What Is Language and Concept Formation?
Language and concept formation concerns the ways words participate in building, activating, distinguishing, and revising mental categories. A concept is a flexible body of knowledge used to recognize an instance, draw inferences, anticipate properties, and relate experiences. A category is a grouping defined by some criterion; the concept supporting it may include typical examples, structural relationships, expectations, and causal knowledge. A verbal label is a conventional or newly learned linguistic expression that can refer to a category or a property. Learning the word bird and learning what makes birds a useful category are interdependent achievements, but they are measured separately in psychological research.
Labels influence concept learning by making category membership salient, directing attention toward shared properties, supporting retrieval of related knowledge, and helping people coordinate distinctions with other speakers. Experiments show causal effects on particular classification and generalization tasks; the direction and magnitude depend on whether the labels are informative, consistent, developmentally meaningful, and relevant to the learner's goal. Foundational demonstrations include category-learning experiments with unfamiliar objects and studies of young infants' emerging links between nouns and kinds (Lupyan et al., 2007) (Waxman and Markow, 1995).
A label, a category, and a concept are different levels of explanation
Consider three unfamiliar animals with different coats but the same anatomical feature. An experimenter might call all three dax. The label supplies a cue that they belong together; repeated observation may reveal the shared anatomy; subsequent evidence may support a causal theory about how that anatomy works. A learner can repeat dax without understanding the commonality, can infer the commonality before learning a word, or can use the word to search for the commonality. Sound reproduction, category assignment, and conceptually productive generalization therefore need separate outcome measures.
This distinction also clarifies the term abstraction. Abstraction is the selective representation of properties or relations shared across different instances. It may involve perceptual similarity, statistical regularities, functions, explanations, or relational structures. Naming can facilitate abstraction when it leads the learner to prioritize a stable pattern. When labels are applied inconsistently or too broadly, the same mechanism may conceal distinctions that matter. A rigorous account consequently studies both gains and errors.
Why Names Can Change What Learners Notice
Attention to relevant features
A label typically co-occurs with experiences and supplies a compact cue about how the listener should group them. In an unfamiliar category task, objects can vary simultaneously in shape, texture, color, motion, and hidden attributes. The learner must decide which differences predict category membership. Hearing a common label can direct sampling toward features shared by named members; hearing two names can encourage comparison of features that separate categories. This attentional interpretation is compatible with selective activation of object representations, but it does not establish a universal neural route for word-to-perception influence.
The label-feedback hypothesis proposes that linguistic activation can modulate perceptual and conceptual processing in real time. It draws on effects of labeling in identification and discrimination tasks as well as learning. Its predictions concern changing processing states rather than requiring that vocabulary permanently fixes visual perception. The framework is influential, and particular mechanisms remain a subject of experimental comparison (Lupyan, 2012).
Reducing search and working-memory demands
Category learning requires a learner to compare current instances with remembered cases and provisional rules. A recurring word can serve as a retrieval handle: on hearing the word, the learner can bring previously experienced examples or prototypical characteristics into consideration. This may reduce the effort of maintaining multiple cases separately. The benefit is conditional; if the name retrieves misleading examples, the same shorthand can promote an incorrect decision. Researchers should distinguish better memory organization from an independent change in perceived stimulus features.
Labels also allow category knowledge to be rehearsed internally and transmitted between people. A teacher who names a pattern can return to it throughout an extended explanation, and a child can ask whether a new example is another instance. These are plausible routes by which social language supports learning. Laboratory performance on simple visual categorization tasks does not by itself demonstrate lasting improvements in classroom achievement, so educational transfer requires dedicated assessment.
Shared labels and category boundaries
Calling two perceptually different things by the same name encourages a learner to consider a common class; different names can highlight potential contrasts. The inference is probabilistic. Multiple categories can apply simultaneously: a robin can be a bird, an animal, a pet in a particular household, or something seen at a feeder. Each category selects different properties. Label-driven selection is thus naturally understood as task-dependent organization of conceptual space.
In a study of superordinate categories, Rissman and Lupyan compared prompts consisting of a category label, a list of examples, or a definition. Responses elicited by a label tended to be more typical of the category than responses elicited by alternative prompts, with variation across categories. The finding shows that a word can be a distinctive route into existing conceptual knowledge. It does not show that words contain complete definitions or that every category is represented by a single prototype (Rissman and Lupyan, 2024).
