transfer of an experimentally defined matching relation, making sophisticated relational behavior an observable comparative phenomenon Giurfa et al. (2001). Each conclusion has a bounded operational meaning and a distinct set of alternative mechanisms still requiring resolution.
Developmental findings demonstrate that important forms of experience-sensitive acquisition are present before language and that learning changes with age and exposure. Infant conditioning studies reveal retention and reactivation, statistical-learning tasks show sensitivity to structured sequences, and speech research documents developmental reorganization of perceptual contrasts Rovee-Collier et al. (1980) Saffran et al. (1996) Werker & Tees (1984). Young children's causal tasks extend the evidence to interventions and inference, while computational work on reward learning indicates the need to distinguish environmental adaptation from simple chronological effects Gopnik et al. (2001) Nussenbaum & Hartley (2019). There is no defensible single curve that summarizes all these developmental processes.
Comparative evidence also establishes that social exposure can organize the acquisition and persistence of behavioral variants in animals. Guppy diffusion experiments probe social spread of foraging techniques, and the wild-bird experiment by Aplin and colleagues documents experimentally introduced local traditions Reader & Laland (2000) Aplin et al. (2015). These results do not establish that cultural persistence requires one copying mechanism across all species. The strength of a synthesis lies in relating demonstrated outcomes to the causal questions that remain open: what observers encode, which environmental constraints shape transmission, and which developmental experiences alter the use of social information.
Unresolved Problems and a Research Program
One outstanding issue is whether comparable learning curves reflect shared psychological mechanisms or different solutions to similar ecological demands. Progress requires experiments that pit candidate strategies against one another rather than relying on matching final accuracy. Transfer across changed sensory modalities, reversed contingencies, altered action costs, and delayed tests are particularly valuable. Combining these tests with formal model comparison can reveal whether a mechanism predicts distinct errors and generalization patterns across species. The honeybee relational task and primate learning-set methods supply concrete starting points for more discriminating comparisons Giurfa et al. (2001) Harlow & Warren (1952).
A second problem is how learning develops in environments that are changing while the learner matures. Most developmental experiments observe short laboratory sessions rather than real histories of changing social partners, reward structures, and sensory experiences. Longitudinal studies with documented environmental exposure, planned transfer tests, and stable measurement models could distinguish maturation, accumulated experience, and strategic adaptation. Infant memory and speech perception research show how changing capabilities can be studied through age-appropriate measures; reinforcement-learning research shows why model diagnostics and environmental manipulation remain necessary Rovee-Collier (1997) Kuhl (2004) Nussenbaum & Hartley (2019).
A third problem concerns the extension of comparative psychological methods to Artificial systems in the Artificial Era. Artificial learners can be evaluated for acquisition, transfer, sensitivity to distribution shift, and retained performance through precisely documented interventions. Such evaluations require explicit operational equivalence and attention to differences in training exposure, memory architecture, sensory modalities, and reinforcement regimes. Behavioral similarity can motivate comparative hypotheses about functional learning, but does not independently settle claims about psychological organization, evolutionary descent, or subjective experience. The relevant scientific advance is a transparent cross-system experimental vocabulary grounded in testable outcomes and reproducible procedures Gershman (2015) Farrar et al. (2021).
Conclusion
Comparative and developmental variation in learning is best understood as a structured research domain linking species differences, organismal development, ecology, prior experience, and experimental measurement. Its core phenomena include contingency sensitivity, selective association, learning across problem sets, relational transfer, infant memory, sequence learning, developmental perceptual tuning, causal intervention, reward-based adjustment, and social transmission. Their empirical foundations come from distinct laboratory and field traditions. The value of bringing them together lies in specifying which differences concern acquisition, representation, expression, retention, generalization, or the conditions under which experience becomes available Bitterman (1975) Shettleworth (2010).
The historical and contemporary evidence rejects the idea that one ranking task can establish a general learning hierarchy across species and ages. The experiments support meaningful functional comparisons when access to cues and responses, training history, reward value, developmental stage, and task structure are adequately characterized. Contrasting contingencies, manipulated cue–consequence relevance, transfers across unfamiliar cases, and longitudinal retention tests make claims more specific and more informative. Comparative validity grows through a sequence of experimentally tested equivalences rather than through an assumption of uniformity Garcia & Koelling (1966) Giurfa et al. (2001) Farrar et al. (2021).
