Word-Learning Biases¶
A family of default assumptions that help children narrow the possible meanings of novel words, including whole-object, taxonomic, mutual-exclusivity, noun-category, and shape biases.
Core Idea¶
A new word could logically name an object, part, color, material, action, owner, or countless other relations. Word-learning biases are defeasible expectations that make this indeterminacy manageable. Whole-object, taxonomic, mutual-exclusivity, noun-category, and shape biases rank some interpretations above others before exhaustive evidence is available.
These biases are not rigid innate dictionary rules. Syntax, social cues, known vocabulary, function, and repeated evidence can override them, and researchers dispute how language-specific each preference is. Their explanatory value lies in showing how learners narrow a hypothesis space quickly while remaining able to revise a provisional mapping.
Structural Signature¶
Sig role-phrases:
- novel word — creates a mapping problem with many logically possible referents It is essential. Counterfactual: A fully known word does not require a new lexical constraint.
- candidate meanings — supply objects, parts, properties, actions, and categories that could be named It is essential. Counterfactual: Without ambiguity there is nothing for a bias to narrow.
- default bias — ranks candidates using whole-object, taxonomic, exclusivity, noun, or shape expectations It is essential. Counterfactual: Removing the ranking mechanism reduces learning to unconstrained association.
- linguistic and situational cues — can reinforce or override a default interpretation It is essential. Counterfactual: Treating biases as mandatory rules contradicts evidence of contextual flexibility.
- mapping outcome — assigns and later extends the word to an inferred referent or category It is essential. Counterfactual: Attention without a lexical mapping is not word learning.
- developmental learner — brings vocabulary, cognition, and experience that shape bias strength It is diagnostic. Counterfactual: Adult-like interpretation cannot be assumed across age and learning history.
What It Is Not¶
- They are not guarantees that every child selects the same referent.
- They are not explicit grammatical rules taught to the learner.
- They are not one single mechanism; several differently triggered preferences form the family.
- They are not merely visual preferences unless they influence word-to-meaning mapping.
- Closest near-miss. General attentional salience can resemble shape or whole-object preference but must affect lexical mapping, not just looking behavior.
Scope of Application¶
- Novel noun mapping. Whole-object and mutual-exclusivity expectations narrow initial referents.
- Category extension. Taxonomic and shape cues guide which new instances share a learned name.
- Developmental comparison. Age, vocabulary, and experience reveal changes in bias strength and interaction.
- Cue competition. Experiments test how syntax, social intention, function, and perceptual similarity override defaults.
Clarity¶
Name the specific bias, learner age and vocabulary, novel word class, candidate referents, cue structure, dependent measure, and competing explanation. A choice consistent with a bias is not diagnostic if the alternative was unfamiliar, less salient, or grammatically impossible. Defaults should be described probabilistically rather than as categorical rules.
Manages Complexity¶
Biases reduce an effectively unlimited referential space to a tractable set of hypotheses. Researchers can study one constraint at a time, but actual learning combines several cues whose weights change with development. The framework manages this complexity by separating candidate generation, default ranking, contextual override, and later extension.
Abstract Reasoning¶
- Specify the novel expression and all plausible referents available to the learner.
- Identify which whole-object, taxonomic, exclusivity, noun, or shape expectation predicts the choice.
- Control salience, familiarity, syntax, and speaker cues that could independently determine the response.
- Measure mapping or extension rather than attention alone.
- Introduce counterevidence to test whether the preference is defeasible.
- Compare age, vocabulary, and nonlinguistic performance before assigning domain specificity.
Knowledge Transfer¶
The framework transfers among languages and learning tasks only after accounting for their lexical and grammatical cues. A general decision heuristic is not a word-learning bias unless it structures mapping between linguistic forms and meanings. The portable cargo is hypothesis-space reduction under referential uncertainty; the weight and even presence of each bias stop at the learner and language studied.
Examples¶
Applied / In Practice¶
A child hears 'dax' while shown a novel truck and maps it to the vehicle rather than its wheel or color.
Mapped back: constraint → The whole-object default ranks the complete object among many logically possible referents..
Applied / In Practice¶
Given a familiar cup and unfamiliar tool, a child maps a new noun to the tool.
Mapped back: contrast → The known label makes a second meaning for the cup less preferred, though context could override it..
Applied / In Practice¶
After learning a novel solid object's name, a child extends it to another object with similar shape.
Mapped back: category cue → Shape provides a default basis for extending some count-noun categories..
Structural Tensions¶
T1 — Fast Constraint versus Contextual Flexibility. Strong defaults reduce ambiguity but must yield to syntax, speaker intention, function, and explicit contrast.
Diagnostic: Test preference strength under competing cues rather than coding a bias as an inviolable rule.
T2 — Language-Specific Principle versus General Cognition. Observed mappings may arise from lexical knowledge, perceptual organization, attention, or broader category learning.
Diagnostic: Use matched nonlinguistic tasks and developmental comparisons to locate the source of the effect.
Structural–Framed Character¶
The family has a structural role in constraining mappings but is developmentally and linguistically framed. Experimental preferences are probabilistic and can arise from interacting cognitive systems. The abstraction organizes evidence without deciding that every named bias is innate or language-specific.
Structural Core vs. Domain Accent¶
The skeleton is default ranking under underdetermination followed by revisable learning. Developmental psycholinguistics supplies novel words, referents, child learners, lexical categories, social cues, and extension tests. Without linguistic mapping the same logic becomes general inductive bias.
Instantiates / Related Primes¶
-
Approved root. The frozen graph contains no validated parent edge for the bias family.
-
Related — mutual exclusivity, shape bias, and whole-object assumption. These are member constraints rather than names for the entire family.
Neighborhood in Abstraction Space¶
Word-Learning Biases sits in a crowded region of the domain-specific corpus (26th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.
Family — Perception, Narrative & Moral Cognition (13 abstractions)
Nearest neighbors
- Lexical Hypothesis — 0.91
- Paraphrase — 0.90
- Medical Subject Headings — 0.90
- Type Error — 0.90
- Absolute Metaphor — 0.88
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
- Inductive bias. Tell: The broader idea of preferential hypotheses across learning problems.
- Mutual exclusivity. Tell: One bias favoring distinct labels, not the complete set.
- Joint attention. Tell: A social cue to reference that can interact with but is not itself a lexical bias.
- Vocabulary knowledge. Tell: Existing learned mappings can mimic or modulate bias-driven choices.
References¶
- Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Word_learning_biases (revision 1315238321).
- Preserved source candidate: https://linkinghub.elsevier.com/retrieve/pii/0022096592900274
- Preserved source candidate: https://linkinghub.elsevier.com/retrieve/pii/088520149290004B
The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.