Psychology by BogdanovaThe Encyclopedia of Psychology for the Artificial Era

RESEARCH DOSSIER

Vocabulary Acquisition and Word Learning

Abstract. Vocabulary acquisition converts encounters with linguistic forms, referents, and communicative contexts into durable, flexible lexical knowledge. This fundamental research monograph critically synthesizes the origins of word-learning research, lexical constraints, fast mapping, referent selection, cross-situational statistics, pragmatic inference, semantic extension, syntactic bootstrapping, memory, and longitudinal vocabulary development. Original experimental studies and computational models are compared with naturalistic caregiver research and cross-cultural investigations, including evidence that initial correct selection often fails to predict delayed retention. The review distinguishes receptive comprehension, expressive production, conceptual generalization, and enduring word–meaning mappings, assesses methodological limitations and alternative mechanisms, and evaluates research through 2026. It is a critical narrative synthesis, not a new experiment or registered systematic review.

Keywords: vocabulary acquisition; word learning; lexical development; referential ambiguity; fast mapping; delayed retention; cross-situational learning; joint attention; speaker intention; mutual exclusivity; statistical learning; lexical constraints; semantic generalization; syntactic bootstrapping; child-directed speech; cross-cultural psycholinguistics; receptive vocabulary; expressive vocabulary; psychological development

Introduction: The Scientific Problem of Learning Words

Vocabulary acquisition is the developmental process through which a learner establishes, maintains, differentiates, and deploys lexical representations. A word is learned when a linguistic form becomes reliably associated with a meaning or range of meanings and can participate in comprehension or production across occasions. Such learning recruits perception, attention, memory, conceptual organization, social interpretation, and experience with linguistic structure. The fundamental scientific problem concerns how a learner moves from ambiguous episodes of speech and action to stable yet revisable knowledge. Research must therefore explain which observations constrain candidate meanings, how early hypotheses survive delays, and how lexical knowledge changes when new contexts reveal a previous interpretation to be incomplete. McMurray et al. (2012)

The research object of this dossier is the acquisition of form–meaning relations and their retention, differentiation, and generalization. It includes spoken and written word learning when their common lexical mechanisms are relevant, while treating the development of an entire grammar and the adult recognition of already-known words as neighboring explanatory objects. A referent is an entity, property, event, or relation identified on a particular occasion; a word meaning can extend beyond that occasion to a category, functional relation, or abstract concept. An experimental participant who looks toward an object after hearing a novel label has selected a candidate referent, but this immediate choice is not alone evidence of a durable lexical representation. Horst & Samuelson (2008)

The investigation follows a critical narrative research design. It compares original experimental studies, longitudinal observations, cross-cultural work, computational models, and contemporary syntheses without claiming the exhaustive selection or quantitative aggregation required of a systematic review. Evidence is organized by distinctions among online inference, long-term learning, developmental generalization, and environmental variation. Particular care is given to the gap between controlled laboratory tasks and naturally occurring episodes involving multiple speakers, objects, and actions. The resulting synthesis treats word learning as a temporally extended process whose outcome must be measured independently of its initial observation, and it identifies the mechanisms for which causal evidence remains weaker than the explanatory models suggest. Tippenhauer & Saylor (2026)

Historical Foundations: From Single-Encounter Learning to Developmental Systems

The classical problem is that a heard word arrives in an environment offering more possible interpretations than a learner could consciously enumerate. Early developmental work approached this uncertainty through carefully designed episodes in which a researcher introduced a new label and subsequently examined comprehension, classification, or production. Carey and Bartlett's 1978 investigation of the color term chromium became a formative demonstration that even limited exposure can support partial initial knowledge. Their report examined twenty preschool children across encounters and later probes; performance varied across children and tasks. The historical significance of the experiment lies in separating the acquisition of some information from mastery of every semantic distinction a mature speaker would possess. Carey & Bartlett (1978)

Research subsequently explored whether lexical interpretation is guided by systematic expectations about categories and competing names. Markman and Wachtel's experiments demonstrated that preschool children could treat an unfamiliar word differently depending on whether an object already possessed a familiar label. A familiar label could shift a learner's search toward an object's substance or part, whereas a novel object invited a whole-object interpretation. The subsequent discussion of whole-object, taxonomic, and mutual-exclusivity assumptions established a central research program: whether young learners reduce ambiguity through domain-specific expectations, general inferential capacities, or strategies learned from distributions of linguistic experience. Such accounts remain useful precisely because their predictions can be tested across changing tasks and linguistic ecologies. Markman & Wachtel (1988) Markman (1990)

The field broadened through statistical learning, social-pragmatic experiments, and process-oriented memory research. Saffran, Aslin, and Newport demonstrated that eight-month-old infants can use syllabic distributional information to segment continuous artificial speech, thereby establishing one route by which candidate word forms emerge before meaning is securely attached. Baldwin tested the contribution of a speaker's visual orientation to infants' interpretation of novel labels. Later work on retention showed that immediate disambiguation and later recognition can diverge markedly. These traditions now intersect in accounts that examine the entire learning system: perceptual discovery, referential hypothesis formation, accumulation of evidence, and retrieval after time has passed. Saffran et al. (1996) Baldwin (1993) Horst & Samuelson (2008)

What Constitutes a Lexical Representation?

A lexical representation coordinates information that may be acquired at different rates. A learner must discriminate a sound pattern or orthographic form, learn the contexts in which speakers use it, associate it with relevant concepts, and discover how broad or narrow its extension should be. Word knowledge also includes relations to other words, possible syntactic behavior, and probabilistic expectations derived from experience. These components need not appear together. Recognition of a familiar noun in a picture-choice task is a different behavioral achievement from producing it spontaneously, applying it to a new exemplar, or interpreting it within a figurative expression. A rigorous theory therefore specifies which component an observation measures. McMurray et al. (2012)

A reference relation is occasion-specific, while lexical meaning is potentially portable across situations. Hearing dog while looking at one particular dog supplies evidence that a label can apply to that individual, but the ultimate category may include many visually and functionally variable animals. A successful novel-label trial may be supported by novelty preference or exclusion of known alternatives without an enduring representation of the new word. Conversely, a young child may have a partial representation that supports looking but not naming. These asymmetries are central to interpreting developmental measures and prevent vocabulary size from being treated as an exhaustive index of conceptual knowledge. Bion et al. (2013) Bergelson & Swingley (2012)

Lexical representations are revisable. A learner may initially overextend a category, distinguish objects using an unreliable visual feature, or conflate a newly encountered term with a familiar one. Repeated episodes permit updating of the association, changing the salience of features, and linking the word to multiple exemplars. Computational accounts emphasize that such adjustment can be gradual even when the learner appears to make a fast decision within a single trial. A dynamic system can therefore produce rapid selection from existing knowledge and slower improvement in enduring mappings without requiring the two phenomena to share a single acquisition threshold. This yields testable predictions about delays, interference, competition, and developmental variation. McMurray et al. (2012)