Experimental Evidence That Labels Facilitate Learning
Novel-category learning with redundant names
Lupyan, Rakison, and McClelland trained participants to sort unfamiliar alien-like figures into categories associated with approach or avoidance decisions. All participants received performance feedback; some additionally heard nonsense names paired with categories. These labels were redundant with other information about the correct response. Nevertheless, labeled training facilitated category acquisition relative to the comparison condition. A follow-up comparison with nonverbal associations suggested that the advantage was not simply the presence of an extra paired cue (Lupyan et al., 2007).
The logic of this design is important. Because the names added no necessary classification information, the improvement points toward effects on learning or representational organization rather than merely supplying the missing answer. It does not identify a single underlying mechanism. Labels might alter attention, consolidate category representations, influence retrieval strategies, or change expectations about the experimenter's intended structure. Subsequent studies try to separate these alternatives.
Immediate gains and possible persistence after training
Fotiadis and Protopapas compared category training with a paired-associate task and examined labeled versus harder-to-name artificial categories. Labels increased categorization accuracy during learning, whereas a comparable advantage did not appear in the noncategorization association task. Their post-training visual discrimination and eye-movement observations offered some indication of continued effects after learning, although the sustained evidence was more limited than the immediate facilitation result. This is useful evidence for the specificity of label benefits to categorization demands (Fotiadis and Protopapas, 2023).
The distinction between immediate performance and enduring conceptual change deserves close attention. A learner may use a recently heard name to complete a task quickly while still lacking stable category knowledge. A delayed, unlabeled test, a transfer test with new objects, and a measure of inference about unobserved properties probe different aspects of learning. A report that labels improved accuracy during training should not be restated as proof of permanent conceptual transformation.
Labels compared with nonlinguistic cues
A rigorous control condition needs to compare verbal labels with other cues that are equally visible, memorable, and presented in an equally credible manner. Gervits, Johanson, and Papafragou used labels, numbers, and symbols to guide classification of perceptually ambiguous novel natural kinds. They manipulated whether the cues were described as relevant, irrelevant, or left unspecified. Across their experiments, labels exerted stronger influence on category boundaries than the alternative symbols, even while instructions about relevance also mattered (Gervits et al., 2023).
These results strengthen the case that familiar linguistic conventions confer more than generic stimulus salience. They also demonstrate why task framing is part of the causal story. If an investigator tells participants that a signal matters, the message may change how participants allocate effort. A claim that labels have an intrinsic and context-free psychological power would exceed what such task-dependent results establish.
What Infant Studies Reveal About Early Conceptual Development
Words as invitations to categorize
Waxman and Markow studied 12- to 13-month-old infants at the early stages of word learning. Their experiments examined whether novel words help infants treat distinct objects as members of a common kind. The resulting noun–category linkage became a central finding in developmental accounts: hearing a count noun can encourage attention to category structure rather than treating every object as isolated. The developmental interpretation concerns an emerging bias in the context of language acquisition, not a complete adult-like theory of natural kinds (Waxman and Markow, 1995).
A naming event carries several signals at once. There is the acoustic form of speech, the temporal pattern of the event, the adult's communicative behavior, and the potential assumption that a word applies beyond a single object. Careful infant work attempts to isolate which elements alter looking behavior. Interpretations should respect the limited information carried by a single novelty-preference response: looking time reflects attention under a specific procedure and is not a direct verbal report of a mature concept.
Speech compared with tones in early infancy
Ferry, Hespos, and Waxman tested 46 infants aged three to four months using familiarization trials accompanied by speech or tone sequences. At test, infants in the speech condition showed evidence consistent with object categorization that was absent in the tone comparison. The experiment supports an early connection between speech and the organization of object experience, before the infants could produce the words. Its narrow age range and method constrain claims about the later development of symbolic meaning (Ferry et al., 2010).
Early sensitivity to speech should be interpreted developmentally. Speech might serve as an especially engaging social or auditory signal, and some accounts emphasize the emergence of expectations that words refer to kinds. Investigators need matched controls for sound properties and appropriate follow-up tests before choosing between perceptual engagement and category-reference explanations. An effect of hearing speech in infancy is not equivalent to evidence that infants possess the named adult category.
When names disrupt categories
Plunkett, Hu, and Cohen trained ten-month-old infants with artificial visual categories while varying the relationship between words and categories. Distinct labels that matched categories permitted learning comparable to an unlabeled condition. Uncorrelated labels disrupted categorization, while one label repeated across different groups promoted a broader grouping. Crucially, categorization was assessed without presenting the labels at test. The results showed that naming conditions can alter later grouping even in preverbal infants (Plunkett et al., 2008).