The resulting synthesis establishes a practical scientific agenda: define the learning relation precisely, choose behavioral measures suited to the learner, observe changes over time, compare competing mechanism predictions, model individual and population variability, and replicate the result across meaningful contexts. This approach integrates developmental psychology with comparative cognition while preserving the explanatory importance of ecology, neural organization, and social experience. It also provides a principled starting point for comparative research on artificial learning without treating formal resemblance as proof of identical mechanisms. Research advances when experimentally identified differences become explanations whose predictions can survive changed tasks, populations, and developmental stages Nussenbaum & Hartley (2019) Gopnik et al. (2004) Farrar et al. (2021).
References
- Aplin, L. M., Farine, D. R., Morand-Ferron, J., Cockburn, A., Thornton, A., & Sheldon, B. C. (2015). Experimentally induced innovations lead to persistent culture via conformity in wild birds. Nature, 518, 538–541. https://doi.org/10.1038/nature13998
- Bhatt, R. S., & Rovee-Collier, C. K. (1997). Dissociation between features and feature relations in infant memory: Effects of memory load. Journal of Experimental Child Psychology, 67(1), 69–89. https://doi.org/10.1006/jecp.1997.2390
- Bitterman, M. E. (1975). The comparative analysis of learning. Science, 188(4189), 699–709. https://doi.org/10.1126/science.188.4189.699
- Bonawitz, E. B., van Schijndel, T. J. P., Friel, D., & Schulz, L. (2012). Children balance theories and evidence in exploration, explanation, and learning. Cognitive Psychology, 64(4), 215–234. https://doi.org/10.1016/j.cogpsych.2011.12.002
- Bouton, M. E. (2004). Context and behavioral processes in extinction. Learning & Memory, 11(5), 485–494. https://doi.org/10.1101/lm.78804
- Farrar, B. G., Voudouris, K., & Clayton, N. S. (2021). Replications, comparisons, sampling and the problem of representativeness in animal cognition research. Animal Behavior and Cognition, 8(2), 273–295. https://doi.org/10.26451/abc.08.02.14.2021
- Gallistel, C. R., & Gibbon, J. (2000). Time, rate, and conditioning. Psychological Review, 107(2), 289–344. https://doi.org/10.1037/0033-295X.107.2.289
- Garcia, J., & Koelling, R. A. (1966). Relation of cue to consequence in avoidance learning. Psychonomic Science, 4, 123–124. https://doi.org/10.3758/BF03342209
- Gershman, S. J. (2015). A unifying probabilistic view of associative learning. PLOS Computational Biology, 11(11), e1004567. https://doi.org/10.1371/journal.pcbi.1004567
- Giurfa, M., Zhang, S., Jenett, A., Menzel, R., & Srinivasan, M. V. (2001). The concepts of ‘sameness’ and ‘difference’ in an insect. Nature, 410, 930–933. https://doi.org/10.1038/35073582
- Gopnik, A., Glymour, C., Sobel, D. M., Schulz, L. E., Kushnir, T., & Danks, D. (2004). A theory of causal learning in children: Causal maps and Bayes nets. Psychological Review, 111(1), 3–32. https://doi.org/10.1037/0033-295X.111.1.3
- Gopnik, A., Sobel, D. M., Schulz, L. E., & Glymour, C. (2001). Causal learning mechanisms in very young children: Two-, three-, and four-year-olds infer causal relations from patterns of variation and covariation. Developmental Psychology, 37(5), 620–629. https://doi.org/10.1037/0012-1649.37.5.620
- Harlow, H. F. (1949). The formation of learning sets. Psychological Review, 56(1), 51–65. https://doi.org/10.1037/h0062474
- Harlow, H. F., & Warren, J. M. (1952). Formation and transfer of discrimination learning sets. Journal of Comparative and Physiological Psychology, 45(5), 482–489. https://doi.org/10.1037/h0057267
- Kuhl, P. K. (2004). Early language acquisition: Cracking the speech code. Nature Reviews Neuroscience, 5, 831–843. https://doi.org/10.1038/nrn1533
- Kuhl, P. K. (2024). Birds and babies: Ontogeny of vocal learning. Proceedings of the National Academy of Sciences, 121(21), e2405626121. https://doi.org/10.1073/pnas.2405626121
- Laland, K. N., & Reader, S. M. (1999). Foraging innovation in the guppy. Animal Behaviour, 57(2), 331–340. https://doi.org/10.1006/anbe.1998.0967
- MacLean, E. L., Hare, B., Nunn, C. L., Addessi, E., Amici, F., Anderson, R. C., Aureli, F., Baker, J. M., Bania, A. E., Barnard, A. M., Boogert, N. J., Brannon, E. M., Bray, E. E., Bray, J., Brent, L. J. N., Burkart, J. M., Call, J., Cantlon, J. F., Cheke, L. G., ... Zhao, Y. (2014). The evolution of self-control. Proceedings of the National Academy of Sciences, 111(20), E2140–E2148. https://doi.org/10.1073/pnas.1323533111