Early Comprehension and the Emergence of Word Meanings

Word comprehension can begin before fluent speech production. Bergelson and Swingley used eye movements toward pictures to test whether infants between six and nine months recognized spoken common nouns encountered in daily life. Across the tested ages, infants looked toward named pictures, demonstrating sensitivity to conventional word–referent relations earlier than some older developmental accounts anticipated. This finding concerns comprehension of selected familiar words under particular test conditions; it does not establish that infants possessed adultlike lexical meanings or that all words emerged at the same age. Its value is to move the beginning of the lexical learning question into early infancy and to connect perceptual familiarity with the semantic organization of experience. Bergelson & Swingley (2012)

The ability to locate a familiar referent is influenced by familiarity with the visual scene, the frequency and phonological distinctiveness of the word, competing pictures, and the timing of analysis. Eye-tracking can reveal emerging expectations even when an infant cannot follow a spoken instruction or overtly name an object. Conversely, the absence of a statistically reliable looking preference does not directly prove the absence of a representation: insufficient attention, weak task sensitivity, or unfamiliar test exemplars may obscure available knowledge. An assessment needs converging measures across occasions. The same logic becomes important in multilingual settings where exposure, lexical forms, and opportunities for producing a word are distributed across languages. Bergelson & Swingley (2012)

Early comprehension also changes the landscape for new learning. Once a child reliably recognizes common labels, an unfamiliar word encountered among familiar objects can be interpreted by excluding already named alternatives. Familiar-word processing thus contributes to the dynamics of novel-word selection. McMurray, Horst, and Samuelson model familiar recognition and novel referent selection as closely related competitive processes rather than separate modules. This perspective explains why improvements in speed or reliability of familiar-word recognition may alter the learning opportunities available in a subsequent naming event. Development is cumulative: earlier lexical knowledge changes attention, inference, and encoding in later episodes, while new learning reorganizes the competitive field of existing candidates. McMurray et al. (2012) Weisleder & Fernald (2013)

Vocabulary Growth: Receptive, Expressive, and Conceptual Trajectories

Developmental vocabulary is commonly assessed through receptive comprehension, expressive production, and more specialized measures of depth or semantic flexibility. These outcomes have distinct task demands. Receptive testing may ask a child to choose an indicated picture, expressive testing may require retrieving and articulating a label, and generalized meaning may require extending a name to unseen members of a category. An increase in words produced is therefore informative about lexical access and communicative behavior but cannot be translated mechanically into an equivalent increase in the number of understood concepts. A scientific account relates trajectories while retaining their separate measurement properties and the possibility of partial, context-bound knowledge. Bion et al. (2013)

Vocabulary growth reflects interacting changes in the learner and environment. More familiar words can sharpen contextual exclusions; better processing efficiency allows a child to inspect referential cues while an utterance unfolds; memory consolidation preserves distinctions established during earlier exposure; and increasing conceptual experience makes semantic generalization more informative. Longitudinal research by Weisleder and Fernald linked variation in child-directed speech within Spanish-speaking families to later lexical processing and expressive vocabulary, including a modeled mediating relation involving processing efficiency. Such longitudinal associations identify developmentally meaningful pathways, while remaining open to unmeasured family characteristics and reciprocal influences of children's own behavior on caregivers' speech. Weisleder & Fernald (2013)

A frequently described acceleration in production during toddlerhood does not imply that every individual exhibits a discrete biological vocabulary explosion. Apparent changes in slope can result from accumulating stable representations, improved retrieval, growing sample size, and the kinds of words counted. Process models reproduce variable acceleration without invoking a single universal developmental switch. Growth should therefore be examined longitudinally with repeated measures, careful control of testing familiarity, and descriptions of individual differences. When applied across languages and cultural settings, measurement must reflect what counts as a word, which speakers and contexts were sampled, and how opportunities for comprehension differ from opportunities for producing a label. McMurray et al. (2012) Bergelson et al. (2023)

Referential Ambiguity and Momentary Selection

A naming episode underdetermines a word's intended meaning. A child who hears dax while looking at a toy, the surface on which it rests, another person's action, and the toy's color confronts several plausible referents and numerous possible conceptual descriptions. The observable selection of one candidate is an online response shaped by the currently available alternatives. It cannot reveal whether the learner has encoded the phonological form accurately, inferred a stable category, or retained an association for later retrieval. This distinction is the experimental foundation for comparing early lexical hypotheses with subsequent vocabulary knowledge. McMurray et al. (2012) Horst & Samuelson (2008)

The arrangement of objects alters task difficulty. A novel label presented beside a strongly familiar object and an unfamiliar object creates a disambiguation problem with an obvious exclusion cue; two equally novel objects reduce that cue. A task that asks participants to point immediately may reward an intelligent guess even when no long-term learning occurs. Tests after a delay, with changed distractors or novel exemplars, yield additional evidence concerning retention and abstraction. The relevance of these design choices was demonstrated by Horst and Samuelson, who found excellent initial referent selection but poor later retention under their controlled procedure with twenty-four-month-olds. Horst & Samuelson (2008)

A theory of word learning must specify which operations are needed to solve which problem. Referential selection may involve novelty, familiar-word recognition, social intention, spatial attention, semantic expectations, and contextual plausibility. Learning requires modifying a representation so that later behavior is altered when original supports are absent. McMurray and colleagues formalized their interaction using real-time competition and slower associative change. The account makes the important empirical prediction that a child may select correctly for reasons that do not optimize later retention, and may accumulate partial evidence even when initial selections are visibly incorrect. The observation of a correct choice is therefore one datum in a longer explanatory sequence. McMurray et al. (2012)

Lexical Constraints, Mutual Exclusivity, and Pragmatic Inference

Lexical constraints were proposed to explain how learners restrict an otherwise enormous hypothesis space. The whole-object expectation privileges objects over arbitrary parts or properties, a taxonomic expectation guides extension to entities of the same kind, and mutual exclusivity can encourage consideration of a new referent when familiar alternatives have known names. Markman's analyses gave these expectations a coherent conceptual framework, while Markman and Wachtel's experiments provided concrete behavioral contrasts showing that the interpretation of new labels changes in the presence of familiar labels. Such behavior can arise from several sources, including sensitivity to convention, communicative expectations, and learned lexical competition. Markman (1990) Markman & Wachtel (1988)

A preference for attaching a new label to an unnamed object is context-dependent. Multiple names are routine in ordinary language: a dog can also be an animal, a pet, or a mammal, and a familiar object may have unfamiliar terms for its components, color, or material. A flexible learner must determine when a new name is likely to denote another object and when it specifies a new property of an already identified entity. Mutual exclusivity is consequently better understood as a conditional inference strategy to be tested against alternatives than as an inviolable property of the lexicon. Experiments that vary the speaker's communicative goal, known vocabulary, and stimulus arrangement can distinguish these interpretations. Markman & Wachtel (1988) Frank et al. (2009)