This finding prevents a simplistic rule that verbalization always helps. A misleading label can privilege an inappropriate division of the stimulus space; one common name can encourage overgeneralization. The best interpretation is that linguistic signals become part of the evidence the learner uses to infer category structure. Better learning depends on the relationship between the linguistic organization and the structure the task actually tests.
Labels, Similarity, and Inductive Generalization
From grouping to predicting hidden properties
Recognizing two objects as members of a category is only one use of a concept. Inductive generalization asks whether a learner projects a new fact from one instance to another: if one unfamiliar bird has a particular biological feature, is a second bird likely to share it? A shared name may support this inference by signaling a common kind, but visual similarity, known causal relations, and knowledge about natural kinds can also govern the answer. Classification and inference outcomes should therefore be examined separately.
Johanson and Papafragou studied four-year-old children and adults who classified ambiguous objects using category names, observable facts, or unobservable facts. Both names and facts affected how stimuli were grouped, including when those cues conflicted with perceptual similarity. Their results establish that linguistic names are influential while showing that concept-relevant information other than labels can be comparably powerful. An accurate theory must therefore accommodate inference from factual and causal knowledge (Johanson and Papafragou, 2016).
The feature-based versus category-marker debate
One theoretical disagreement concerns what a label is psychologically, especially early in development. A feature-based approach treats the heard or seen word as one more attribute accompanying an object, capable of increasing overall similarity between instances. A category-marker account treats a common label as a signal that objects belong to the same kind and that their unobserved properties may be shared. Both approaches predict many effects of consistent naming, so discriminatory evidence requires conditions in which labels conflict with appearance.
Deng and Sloutsky found developmental differences when category names were compared with other stimulus information in induction tasks. Their experimental results support a trajectory in which young children's use of labels can resemble their use of other features, while adult learners can give labels a more category-defining role. Sloutsky and Fisher likewise examined the possibility that early labels contribute to perceived similarity. These results complicate claims that every listener interprets words as abstract kind markers from the outset (Deng and Sloutsky, 2013) (Sloutsky and Fisher, 2012).
Development changes the meaning of a word cue
As vocabulary, conceptual knowledge, and conversational experience grow, learners acquire expectations about what different expressions are used to denote. The utterance 'This is a blicket' and the sentence 'It has blicket-colored wings' can imply distinct kinds of grouping. A novel noun may invite category search, whereas a description of a property can draw attention to one dimension. The distribution of previous linguistic experience helps explain why two individuals may use the same surface cue differently.
Linguistic conventions interact with causal explanation. A preschooler might initially group insects by color; a more experienced learner might know that species membership is better tracked by anatomy and reproduction. Naming helps when it activates knowledge suited to the problem. The availability of a noun does not determine which inferential framework is scientifically appropriate; expertise, evidence and task purpose remain central.
Concept Formation Without Explicit Words
Perceptual similarity and statistical structure
People and other animals can discriminate and generalize patterns without verbal names. Infants also form expectations about objects before they can produce conventional labels. Category formation can emerge from repeated exposure to common shapes, trajectories, co-occurrences, or event regularities. The question investigated here is how adding a word changes those preexisting learning processes, not whether every concept originates in language.
This matters experimentally because studies often compare labeled training against unlabeled training, but the unlabeled group is still learning. A label advantage means better performance under specified conditions. It does not imply that the control group had no conceptual representations, nor does a null label effect demonstrate the absence of any meaningful language–thought interaction.
Comparison and alignment
Placing two examples side by side can draw attention to shared structure even before a learner knows their names. Research on comparison suggests that comparing multiple exemplars can lead young children toward taxonomic categories that they may not infer from a single example. This is a route to conceptual generalization that can operate alongside naming. The label can then summarize the comparison or act as a cue to undertake it (Namy and Gentner, 2002).
A teaching activity that supplies several varied examples, contrasting nonexamples, and an accurate category name provides multiple sources of structure. If learners succeed, the result cannot be attributed solely to the label. Tests must independently manipulate naming, example diversity, comparison opportunities and feedback to identify the contribution of each.
Theories of How Words Affect Concepts
Label-feedback and flexible conceptual activation
Under label-feedback accounts, words recruit or bias information relevant to a category while processing is underway. A category name can make some features more accessible, potentially facilitating a difficult identification or prioritization judgment. Lupyan's theoretical synthesis treats linguistic modulation as flexible and task-sensitive. It is a mechanistic proposal supported by several converging findings, while claims about which processing stage changes must be tested directly (Lupyan, 2012).