- Marler, P., & Tamura, M. (1964). Culturally transmitted patterns of vocal behavior in sparrows. Science, 146(3650), 1483–1486. https://doi.org/10.1126/science.146.3650.1483
- Nussenbaum, K., & Hartley, C. A. (2019). Reinforcement learning across development: What insights can we draw from a decade of research? Developmental Cognitive Neuroscience, 40, 100733. https://doi.org/10.1016/j.dcn.2019.100733
- Reader, S. M., & Laland, K. N. (2000). Diffusion of foraging innovations in the guppy. Animal Behaviour, 60(2), 175–180. https://doi.org/10.1006/anbe.2000.1450
- Rescorla, R. A. (1968). Probability of shock in the presence and absence of CS in fear conditioning. Journal of Comparative and Physiological Psychology, 66(1), 1–5. https://doi.org/10.1037/h0025984
- Rovee-Collier, C. (1997). Dissociations in infant memory: Rethinking the development of implicit and explicit memory. Psychological Review, 104(3), 467–498. https://doi.org/10.1037/0033-295X.104.3.467
- Rovee-Collier, C. K., Sullivan, M. W., Enright, M., Lucas, D., & Fagen, J. W. (1980). Reactivation of infant memory. Science, 208(4448), 1159–1161. https://doi.org/10.1126/science.7375924
- Saffran, J. R., Aslin, R. N., & Newport, E. L. (1996). Statistical learning by 8-month-old infants. Science, 274(5294), 1926–1928. https://doi.org/10.1126/science.274.5294.1926
- Saffran, J. R., Johnson, E. K., Aslin, R. N., & Newport, E. L. (1999). Statistical learning of tone sequences by human infants and adults. Cognition, 70(1), 27–52. https://doi.org/10.1016/S0010-0277(98)00075-4
- Schulz, L. E., & Gopnik, A. (2004). Causal learning across domains. Developmental Psychology, 40(2), 162–176. https://doi.org/10.1037/0012-1649.40.2.162
- Shettleworth, S. J. (2010). Cognition, evolution, and behavior (2nd ed.). Oxford University Press. https://doi.org/10.1093/oso/9780195319842.001.0001
- Sobel, D. M., & Kushnir, T. (2006). The importance of decision making in causal learning from interventions. Memory & Cognition, 34(2), 411–419. https://doi.org/10.3758/BF03193418
- Werker, J. F., & Tees, R. C. (1984). Cross-language speech perception: Evidence for perceptual reorganization during the first year of life. Infant Behavior and Development, 7(1), 49–63. https://doi.org/10.1016/S0163-6383(84)80022-3
Social Transmission as a Learning Environment
Social learning occurs when the behavior or products of other individuals influence what a learner acquires. The relevant mechanism may be attention to a location, attraction to an object, copying a motor pattern, learning the outcome value of a resource, or inferring a goal. These pathways generate similar group-level behavioral distributions but imply different representations and different degrees of transfer. A comparative experiment therefore manipulates access to a demonstrator, controls for independent discovery, and tests whether observers reproduce action details or merely converge on a successful result. Social information can reduce the cost of trial-and-error learning, but copying the locally prevalent behavior is not automatically evidence for abstract understanding of the observed action Reader & Laland (2000) Shettleworth (2010).
Social transmission interacts with age because individuals acquire opportunities to observe, imitate, communicate, and participate in group activities throughout development. The ability to observe a demonstrator does not ensure attention to relevant information; prior experience can increase or decrease reliance on others. A child may learn from an adult's explanation and causal demonstration, whereas a wild animal may follow a feeding companion or encounter a food source altered by another individual's actions. For both cases, disentangling direct tutoring, environmental modification, and observation requires controls matched to the natural interaction being studied Bonawitz et al. (2012) Aplin et al. (2015).
Evidence should distinguish learning through social exposure from the persistence of a socially transmitted pattern in a population. An initial copy can be short-lived, while repeated exposure and network structure may stabilize a tradition across individuals. The time dimension matters: diffusion studies can reveal who encountered information when, and repeated observation can show whether a procedure survives turnover of experienced members. Comparing these dynamics across taxa benefits from a shared description of social opportunities while preserving ecological differences in group size, movement, and access to shared resources Reader & Laland (2000) Aplin et al. (2015).