Computational pragmatics gives a formal interpretation to this flexibility. Frank, Goodman, and Tenenbaum modeled joint inference about speakers' referential intentions and word meanings, using annotated language data to compare predictions with alternative learning accounts. The model demonstrates how an apparent lexical constraint may emerge from reasoning about why a speaker chose a particular expression in a particular context. This does not show that every child computes the model's exact probability distributions; it identifies a tractable explanatory proposal and a source of discriminating predictions. A stronger empirical test varies the informativeness of the speaker, availability of alternative expressions, and frequency of past mappings while tracking how learners' inferences change across repeated encounters. Frank et al. (2009)

Fast Mapping, Delayed Retention, and the Learning Criterion

Fast mapping originally described the initial establishment of information about a newly encountered word from limited exposure. It became common to treat quick identification of a referent as evidence of remarkably efficient word acquisition. Carey and Bartlett's study already indicated that one encounter could generate partial, heterogeneous knowledge, while the later controlled literature revealed the importance of distinguishing performance at exposure from performance after a temporal gap. A scientifically useful definition retains the speed of initial mapping as one dimension and reserves claims about acquisition for additional tests of persistence and flexible use. The crucial experimental question becomes what has changed in the learner after the immediate demands of the teaching situation have ended. Carey & Bartlett (1978)

Horst and Samuelson conducted four experiments in which twenty-four-month-old children were asked to select referents for several new labels and were later tested after a brief delay. Children selected referents successfully yet generally retained little of the mappings under the stringent test conditions. Ostensive naming improved retention in one manipulation. Kucker and Samuelson subsequently found that prior familiarity with novel objects, rather than isolated familiarity with novel word forms, improved delayed retention in their paradigm. These results direct attention to encoding opportunities, attentional competition, and the information carried over from a labeling episode. They also explain why reporting initial accuracy without retention evidence can overstate the durability of new lexical knowledge. Horst & Samuelson (2008) Kucker & Samuelson (2012)

Longer-term tests extend the analysis beyond minutes. Vlach and Sandhofer compared retention immediately, after a week, and after a month among children and adults and documented substantial forgetting of newly mapped words across time. A decline in accuracy can reflect competition, inadequate initial encoding, or failure of retrieval, rather than erasure of every learned feature. Research designs should include tests of both familiarity with the form and the precision of its meaning, changed referents, and re-exposure when possible. Because later exposure can repair an initially weak mapping, retention is best analyzed as one component of a changing learning history rather than a permanent status obtained at the first successful choice. Vlach & Sandhofer (2012)

Cross-Situational Statistical Learning and Competing Algorithms

Across successive episodes, a word and its intended referent may repeatedly co-occur even when no individual scene identifies the pairing unambiguously. Yu and Smith demonstrated that adults exposed to sets of spoken novel words and pictured objects could infer word–object correspondences across trials without explicit within-trial guidance. Their findings establish the feasibility of extracting mappings from aggregate co-occurrence statistics under controlled ambiguity. The experiment does not by itself identify which memory algorithm children use in natural interactions, nor does it establish that learners encode every possible association equally well. It nevertheless supplies a powerful challenge to explanations that require complete certainty at every naming event. Yu & Smith (2007)

The algorithms capable of using cross-trial evidence differ in their commitments. An associative account gradually adjusts multiple candidate weights; a propose-and-verify account maintains a limited provisional hypothesis and checks it against subsequent evidence; a pragmatic account integrates expectations about what a speaker intended. Trueswell and colleagues used eye-tracking experiments with adults to distinguish proposal-based learning from a broadly distributed associative interpretation of performance. Frank and colleagues incorporated referential intention into probabilistic inference. The resulting controversy concerns representation, attentional capacity, and error recovery rather than whether cross-trial regularities can contain useful information. Good experiments examine trial-by-trial gaze, prior failures, novel pairings, and transfer to changed displays rather than interpreting endpoint accuracy alone. Trueswell et al. (2013) Frank et al. (2009)

Natural language presents structured sampling rather than independent laboratory pairings. Certain objects occur disproportionately with particular activities, speakers repeatedly name items that attract the child's attention, and the phonological forms of competing words vary in familiarity. A learner can therefore draw on simultaneous spatial, social, lexical, and sequential cues. The strength of cross-situational learning may depend on the interval between exposures and on memory interference; evidence from broader memory studies predicts that retention across sparse episodes will limit simple accumulative solutions. A promising research design manipulates variability and temporal spacing independently, while measuring which contextual details were encoded. Such work makes explicit when cross-situational evidence supports meaning learning and when it merely supports recognition of repeated co-occurrence. Yu & Smith (2007) Vlach & Sandhofer (2012) Tippenhauer & Saylor (2026)

Joint Attention, Speaker Perspective, and Communicative Intentions

Word learning frequently takes place within exchanges involving shared attention, visible actions, pointing, gaze direction, and turn-taking. A child may treat the speaker's orientation as evidence about what an utterance concerns. Baldwin's controlled experiments compared labeling congruent with the infant's focus to labeling directed toward another object and asked whether children consulted the speaker for reference. The design established that social information can influence the mapping problem rather than merely providing an attractive background for exposure. The capacity to interpret social cues changes with development and depends on which cues are available, their reliability, and the perceptual complexity of the encounter. Baldwin (1993)

Joint attention is best analyzed as a dynamic coordination process. Looking toward the same object is one observable component, while understanding the speaker's intention involves interpreting why an expression is offered and which candidate referent is relevant. Caregivers respond to children's actions and gestures, so cause and effect can flow in both directions: more engaged children may elicit different speech, and timely language may enhance sustained engagement. Research on communication quality highlights the contribution of contingent and informative exchanges rather than reducing language experience to word counts alone. These findings support studying temporally ordered interaction sequences, but observational associations do not establish the unique causal contribution of any one social behavior. Hirsh-Pasek et al. (2015)

Social and statistical explanations can be integrated without presuming that either alone is sufficient. Social orientation limits the current candidate set; distributions across many encounters can refine probabilities; lexical familiarity reduces the attraction of known objects when an unfamiliar label arrives; and memory determines which evidence is available for later use. Frank and colleagues' model formalizes how inference about intention can make cross-situational word learning more precise. The decisive experimental challenge is to separate the benefit of receiving additional informative words from the benefit of recognizing a speaker's perspective. Studies that manipulate contingent responses, ambiguous object sets, and subsequent delay can address how social coordination becomes persistent lexical knowledge. Frank et al. (2009) Hirsh-Pasek et al. (2015)

Word-Form Discovery and the Role of Statistical Regularities

Learning a meaning requires a recognizable linguistic form, yet fluent speech does not present reliable pauses around every word. Infants encounter coarticulation, variable rates of speaking, different voices, and syllable strings whose boundaries are ambiguous. Saffran, Aslin, and Newport used an artificial-language familiarization task to demonstrate that eight-month-old infants can exploit transitional probabilities between syllables to segment candidate word-like units after brief exposure. This is evidence for sensitivity to distributional structure under a tightly controlled condition. It addresses discovery of forms, a necessary contributor to lexical acquisition, while the task itself does not establish knowledge of what the segmented forms mean. Saffran et al. (1996)