An important implication is that the same object can be perceived and described under many categories. Calling a kitchen object a tool foregrounds different properties than calling it a decorative item. The situation, goal, and label jointly determine the active conceptual frame. Evidence for flexible access is more informative than a universal claim that words permanently reshape sensory representations.
Feature-weighting and associative learning
A feature-weighting model explains label effects through changed attention to dimensions and learned associations between cues and outcomes. If a category boundary depends on shape and a consistent word predicts correct grouping, learners may learn to emphasize shape. A word can function as a highly reliable predictor under supervised training. Models differ in whether language has a privileged representational status or influences cognition because of its learned reliability and communicative significance.
The feature-based findings in young children and the apparent category-marker advantage in some adult tasks suggest that a single model may need age-dependent assumptions about learned associations and how the listener interprets naming. Mechanistic adequacy is assessed by whether a model predicts transfer to new exemplars, response under verbal interference, and behavior when words conflict with visual evidence (Deng and Sloutsky, 2013) (Gervits et al., 2023).
Words as retrieval cues for prototypes and exemplars
A concept can be represented through remembered individual cases, summary tendencies, structured relational information, or combinations of these. A category word may activate highly typical members more readily than atypical members, as reflected in the superordinate-category generation findings. This does not require assuming that the category has a single stored essence. Rather, the cue can bias retrieval toward an accessible region of conceptual space (Rissman and Lupyan, 2024).
The difference between typicality and membership is crucial. A penguin may be atypical of common bird imagery but still a bird. If a teaching method reinforces only typical examples, a learner may mistake frequent properties for necessary conditions. A robust learning design checks whether the learner can accommodate atypical members and reject deceptively similar nonmembers.
Causal, theory-based, and socially communicated concepts
Some categories are learned through explanations of hidden structure: why a medical instrument works, what defines a biological kind, or why a social role carries particular duties. Labels can compress these explanations and make them reusable across conversations. But a name alone can produce a feeling that a phenomenon has been explained. Giffin, Wilkenfeld, and Lombrozo found that named categories could increase participants' satisfaction with explanations and promote inferences about underlying causes, even when the explicit information was limited (Giffin et al., 2017).
The educational implication is straightforward: a fluent label is a cue to a possible explanation, and the explanation itself still needs evidence. Learners can be asked what the category predicts, what would count as a counterexample, and what distinguishes it from nearby categories. These questions assess whether the label supports a usable conceptual model.
Abstraction, Transfer, and Conceptual Change
Learning beyond the training examples
A learner has formed a broadly useful concept when knowledge extends to cases never presented during training. Transfer can be near, involving new examples that resemble familiar ones, or far, involving new contexts and superficially different materials that share an underlying principle. Naming can potentially support transfer by providing a stable handle for what should generalize. Claims about such transfer require tests on new items, since memorizing the training set can produce perfect classification without abstraction.
Experiments that measure only speed or accuracy during training leave this issue unresolved. A better evaluation includes a novel-example test, an explanation task, a delayed test, and questions about exceptions. A label-driven advantage on one measure may coexist with an unchanged or reduced ability to generalize. These outcomes help distinguish short-term cue use from durable learning.
Conceptual change when a category name is revised
Conceptual change occurs when a learner reorganizes an earlier understanding in light of new information. Renaming a category can be part of that process: a broad everyday label may divide into two scientific categories once a newly discovered distinction becomes important. Conversely, two previously separated groups may be understood as members of one higher-order class. Such revisions depend on causal and evidential learning as well as terminology.
A concrete example comes from biological classification. Learning an accurate genus or species distinction can guide questions about ancestry and shared properties, while simply drilling Latin names cannot substitute for learning the biological evidence. Words organize the search space; observation and explanation govern whether the proposed boundaries deserve to be retained.
When verbal labels are too broad, inconsistent, or misleading
Label-based learning can fail when several conceptually different things receive one umbrella name, or when the same category is assigned unrelated names without explanation. Overly broad naming may encourage overextension; inconsistent naming may obscure stable patterns. The infant study by Plunkett and colleagues shows a particularly clear causal example, because the naming manipulation changed classification behavior at a later unlabeled test (Plunkett et al., 2008).
A misleading name can also create a false confidence in explanation. The mere presence of a technical term may imply that an underlying mechanism is known, even when the term only describes a pattern. Responsible scientific education links a category name to operational criteria, discriminating evidence, known exceptions and meaningful uncertainty (Giffin et al., 2017).
How Researchers Test the Causal Role of Labels
Controlled category-learning paradigms
Investigators commonly create unfamiliar visual objects, randomly assign participants to conditions, and vary whether a word accompanies each object or category. The task may involve supervised classification, passive familiarization, free sorting, novelty preference or generalization to unseen stimuli. Random assignment supports causal inference about the experimental naming manipulation, provided groups receive equivalent exposure and test conditions.
The most informative comparison does not merely contrast speech with silence. It considers unrelated sounds, numbers, symbols, repeated nonlinguistic signals, or meaningful facts, matched as closely as practical for timing and salience. This matters because a label may change the learner's belief about which cues the experimenter regards as important (Gervits et al., 2023) (Johanson and Papafragou, 2016).
Verbal interference and testing without the cue
One way to probe online linguistic mediation is to add a concurrent verbal task, limiting access to language-related resources while participants categorize objects. An effect that weakens under verbal interference suggests a task-dependent contribution of verbal processing, though interference also imposes broader attention and memory demands. A delayed test presented without labels helps assess persistence beyond immediate cueing.
Interference paradigms and unlabeled transfer tests answer different questions. The former asks whether language supports processing during the task. The latter asks whether earlier linguistic experience changed subsequent performance. Neither test alone reveals an exhaustive neural mechanism. Researchers should analyze accuracy, reaction time, gaze, error patterns and generalization where appropriate.
Distinguishing category learning from performance strategies
A participant may quickly learn that an experimenter expects objects called dax to be grouped together. Compliance with this instruction can improve accuracy even before robust conceptual understanding develops. Conversely, a participant might acquire a category but perform poorly because the task is confusing or memory-demanding. Experiments can probe these possibilities by withholding labels at test, using novel items, asking for justifications, and comparing tasks that require inference versus simple label matching.
The stronger claim that labels change conceptual organization requires convergent evidence across measures and ideally across different contexts, laboratories and participant samples. Significant results in one artificial category-learning task warrant a precise claim about that task. Long-term development, real-world problem solving and classroom effectiveness require additional research.
Individual and Developmental Differences
Age, vocabulary, and accumulated knowledge
The influence of labels changes with the learner's experience. Infants may respond to properties of speech and emerging noun–category expectations; preschoolers can treat labels as salient features or kinds depending on context; adults have richer lexical semantics, stronger expectations about naming conventions, and more elaborate causal knowledge. The developmental literature shows both continuities and differences rather than a single fixed word effect (Ferry et al., 2010) (Deng and Sloutsky, 2013).
Vocabulary knowledge can affect whether a word efficiently activates a useful concept. Familiar technical terminology can assist experts yet confuse beginners who have not learned its meaning. Instructional design should therefore diagnose what the learner can already infer, rather than assuming that giving everyone the same noun has the same psychological consequence.
Task expertise and domain knowledge
A specialist sometimes disregards superficial similarity in favor of features that reliably signal function or causal structure. For example, a botanist may classify visually dissimilar plants together based on reproductive properties. An unfamiliar category name can help a novice search for such distinctions, but an expert's conceptual organization is supported by extensive examples, principles and feedback. Label training should be evaluated against the knowledge structures the learner already has.
Professional vocabularies are valuable partly because they stabilize distinctions that can be used repeatedly in coordinated work. Yet terminology can be adopted faster than conceptual skill. Accuracy on definitions, diagnostic descriptions, and novel cases may diverge, so knowledge assessment should incorporate explanations and application, not only recall.
Multilingual naming and cross-language variation
Languages differ in how vocabulary conventionally partitions experience. This creates opportunities to study whether speakers preferentially use category divisions made salient by familiar words. The core question in the present article is the local effect of a label on learning and conceptual access. Broad claims about linguistic relativity require distinct comparisons across language communities, careful measurement of confounds, and evidence about what changes when labels change.
For multilingual learners, two names may refer to overlapping categories while evoking different examples, cultural practices, or conversational contexts. Teaching should reveal these correspondences explicitly. Such variation does not imply that the learner has incompatible minds or lacks the ability to understand a category outside a preferred language.
Applications to Education and Concept Teaching
Introduce names together with examples and contrasts
A useful teaching sequence begins with a concrete task, introduces a stable category word, presents several varied positive examples and carefully chosen contrasts, and then asks learners to explain what is shared. The word helps learners discuss and retrieve the category; the examples show what the word has to accomplish. A name should be paired with reasons for inclusion and exclusion, because category accuracy depends on recognizing boundaries as well as central cases.
The experimental basis supports treating accurate, consistent labels as potentially helpful cues to learning. However, the size of an educational benefit depends on domain, age, instructional sequence and outcome. Laboratory effects should not be translated into guaranteed gains in mathematics, science or literacy without direct teaching studies (Lupyan et al., 2007) (Fotiadis and Protopapas, 2023).