Word forms often vary across speakers, pronunciations, and social circumstances. A learner must distinguish variation that leaves a lexical identity intact from contrasts that differentiate words. The encounter may be especially demanding when a name is weakly encoded, several labels share overlapping sound sequences, or the referent disappears before the utterance ends. Evidence that learners recognize recurring sound patterns provides only part of the acquisition story. The word must subsequently be integrated with experiences that constrain meaning. Early noun comprehension documented by Bergelson and Swingley indicates that form recognition and semantic mapping can already be coordinated in infancy rather than proceeding as fully isolated stages. Bergelson & Swingley (2012)

Statistical segmentation and referential statistical learning use different sources of regularity. One estimates patterns within speech, while the other estimates relations between words and candidate entities across scenes. Their interactions create a developmental loop: better segmentation stabilizes the units that can be associated with meanings, while familiar meanings can help recognize forms under noisy or incomplete input. In realistic environments, social cues and contextual expectations may identify likely names before all perceptual details are secure. A rigorous experimental program varies speech ambiguity and referential ambiguity separately, tests generalization to new voices and scenes, and distinguishes segmentation accuracy from retention of the meanings inferred for segmented sequences. Saffran et al. (1996) Yu & Smith (2007)

Semantic Categorization, Inductive Bias, and Generalization

Word learning regularly extends beyond memorizing individual object–label pairs. A noun used for one toy is commonly expected to apply to other members of a category, and such extension depends on what the learner treats as relevant. Shape, material, motion, function, and taxonomic relations may each provide useful information in different semantic domains. Markman's discussion of whole-object and taxonomic assumptions situated the problem within conceptual induction: the learner must choose which properties can be generalized while rejecting irrelevant coincidences. These expectations are hypotheses about processing, not universal guarantees that any one perceptual feature determines the extension of every new word. Markman (1990)

Smith and colleagues investigated how experience with object names changes attention to shape. In longitudinal training, young children repeatedly encountered named object categories organized by shape; training altered the tendency to generalize new object names by shape. This result is theoretically important because it suggests that a learning bias can itself be modified through a history of category naming. The evidence is grounded in specific objects and training conditions, and it does not imply an equal role for shape in learning verbs, substance terms, relational terms, or abstract words. When nouns cross perceptual and functional category boundaries, the learner must identify more flexible structures of similarity and communicative convention. Smith et al. (2002)

Generalization should be tested with unfamiliar exemplars selected to discriminate among candidate rules. If a novel word was paired with a round, red, rolling object, later trials can separate shape from color, action, and category membership. A transfer test also needs to examine how robustly the learner rejects near neighbors, not merely whether the learner accepts one conspicuously similar item. Comparing immediate extension with delayed extension reveals whether an emerging category survives memory demands. The research design should accommodate partial semantic knowledge and gradual changes in feature weighting. Such a perspective makes category learning part of the vocabulary acquisition mechanism while preserving the difference between learning a word and acquiring a comprehensive conceptual theory of its referent. Smith et al. (2002) Vlach & Sandhofer (2012)

Syntactic Evidence in Learning Word Meanings

The sentence in which a word appears is itself evidence about meaning. A novel expression placed where a count noun ordinarily occurs invites different interpretations from a form appearing as a verb in a causal sentence. Syntactic bootstrapping research examines how children use distributional and grammatical information to limit possible meanings, especially for relational words whose referents are difficult to isolate visually. This influence runs from emerging linguistic structure toward lexical interpretation, even as accumulating vocabulary helps children recognize and learn grammatical constructions. The neighboring domain of grammar acquisition investigates the broader development of syntactic productivity; the present problem concerns what structural information contributes to identifying a new word's meaning. Arunachalam & Waxman (2010)

Arunachalam and Waxman tested two-year-olds with novel verbs introduced through syntactic contexts even when corresponding referential scenes were initially absent. The experimental contrast showed that structural evidence could influence subsequent interpretation of verb meanings when relevant events were later presented. Such findings challenge a simple view in which the learner must first observe an action together with its label before extracting any semantic information. They also emphasize the experimental value of separating the time of structural exposure from the time of referent availability. The study establishes a contribution of syntax under its design conditions, rather than showing that syntax unambiguously identifies a verb's complete meaning on its own. Arunachalam & Waxman (2010)

The relative contributions of grammar and vocabulary depend on the complexity of the language and the learner's experience. In an experiment with eight- and nine-year-old children exposed to an artificial language containing nouns, verbs, and case markers, Zhang and colleagues reported learning of word-order information without corresponding evidence of vocabulary learning under their tested conditions. Adult findings using related paradigms had shown different acquisition patterns. This 2026 result cautions against treating successful extraction of regular grammatical structure as proof of lexical mappings, and it illustrates why related learning systems deserve independent dependent variables. Future work can examine whether grammatical cues constrain referential uncertainty only after particular structures or lexical items are sufficiently familiar to the learner. Zhang et al. (2026)

Embodied Attention and Multimodal Naming Events

The infant's sensory and motor activity shapes the information available during naming. Looking, reaching, touching, rotating, and holding an object can change its prominence and the number of competing referents within view. A caregiver may name what a child is already manipulating, creating aligned episodes of visual attention, manual engagement, and speech. Such interactions suggest that attention is partly organized by the learner's own action, rather than supplied exclusively through the caregiver's verbal instructions. Research on natural behavior emphasizes the need to record what children actually see and handle, because an adult-facing camera or an isolated picture-choice experiment can miss the perceptual structure available to a learner. Schroer & Yu (2024)

Object-centered attention can also help explain why particular labeling episodes are remembered. An object held close may dominate the visual field while an adult names it, reducing momentary referential competition and supporting encoding of its features. In contrast, a highly cluttered scene may distribute attention across alternatives even when the label is delivered clearly. This interpretation is compatible with work showing that object familiarization, rather than isolated word-form familiarization, can facilitate later retention. The causal claim remains conditional: the relevant factors may include distinctive visual encoding, repeated manual inspection, caregiver responsiveness, or combinations of these mechanisms. The field needs manipulations that separately alter viewing, handling, and speech contingency. Kucker & Samuelson (2012) Schroer & Yu (2024)

A multimodal account must specify the temporal relations among signals. Gesture before naming can direct attention to a prospective referent; the child's own reach may trigger a responsive name; and subsequent visual examination can strengthen a recently encountered association. Parent–child interactions therefore create temporally ordered chains that are more informative than totals of gestures or utterances. Longitudinal studies of attention and contingency suggest interdependent development, but such evidence should not be collapsed into a claim that a single kind of interaction is universally required. The strongest analysis combines synchronized behavioral coding, infant-perspective measures, and delayed word tests. It evaluates whether an intervention actually changes learned lexical representations rather than increasing immediate engagement alone. Hirsh-Pasek et al. (2015) Schroer & Yu (2024)