Use exceptions and nonexamples to prevent overgeneralization
If all examples of a category share an accidental surface feature, the learner may infer that it defines the category. Diverse examples and counterexamples prevent the word from becoming a substitute for scrutiny. In science instruction, an accurate category is often refined by discussing borderline instances and exceptions. Asking 'What would make this example belong elsewhere?' tests the learner's rule.
An appropriate sequence may alternate naming with unlabeled classification. First, students use the term while examining examples. Later, they must classify new material without hearing the word in advance, describe which evidence mattered, and defend the distinction. This helps teachers identify whether the label has become a useful conceptual resource or remains only a remembered sound.
Support conversation, retrieval, and feedback
The practical value of a label is partly social. A shared term enables a teacher and learner to refer to the same pattern across lessons, ask questions, and correct misunderstandings. Brief explanations, spaced retrieval opportunities, and feedback about why an example fits can reinforce the conceptual link. These are general instructional design principles; the experiments reviewed here specifically support naming as one influence on categorization, not a standalone comprehensive teaching intervention.
When students are asked to explain a concept in their own words, instructors can see whether they have learned more than the designated term. A productive discussion makes the relation between name, evidence, mechanism and possible exceptions explicit. In heterogeneous classrooms, culturally accessible examples and multilingual clarification can improve opportunities to demonstrate knowledge.
Risks of Treating Names as Explanations
The illusion of understanding
Naming may create the impression that a causal explanation has been supplied. In the experiments by Giffin and colleagues, category labels made otherwise similar explanations more satisfying and encouraged an inference that an underlying cause existed. Such a response may be reasonable when names ordinarily track explanatory kinds, yet can mislead when a label does little more than restate the observed phenomenon (Giffin et al., 2017).
A scientific term should therefore prompt questions about measurement and mechanism. What evidence establishes the category? Which observations would count against it? What does the term predict beyond its own definition? These questions are especially important when labels are used to explain behavior, learning styles, social differences, or psychological characteristics.
Essentialism, stereotyping, and false boundaries
Human categories sometimes invite an assumption that every member possesses a hidden shared essence. Such an assumption can aid predictions in domains with stable underlying causal structure, but it can overstate coherence when group boundaries are conventional, overlapping or socially negotiated. A name can also direct attention toward category differences while hiding substantial variation within the group. The demonstrated power of names to steer categorization makes accuracy and context important.
Scientific education should differentiate a useful analytical category from a claim of immutable human nature. Especially for social categories, words can be politically and historically consequential. Experimental results about novel objects provide no license to infer that arbitrary human group labels capture fixed cognitive or biological essences.
Language, Thought, and the Limits of Linguistic Relativity
Specific labeling effects and broad theories
Experiments where participants are randomly given category names can show that language influences behavior in that controlled setting. This is a narrower and more testable proposition than the idea that an entire language determines what its speakers can think. Strong linguistic determinism makes far-reaching claims about conceptual possibility; local label-based modulation concerns how available words guide attention, representation and choice in a task (Lupyan, 2012).
The strongest account distinguishes an individual's temporary strategy, a learned category representation, the conventional structure of a lexicon, and wider linguistic experience. These levels can interact, but evidence at one level cannot automatically prove an effect at every other level. Researchers also have to distinguish different uses of 'perception': discrimination performance, attention, memory and conscious experience are not interchangeable measures.
Concepts can be revised as evidence grows
A language user can learn that an everyday word covers several scientific kinds, discover an unfamiliar distinction, and adopt a new name. Conceptual systems are revisable because perceptual input, inferential reasoning, social feedback, and explicit definitions all supply information. Language provides an unusually efficient set of representational cues, while valid conceptual change is anchored in improved prediction and explanation.
Accordingly, a finding that labels shift a category boundary supports a causal role for naming under particular conditions. It does not entail that speakers are confined to the categories their language already names. The vocabulary available in a community is one resource for reasoning, and humans continually create, borrow and refine terms to express newly learned distinctions.
Research Quality, Methodological Limits, and Competing Interpretations
Sample size, artificial stimuli, and replication
Many influential labeling experiments use unfamiliar drawings, arbitrary names, and a brief laboratory task. Such methods offer experimental control but only partial ecological validity. A newly learned boundary between invented aliens may not behave like a scientifically rich concept acquired over years. Sample sizes in particular infant experiments can be modest, looking-time measurements require careful analysis, and the literature includes heterogeneous conditions that limit simple generalization (Ferry et al., 2010) (Plunkett et al., 2008).