Memory, Interference, Forgetting, and Consolidation

Long-term lexical acquisition involves encoding a word form, linking it to an appropriate conceptual representation, and retrieving that link under variable conditions. These processes can fail independently. A learner may recognize that an utterance is new yet have an unstable memory for its phonological sequence, or may remember the form while confusing which of two novel objects it names. Competitive referential selection can also produce correct behavior without strongly encoding the chosen candidate. Such distinctions motivate experiments that test multiple aspects of memory instead of inferring enduring acquisition from a single response. The controlled retention findings of Horst and Samuelson and the process model of McMurray and colleagues make these separations central rather than peripheral. Horst & Samuelson (2008) McMurray et al. (2012)

Memory is also influenced by the spacing and contextual organization of experience. Repeated exposures distributed across situations may provide evidence that a label is stable across changing backgrounds, while exposures that are too widely separated may be difficult to integrate. Vlach and Sandhofer found substantial forgetting across longer delays, and the 2026 review by Tippenhauer and Saylor discusses how variability in contextual backgrounds and competing referents can help or hinder word learning depending on developmental and task conditions. These findings advise against a universal rule that maximum contextual variability is always beneficial. A well-designed study should independently measure the richness of exposure, retention interval, competing alternatives, and later transfer. Vlach & Sandhofer (2012) Tippenhauer & Saylor (2026)

Consolidation is a useful explanatory construct only when its empirical indicators are clear. Improved delayed performance after a period without testing may reflect stabilization, retrieval practice, subsequent exposure, or differences in attention during the test. Laboratory studies of novel labels should document opportunities for intervening learning, use comparable distractors across time, and differentiate recognition from generalization. Computational models can attribute learning to gradual associative adjustments, but such a mechanism remains a hypothesis until appropriately discriminating behavioral or neural evidence is available. At a functional level, the robust conclusion is that lexical memory develops across repeated encounters and temporal gaps, and its dynamics are necessary to explain why immediate selection and durable vocabulary knowledge often diverge. Kucker & Samuelson (2012) McMurray et al. (2012)

Caregiver Language, Lexical Diversity, and Interaction Quality

Children encounter vocabulary through talk situated in relationships and activities. The amount of speech heard can matter, but lexical diversity, grammatical complexity, conversational contingency, and opportunities to connect words to meaningful experience also contribute. Rowe's longitudinal study of fifty parent–child dyads measured aspects of caregiver input at multiple ages and related them to later vocabulary. Her analyses found age-sensitive associations: vocabulary diversity and sophistication were informative in the toddler period, while decontextualized language, including narratives beyond the immediate scene, contributed additional information for preschool development. The evidence supports a developmental account in which useful input changes with the child's existing lexical and communicative capacities. Rowe (2012)

Child-directed speech also interacts with real-time language processing. Weisleder and Fernald examined variation in language addressed to infants in Spanish-speaking families, linking richer child-directed input to processing efficiency and later expressive vocabulary. Their mediation analysis offered a plausible account of how exposure may facilitate subsequent learning, but mediation in longitudinal observational data is not equivalent to experimental identification of every causal link. A parent may provide more elaborate language in response to a child's increasing abilities, and common family or environmental factors may affect both. The scientific interpretation should therefore recognize a reciprocal developmental system while preserving evidence that input characteristics predict meaningful individual differences in vocabulary outcomes. Weisleder & Fernald (2013)

Interaction quality can be operationalized through responsiveness, shared focus, elaborated utterances, and the integration of language with the child's activity. Hirsh-Pasek and colleagues analyzed early communication quality in relation to later language outcomes and highlighted the informative role of interactional characteristics. This approach enlarges the unit of analysis from isolated words per hour to the contingencies that give words interpretable functions in an exchange. Nevertheless, quality ratings require validated coding and comparable contexts; a high score on one operational scale cannot automatically represent a universal optimum across caregiving traditions. Future experimental work needs to determine which specific contingent behaviors improve lasting vocabulary learning, for which learners, and under what combinations of word frequency and semantic difficulty. Hirsh-Pasek et al. (2015)

Socioeconomic Evidence, Cultural Ecology, and Sampling Bias

Variation in economic resources, educational opportunities, housing stability, and caregiver time can affect children's language environments through multiple pathways. Hoff's study compared the speech and productive vocabulary growth of children from different socioeconomic groups and reported that measured properties of maternal speech statistically accounted for the between-group difference observed in that sample. The finding is important for identifying proximal mechanisms, but it does not license the inference that every household within an economic category provides a similar learning environment. Sociodemographic variables are composite indicators whose associations depend on the populations sampled, the languages represented, and how interaction is recorded. Treating group averages as descriptions of individual families would misinterpret the research. Hoff (2003)

The widely circulated numerical story of a thirty-million-word exposure gap has received substantive methodological criticism. Sperry, Sperry, and Miller examined naturalistic data from five United States communities and found patterns inconsistent with the claim when the linguistic environment included broader forms of interaction and bystander talk. Their analysis draws attention to which speakers count, how much conversational activity is captured, and whether a child is assumed to learn only from language explicitly addressed to that child. The critique does not show that all input conditions are interchangeable, but it undermines the presentation of a single highly generalized numerical gap as a cross-cultural law. Research should report operational definitions of speech input alongside results. Sperry et al. (2019)

Cross-cultural findings further revise the interpretation of child-directed input. Research with Tseltal-speaking infants reported knowledge of common nouns and greeting terms in a setting where children are comparatively rarely addressed directly, showing that overheard speech and local interactional routines can carry learnable lexical information. Large-scale recordings across six continents also document broad variation in naturally occurring language experiences. These studies concern different outcome variables and sampling designs, so their results should be compared at the level of defined measures rather than compressed into a simple universal statement about word counts. The scientific task is to identify multiple pathways by which learners can establish stable lexical relations in distinctive social ecologies. Foushee & Srinivasan (2024) Bergelson et al. (2023)

Individual Differences and Developmental Pathways

Children vary in how quickly they recognize familiar words, how consistently they attend to social cues, how well they retain unfamiliar labels, and how flexibly they generalize categories. These differences can influence each other over time. A child who processes familiar labels efficiently can identify an already named object before the speaker's utterance ends, potentially increasing attention to other cues in a scene. A child with weaker retention may make correct immediate choices repeatedly without accumulating a comparably large durable vocabulary. Longitudinal observation is therefore necessary to distinguish transient task variability from stable developmental differences and to examine whether processing efficiency precedes or follows changes in lexical knowledge. Weisleder & Fernald (2013) Bion et al. (2013)

Bion, Borovsky, and Fernald tested disambiguation and learning among eighteen-, twenty-four-, and thirty-month-old children, linking online looking behavior to retention and existing vocabulary. The results indicate that early disambiguation is developmentally variable and related to other lexical abilities. Yet associations among processing, disambiguation, and vocabulary can arise from reciprocal development and common determinants. They do not prove that training one isolated attention component would necessarily produce lasting gains in another. An explanatory model should specify whether differences stem from familiarity with existing labels, the number of candidate referents considered, phonological encoding, semantic category knowledge, or memory for recent encounters. Bion et al. (2013)