Judging the evidence requires attention to preregistration where available, independent replication, confidence intervals, age-specific predictions, and the extent of transfer to novel stimuli. The experiments establish that effects exist in defined paradigms; they do not provide a single pooled effect size applicable to every age or instructional domain. Evidence for durable changes in complex real-world knowledge is less direct.
Attention, pragmatic inference, and association are alternative mechanisms
Suppose a child groups two objects after hearing the same name. The response might indicate a shared-kind inference, increased perceptual similarity, attentional focusing, associative cue learning, or a belief about what the experimenter intends. These are meaningfully different mechanisms with overlapping behavioral predictions. Comparisons among labels, symbols and facts, and manipulations of relevance, help make them empirically distinguishable (Johanson and Papafragou, 2016) (Gervits et al., 2023).
No single result settles the feature-versus-category-marker debate across development. The combined record suggests that learned word meanings, acoustic salience, task instructions and categorical knowledge can each contribute. Strong theory requires explaining both successful learning with accurate labels and the errors induced by misleading ones.
Evidence tiers
Established within studied tasks: labels can causally affect categorization, and some experiments show benefits relative to silent or nonlinguistic comparison conditions. Established as a limitation: labels can also disrupt grouping when their correspondence to category structure is misleading. Supported but mechanism-dependent: words guide attention and conceptual retrieval. Contested: whether labels initially function chiefly as features or as privileged category markers across ages. Limited: general claims about long-term educational impact from brief artificial-category experiments (Lupyan et al., 2007) (Plunkett et al., 2008) (Deng and Sloutsky, 2013).
The label-feedback hypothesis is a theoretical explanation with evidence from multiple tasks; it should be described as such. Studies of superordinate category accessibility and causal explanation add further dimensions to the problem but should not be treated as interchangeable demonstrations of concept acquisition. The credibility of an inference depends on the exact task, population, manipulation and measured outcome.
Distinguishing Neighboring Psychological Concepts
Concept formation and semantic memory
Concept formation refers to acquiring and reorganizing representations used for categorization and inference. Semantic memory concerns knowledge of meanings, facts, concepts and relationships accumulated over experience. The two are closely connected: a newly learned label may access preexisting semantic structures, and repeated learning can alter stored knowledge. The broader knowledge-system question has its own canonical treatment in Semantic Memory: Facts, Meaning, and Knowledge.
Concept formation and general psycholinguistics
Psycholinguistics examines perception, comprehension, production, acquisition and use of language across different levels. Naming effects on category learning represent one intersection of language processing and conceptual organization. For the broader disciplinary framework, methods and architecture of human language processing, consult Psycholinguistics: How Language Is Processed in the Mind.
Labels, framing, and judgment
A category label can guide which features a learner notices. A framing effect concerns how the wording or presentation of options influences evaluation and decisions. These processes may interact, yet a category-learning experiment does not automatically establish a decision-framing effect. The canonical treatment of choice wording belongs to Framing Effect: How the Wording of Choices Changes Decisions.
Frequently Asked Questions
Do children need words before they can form concepts?
No. Infants develop perceptual categories and expectations before producing conventional words. Naming provides a further source of structure and can influence categorization remarkably early, including in looking-time paradigms with preverbal infants. The evidence therefore concerns interaction between language and conceptual development rather than a requirement that all concepts begin as words (Ferry et al., 2010) (Waxman and Markow, 1995).
Can teaching a new word improve learning?
Yes, under specific conditions. Accurate, consistent labels can facilitate novel-category learning, especially when they guide attention toward a useful structure or provide a stable reference across examples. Outcomes also depend on examples, explanations and learner knowledge. The evidence does not establish that vocabulary instruction alone guarantees durable understanding (Lupyan et al., 2007) (Fotiadis and Protopapas, 2023).
Can a label make learning worse?
Yes. When words divide stimuli unpredictably or collapse important distinctions, categorization can be disrupted or broadened in misleading ways. Infant experiments manipulating label-to-category correspondence provide direct evidence. A label should be evaluated by the relationships and inferences it supports rather than merely by its presence (Plunkett et al., 2008).
Are words more influential than arbitrary symbols?
In some controlled adult studies, labels influenced classification more strongly than numbers or symbols. The size of the difference depends partly on how participants understand the task relevance of each cue. This supports a distinctive role for linguistic naming in those experiments without establishing that every word always outweighs every alternative signal (Gervits et al., 2023).
Does having a name mean understanding a concept?
No. A learner may correctly repeat a name while failing to identify new members, explain the rule, or draw appropriate inferences. Conversely, a learner may perform meaningful classification without knowing the conventional term. Strong assessments combine naming with unlabeled transfer, counterexamples, explanations and evidence-based predictions.
Does language determine what people are able to think?
Controlled studies demonstrate that available labels can shape attention and classification under particular conditions. They do not establish that thought is confined to existing words. People learn new distinctions through observation, comparison, inference and social instruction, and they can introduce new vocabulary to express them. The precise extent of linguistic influence remains a question for carefully designed tasks and populations.
Conclusion
Language contributes to concept formation by giving learners reusable cues for organizing experience. Labels can highlight shared structure, redirect attention, activate familiar examples, and sustain inferences across encounters. Their effects appear in experiments with infants, children and adults, including studies where names aid learning beyond otherwise sufficient feedback. The same evidence shows that labels can misdirect grouping when linguistic and perceptual information conflict.
The most defensible scientific conclusion is conditional and generative: verbal naming is one important mechanism within a larger system of perception, similarity learning, causal inference, memory, social interaction and conceptual revision. To know whether a word has supported genuine learning, researchers and educators must test what learners can recognize, explain and predict beyond the original naming situation. The concept becomes scientifically useful through the distinctions it makes and the consequences those distinctions successfully organize.
Related Articles
Cluster Language, Thought and Psycholinguistics — the published research chapter situating concept formation within language and thought.
Psycholinguistics: How Language Is Processed in the Mind — cognitive foundations of language representation, comprehension, production, and learning.
Semantic Memory: Facts, Meaning, and Knowledge — knowledge structures, concepts, and semantic retrieval.
Framing Effect: How the Wording of Choices Changes Decisions — a separate canonical account of wording effects in judgment and decisions.
Inner Speech: How Language Supports Thought and Self-Regulation — How self-directed speech supports thinking, learning, and self-regulation.
Embodied Language Comprehension: How Action and Perception Shape Meaning — Connections between linguistic meanings, perception, and action representations.
References
Deng, W., and Sloutsky, V. M. (2013). The role of linguistic labels in inductive generalization. Journal of Experimental Child Psychology, 114(3), 432–455. Source
Ferry, A. L., Hespos, S. J., and Waxman, S. R. (2010). Categorization in 3- and 4-month-old infants: An advantage of words over tones. Child Development, 81(2), 472–479. Source
Fotiadis, F. A., and Protopapas, A. (2023). Immediate and sustained effects of verbal labels for newly-learned categories. Quarterly Journal of Experimental Psychology, 76(8), 1773–1789. Source
Gervits, F., Johanson, M., and Papafragou, A. (2023). Relevance and the role of labels in categorization. Cognitive Science, 47(12), e13395. Source
Giffin, C., Wilkenfeld, D., and Lombrozo, T. (2017). The explanatory effect of a label: Explanations with named categories are more satisfying. Cognition, 168, 357–369. Source
Johanson, M., and Papafragou, A. (2016). The influence of labels and facts on children's and adults' categorization. Journal of Experimental Child Psychology, 144, 130–151. Source
Lupyan, G. (2012). Linguistically modulated perception and cognition: The label-feedback hypothesis. Frontiers in Psychology, 3, 54. Source
Lupyan, G., Rakison, D. H., and McClelland, J. L. (2007). Language is not just for talking: Redundant labels facilitate learning of novel categories. Psychological Science, 18(12), 1077–1083. Source
Lupyan, G., and Thompson-Schill, S. L. (2012). The evocative power of words: Activation of concepts by verbal and nonverbal means. Journal of Experimental Psychology: General, 141(1), 170–186. Source
Namy, L. L., and Gentner, D. (2002). Making a silk purse out of two sow's ears: Young children's use of comparison in category learning. Journal of Experimental Psychology: General, 131(1), 5–15. Source
Plunkett, K., Hu, J.-F., and Cohen, L. B. (2008). Labels can override perceptual categories in early infancy. Cognition, 106(2), 665–681. Source
Rissman, L., and Lupyan, G. (2024). Words do not just label concepts: Activating superordinate categories through labels, lists, and definitions. Language, Cognition and Neuroscience, 39(5), 657–676. Source
Sloutsky, V. M., and Fisher, A. V. (2012). Linguistic labels: Conceptual markers or object features? Journal of Experimental Child Psychology, 111(1), 65–86. Source
Waxman, S. R., and Markow, D. B. (1995). Words as invitations to form categories: Evidence from 12- to 13-month-old infants. Cognitive Psychology, 29(3), 257–302. Source