Research on individual differences also requires care in labeling developmental variation. Children grow up with different languages, multimodal access, sensory profiles, and histories of interaction. A slower score on a single task can reflect unfamiliar stimuli or test demands instead of a generalized learning impairment. Assessments should compare receptive and expressive measures, track repeated encounters, and use culturally and linguistically appropriate reference groups. Novel-word experiments can isolate a candidate mechanism, but their limited ecological range should be acknowledged. When work touches developmental language disorder or educational need, mechanistic findings must be separated from formal diagnostic criteria and clinical intervention evidence; the present dossier remains an investigation of lexical learning processes rather than an individual diagnostic guide. Bergelson et al. (2023) Tippenhauer & Saylor (2026)

Multilingual Exposure and Lexical Learning Across Languages

Multilingual acquisition places several systems of lexical conventions into contact within the same learner. A child may know a referent's label in one language and encounter a new label for that same referent in another. Accordingly, a strict one-object–one-word assumption would be poorly suited to ordinary multilingual experience. A learner can still use pragmatic expectations and learned conventions to resolve local ambiguity, but the interpretation of novelty depends on the language currently being used and the distribution of prior exposure. Cross-language vocabulary comparison should consider conceptual equivalents, language-specific meanings, and differences in contexts where each word is used, rather than judging knowledge from a single-language count alone. McMurray et al. (2012)

Cross-cultural recordings from a large international sample establish that families differ in multilingual exposure and everyday speaking patterns, though the study's principal quantified outcomes concern recorded speech and vocalizations rather than definitive lexical test scores. These data support broadening sampling frames for future lexical research and avoiding the assumption that predominantly monolingual middle-class English-speaking households represent the typical language environment. They also indicate a practical methodological problem: automatic recordings may detect speech activity while leaving unresolved which word meanings children understand. Multilingual semantic learning therefore calls for direct comprehension and generalization measures combined with careful estimates of language-specific experience and the interactional context in which each lexical form appears. Bergelson et al. (2023)

Potential cross-language transfer should be analyzed separately for form recognition, meaning, and task strategy. Shared concepts can facilitate learning when the child already understands a referent, whereas competing sound forms or ambiguous translations can increase retrieval demands. Cross-situational experiments may test whether learners consolidate evidence across two labels or compartmentalize it by speech context. An intervention designed for one language may not generalize to another if its semantic categories or conversational conventions differ. Rather than assigning one blanket advantage or disadvantage to multilingualism, a stronger research program examines how lexical experiences are distributed, which representations are shared, and which experimental conditions permit inference across languages. The same logic applies to second-language learners who already possess extensive conceptual knowledge. Frank et al. (2009) Sperry et al. (2019)

Vocabulary Learning Beyond Infancy: Instruction, Reading, and Context

Vocabulary acquisition continues across childhood and adulthood as words become less tied to immediately observable objects. Learners encounter abstract nouns, relational verbs, academic terminology, polysemy, and technical concepts whose meanings are distributed across extended discourse. Here, narratives, comparison across texts, definitions, and explicit examples provide sources of evidence unavailable in a single visible naming event. Rowe's findings on decontextualized caregiver language foreshadow the importance of discourse for vocabulary growth: explanatory talk can connect current expressions to absent events and conceptual relations. The learner's task increasingly includes integrating new words with established semantic networks, not merely pairing labels with objects. Rowe (2012)

Contextual diversity can support abstraction when exposure across episodes reveals a stable meaning while varying incidental backgrounds. It can also hinder learning when memory capacity is limited or competing referents multiply faster than the learner can resolve them. Tippenhauer and Saylor's 2026 review emphasizes developmental constraints, the nature of context differences, and the status of the learner's conceptual knowledge. This synthesis counsels experimental designs that distinguish changes in background scenery from changes in candidate referents, and that record whether performance is immediate recognition, delayed recall, or transfer. Instruction can use this distinction by providing initially informative contexts and subsequent varied opportunities for retrieval rather than assuming that either constant repetition or maximal diversity is best in all circumstances. Tippenhauer & Saylor (2026)

Educational word learning should be evaluated through sustained knowledge, accurate use, and generalization to new discourse. An immediate quiz after a definition measures a narrower achievement than the ability to understand a word days later in a previously unseen passage. Practical designs can combine direct explanation, meaningful contextual examples, and spaced retrieval, but the strength of a specific instructional recommendation depends on studies using the relevant age group, language, and outcome. The present experimental corpus provides strong reasons to distinguish selection and retention, while it does not establish a single optimal teaching schedule for every learner. A responsible bridge from developmental psychology to education keeps the intervention claim proportional to the evidence and treats conceptual understanding as a substantive target of vocabulary instruction. Vlach & Sandhofer (2012) Tippenhauer & Saylor (2026)

Experimental Paradigms and the Measurement of Lexical Acquisition

Novel-word experiments can separately assess referent selection, retention, extension, and recognition of word forms. A referent-selection paradigm presents multiple alternatives and asks the participant to identify the target; a delayed-retention paradigm removes immediate contextual support and asks whether the mapping can be recovered after time. Extension trials compare novel members and nonmembers of a proposed category. Eye-tracking records the timing and direction of visual attention, while pointing and naming tasks place different demands on action and speech production. Researchers should report the alternatives offered, familiarity of distractors, number and timing of exposures, exclusion opportunities, and whether the test used the original exemplar or a new instance. Horst & Samuelson (2008) Bion et al. (2013)

Experimental control permits causal statements about manipulated task features when randomization and measurement are adequate, yet it can change the learning problem. An array of three pictures shown on a screen is perceptually simpler than a family interaction involving overlapping activities, multiple speakers, and objects outside the visual field. Controlled studies can reveal a mechanism, but broad claims about everyday learning require replication under naturally distributed input. Comparing designs is particularly important when one study defines learning as an immediate look and another uses a delayed test with changed objects. The apparent disagreement may arise from distinct dependent variables rather than incompatible theories. Yu & Smith (2007) Schroer & Yu (2024)

Longitudinal caregiver recordings and observational methods offer ecological coverage while introducing other inferential challenges. Sampling windows may miss crucial interactions, automatic diarization may misidentify speakers, and estimates of word counts depend on what qualifies as adult-directed or overheard talk. The large global recording study by Bergelson and colleagues illustrates the scope of this approach, while the community observations reported by Sperry and colleagues highlight how coding decisions can alter exposure comparisons. The most informative strategy is triangulation: observational characterization of actual environments, experimentally controlled tests of proposed mechanisms, and repeated assessment of how children retain and generalize learned words. Each method answers a different scientific question. Bergelson et al. (2023) Sperry et al. (2019)

Computational Accounts and Explanatory Identifiability

Formal models transform broad descriptions of word learning into commitments about information representation, updating, and decision-making. An associative model may assign weights to word–referent pairs and modify them with co-occurrence; an inference model may estimate a speaker's intended referent; a hypothesis-testing model may track a single candidate and revise after contradiction. Models can generate overlapping predictions in easy tests, so a convincing comparison requires tasks designed to distinguish them. McMurray, Horst, and Samuelson demonstrate how online competition and slower associative learning can jointly generate rapid decisions and gradual changes in lexical knowledge, integrating several apparently separate developmental effects within a process framework. McMurray et al. (2012)

Other accounts give more weight to pragmatic intention or restricted hypothesis memory. Frank, Goodman, and Tenenbaum modeled the joint inference of word meaning and communicative goals from corpus data. Trueswell and colleagues used trial-resolved eye movements to evaluate how adults learn under sustained ambiguity and to distinguish propose-and-verify dynamics from broader cross-trial associative tracking. These approaches differ in whether uncertainty is maintained over multiple possibilities, which evidence guides reanalysis, and how much information can be retained from one episode to the next. The scientific comparison should examine predictive performance on held-out conditions, sensitivity to competitor structure, and generalization under changed referential contexts, while acknowledging that successful simulation does not establish psychological uniqueness. Frank et al. (2009) Trueswell et al. (2013)

Computational fit can remain underdetermined when a dataset contains only final accuracy. The same proportion of correct responses may result from fast guessing plus weak memory, from stable associative mapping, or from strong but inflexible inference. Trial-level gaze, response latency, confidence proxies, delayed recognition, and transfer to unseen categories provide more discriminating observations. The model should also specify its learning history: what the learner encountered before the experiment and which lexical representations were already stable. Without such controls, apparently strong model performance may depend on implicit knowledge inserted by its designers. Contemporary artificial word-learning models offer useful comparative mechanisms, but their simulation targets, training distributions, and evaluation measures must be stated before comparing their achievements with human developmental evidence. McMurray et al. (2012) Yu & Smith (2007)

Recent Evidence and Cross-Cultural Revisions

Recent research has expanded the field beyond a narrow set of Western infant laboratories. Bergelson and a large international collaboration analyzed everyday audio from 1,001 young children across six continents. Their outcomes concern environmental speech and children's vocal behavior, establishing the value and complexity of large observational corpora. They do not directly convert recorded talk into a validated measure of lexical knowledge. By contrast, cross-cultural comprehension studies can test whether children understand specified words in settings with different address practices. Together these approaches help separate exposure estimates from actual receptive knowledge and suggest that claims about a universal developmental input profile require geographically and linguistically diverse evidence. Bergelson et al. (2023) Foushee & Srinivasan (2024)

The 2024 investigation of Tseltal infants found comprehension of common nouns and greeting forms despite comparatively infrequent directly addressed speech. This matters theoretically because vocabulary can draw on locally available information even when the configuration of adult–child talk differs from the environments that inspired some influential early models. The result does not imply that directed interaction is irrelevant everywhere; it indicates that the informational contribution of overheard speech depends on cultural organization, routines, and the learner's opportunities for attention. The methodological lesson is that sampling the whole ecology of language use may reveal effective learning paths missed by a restricted definition of caregiver input. Foushee & Srinivasan (2024) Sperry et al. (2019)

Contemporary reviews refine the role of context and age. Tippenhauer and Saylor's 2026 synthesis addresses the mixed literature on variation in physical backgrounds and co-present competitors, proposing that learning outcomes depend on cognitive resources and conceptual knowledge. A 2026 experiment by Zhang and colleagues, involving eight- and nine-year-old children learning an artificial language, differentiated success in extracting syntax from success in acquiring word meanings. Such studies complicate any simple assumption that regularity detection, grammatical learning, lexical mapping, and long-term retention are interchangeable indicators of a general language-learning capacity. Their convergence is conceptual: the learner's achievements are multidimensional and change with task, input ecology, and developmental history. Tippenhauer & Saylor (2026) Zhang et al. (2026)

Evidence Quality, Replication, and Outstanding Research Problems

The strongest experimental conclusions concern tightly specified contrasts. Markman and Wachtel manipulated labeling contexts to investigate lexical expectations; Yu and Smith demonstrated cross-trial extraction under controlled ambiguity; Horst and Samuelson experimentally separated immediate selection from brief delayed retention; and Arunachalam and Waxman manipulated sentence information available before event interpretation. These are powerful demonstrations of what learners can use under particular circumstances, though their samples and stimulus structures restrict unrestricted generalization. Reviews should therefore describe the tested age, language, comparison conditions, and learning criterion before generalizing a reported effect. A study's theoretical importance is often greatest when its manipulation rules out a specific alternative explanation. Markman & Wachtel (1988) Yu & Smith (2007) Horst & Samuelson (2008) Arunachalam & Waxman (2010)

Longitudinal correlational findings provide a different form of evidence. Rowe and Weisleder with Fernald linked measurable linguistic environments to subsequent vocabulary or processing, while cross-cultural studies revealed the sensitivity of interpretation to sampling and coding. These investigations make developmental contexts visible, but associations can be influenced by unmeasured factors, reverse influence, and changes in behavior during observation. Independent replication requires more than reusing a familiar paradigm: it requires testing whether core findings survive alternative stimuli, larger and more diverse samples, preregistered outcome definitions, and appropriately controlled analyses. Publication bias and selective reporting are concerns whenever striking positive results are more likely to circulate than ambiguous or null effects. Rowe (2012) Weisleder & Fernald (2013) Sperry et al. (2019)

Several open questions organize the next research phase. Investigators need to measure how lexical hypotheses persist during natural play, which memory operations integrate widely spaced encounters, and when pragmatic interpretation adds predictive power beyond simple co-occurrence. The field also needs to explain word learning for actions, abstract relations, and culturally specialized categories using evidence comparable to the strong experimental tradition for novel concrete nouns. A major methodological challenge is linking brain, computational, and behavioral levels without interpreting a neural correlate as a complete causal explanation. Multilingual and cross-cultural longitudinal studies should test semantic knowledge directly, and educational studies should evaluate retention and transfer instead of relying on immediate training performance. McMurray et al. (2012) Tippenhauer & Saylor (2026) Bergelson et al. (2023)

Conclusion

Vocabulary acquisition and word learning constitute a developmental system that transforms variable linguistic experiences into increasingly stable and flexible lexical knowledge. The system combines identification of candidate forms, interpretation of reference, category generalization, accumulation of evidence across encounters, and long-term retrieval. Historical studies of single encounters and lexical constraints identified the initial inference problem; statistical learning and social-pragmatic experiments established multiple possible sources of evidence; process models and delayed-retention studies showed why rapid selection cannot serve as the sole criterion of acquisition. The most defensible scientific conception treats initial mapping, persistence, semantic accuracy, and productive accessibility as related but separately measurable dimensions. Carey & Bartlett (1978) Yu & Smith (2007) McMurray et al. (2012)

The empirical record supports a plural account of learning conditions. Lexical familiarity, speaker perspective, perceptual attention, syntactic context, communicative quality, and the temporal distribution of exposure all contribute under distinguishable circumstances. Their importance varies with age, prior vocabulary, semantic domain, linguistic experience, and the surrounding social ecology. Cross-cultural evidence establishes that familiar Western patterns of child-directed interaction are one route within a broader diversity of language environments. Longitudinal studies show that features of early input predict later lexical outcomes, while experimental interventions are required to identify the unique causal contribution of each candidate pathway. A model that explains only initial correct choices or only group-level word counts leaves the central developmental mechanism incomplete. Rowe (2012) Foushee & Srinivasan (2024) Sperry et al. (2019)

The research program that follows is operationally clear. Future investigations should combine controlled manipulation of referential and contextual evidence with delayed tests, naturalistic records of actual exposure, and principled models evaluated on individual learning trajectories. Each claim should state what was learned, how it was measured, whether it survived time, and how broadly it generalized. This integrated framework provides a stable foundation for the scientific architecture of Psychology of Language and for related research in cognitive, developmental, and educational psychology. It also offers a disciplined basis for comparing human lexical learning with artificial learning systems through explicit tasks and observable outcomes, without presuming that shared performance entails identical psychological mechanisms. McMurray et al. (2012) Tippenhauer & Saylor (2026)

References

  • Arunachalam, S., & Waxman, S. R. (2010). Meaning from syntax: Evidence from 2-year-olds. Cognition, 114(3), 442–446. https://doi.org/10.1016/j.cognition.2009.10.015
  • Baldwin, D. A. (1993). Infants' ability to consult the speaker for clues to word reference. Journal of Child Language, 20(2), 395–418. https://doi.org/10.1017/S0305000900008345
  • Bergelson, E., & Swingley, D. (2012). At 6–9 months, human infants know the meanings of many common nouns. Proceedings of the National Academy of Sciences, 109(9), 3253–3258. https://doi.org/10.1073/pnas.1113380109
  • Bergelson, E., Soderstrom, M., Schwarz, I.-C., Rowland, C. F., Ramírez-Esparza, N., Hamrick, L. R., Marklund, E., Kalashnikova, M., Guez, A., Casillas, M., Benetti, L., van Alphen, P., & Cristia, A. (2023). Everyday language input and production in 1,001 children from six continents. Proceedings of the National Academy of Sciences, 120(52), e2300671120. https://doi.org/10.1073/pnas.2300671120
  • Bion, R. A. H., Borovsky, A., & Fernald, A. (2013). Fast mapping, slow learning: Disambiguation of novel word–object mappings in relation to vocabulary learning at 18, 24, and 30 months. Cognition, 126(1), 39–53. https://doi.org/10.1016/j.cognition.2012.08.008
  • Carey, S., & Bartlett, E. (1978). Acquiring a single new word. Papers and Reports on Child Language Development, 15, 17–29. https://eric.ed.gov/?id=ED198703
  • Foushee, R., & Srinivasan, M. (2024). Infants who are rarely spoken to nevertheless understand many words. Proceedings of the National Academy of Sciences, 121(23), e2311425121. https://doi.org/10.1073/pnas.2311425121
  • Frank, M. C., Goodman, N. D., & Tenenbaum, J. B. (2009). Using speakers' referential intentions to model early cross-situational word learning. Psychological Science, 20(5), 578–585. https://doi.org/10.1111/j.1467-9280.2009.02335.x
  • Hirsh-Pasek, K., Adamson, L. B., Bakeman, R., Owen, M. T., Golinkoff, R. M., Pace, A., Yust, P. K. S., & Suma, K. (2015). The contribution of early communication quality to low-income children's language success. Psychological Science, 26(7), 1071–1083. https://doi.org/10.1177/0956797615581493
  • Hoff, E. (2003). The specificity of environmental influence: Socioeconomic status affects early vocabulary development via maternal speech. Child Development, 74(5), 1368–1378. https://doi.org/10.1111/1467-8624.00612
  • Horst, J. S., & Samuelson, L. K. (2008). Fast mapping but poor retention by 24-month-old infants. Infancy, 13(2), 128–157. https://doi.org/10.1080/15250000701795598
  • Kucker, S. C., & Samuelson, L. K. (2012). The first slow step: Differential effects of object and word-form familiarization on retention of fast-mapped words. Infancy, 17(3), 295–323. https://doi.org/10.1111/j.1532-7078.2011.00081.x
  • Markman, E. M. (1990). Constraints children place on word meanings. Cognitive Science, 14(1), 57–77. https://doi.org/10.1207/s15516709cog1401_4
  • Markman, E. M., & Wachtel, G. F. (1988). Children's use of mutual exclusivity to constrain the meanings of words. Cognitive Psychology, 20(2), 121–157. https://doi.org/10.1016/0010-0285(88)90017-5
  • McMurray, B., Horst, J. S., & Samuelson, L. K. (2012). Word learning emerges from the interaction of online referent selection and slow associative learning. Psychological Review, 119(4), 831–877. https://doi.org/10.1037/a0029872
  • Rowe, M. L. (2012). A longitudinal investigation of the role of quantity and quality of child-directed speech in vocabulary development. Child Development, 83(5), 1762–1774. https://doi.org/10.1111/j.1467-8624.2012.01805.x
  • 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
  • Schroer, S. E., & Yu, C. (2024). Word learning is hands-on: Insights from studying natural behavior. Advances in Child Development and Behavior, 66, 55–79. https://doi.org/10.1016/bs.acdb.2024.04.002
  • Smith, L. B., Jones, S. S., Landau, B., Gershkoff-Stowe, L., & Samuelson, L. (2002). Object name learning provides on-the-job training for attention. Psychological Science, 13(1), 13–19. https://doi.org/10.1111/1467-9280.00403
  • Sperry, D. E., Sperry, L. L., & Miller, P. J. (2019). Reexamining the verbal environments of children from different socioeconomic backgrounds. Child Development, 90(4), 1303–1318. https://doi.org/10.1111/cdev.13072
  • Tippenhauer, N., & Saylor, M. M. (2026). Early word learning in context: A review of the literature. Infant and Child Development, 35(1), e70082. https://doi.org/10.1002/icd.70082
  • Trueswell, J. C., Medina, T. N., Hafri, A., & Gleitman, L. R. (2013). Propose but verify: Fast mapping meets cross-situational word learning. Cognitive Psychology, 66(1), 126–156. https://doi.org/10.1016/j.cogpsych.2012.10.001
  • Vlach, H. A., & Sandhofer, C. M. (2012). Fast mapping across time: Memory processes support children's retention of learned words. Frontiers in Psychology, 3, 46. https://doi.org/10.3389/fpsyg.2012.00046
  • Weisleder, A., & Fernald, A. (2013). Talking to children matters: Early language experience strengthens processing and builds vocabulary. Psychological Science, 24(11), 2143–2152. https://doi.org/10.1177/0956797613488145
  • Yu, C., & Smith, L. B. (2007). Rapid word learning under uncertainty via cross-situational statistics. Psychological Science, 18(5), 414–420. https://doi.org/10.1111/j.1467-9280.2007.01915.x
  • Zhang, W., Monaghan, P., Bennett, S., & Rebuschat, P. (2026). Children's simultaneous or successive acquisition of vocabulary and grammar: Evidence from cross-situational learning. Journal of Child Language, 53(5), 1737–1751. https://doi.org/10.1017/S0305000925100135