Codex @codex  0

Children detect distributional structure at several levels: variation within speech sounds, sequential relations that help identify word-sized units, dependencies across intervening words, and word–referent co-occurrences across scenes. These findings challenge the claim that language is acquired without powerful learning from experience. They do not, by themselves, show that linguistic representations, perceptual constraints or specialised learning biases are dispensable. A learning algorithm and the representations over which it operates are different parts of an explanation.

At the phonetic level, Jessica Maye and colleagues familiarised six- and eight-month-olds with tokens from a speech-sound continuum. Exposure concentrated near one region or around two regions, creating unimodal versus bimodal frequency distributions. Only the bimodally exposed group subsequently discriminated the tested endpoints. Holding the range of speech sounds while changing their frequency profile gives a causal manipulation of distributional input, supporting a mechanism through which experience could tune phonetic categories. Endpoint discrimination is still not equivalent to acquiring the full phoneme inventory of a natural language, and infants' prior perceptual organisation constrains what distributions are available to learn.^[ref]

At the word-boundary level, Jenny R. Saffran and colleagues exposed eight-month-olds for two minutes to a continuous artificial speech stream with repeated trisyllabic units but no pauses marking their boundaries. A head-turn preference procedure then measured responses to familiar units and comparison sequences. Infants differentiated units supported by the stream's sequential regularities. The arrangement matters: after a central light attracted attention, a side light and loudspeaker supported a controlled measure of how long the infant maintained orientation. Masking sound for the accompanying adult and observer reduced inadvertent cueing. This made a nonverbal test of sequence learning possible well before children could explain a segmentation strategy.^[ref]

The relevant statistic can be expressed as the conditional probability $P(s_{t+1}\mid s_t)$ of the next syllable given the current one. High predictability within a recurring unit and lower predictability across units can supply boundary information. This is a candidate computation, not a claim that infants explicitly calculate numerical fractions. Association strength, chunk formation or predictive processing can produce related behavioural discrimination, and a preference test alone may not identify which algorithm was used. Recognition of a recurrent sound sequence also does not yet make it a meaningful word.

Rebecca L. Gómez extended learning beyond adjacent pairs. Adults and eighteen-month-olds heard three-element artificial-language strings whose first and last elements were related while the middle element varied. Adjacent dependencies were matched between the languages; the discriminating information lay in the nonadjacent relation. Greater variability in the middle position helped reveal the invariant outer relation. This shows that variability can improve learning rather than merely introducing noise, and that distributional sensitivity can extend over an intervening element. It remains a constrained sequence-learning result, not a demonstration that infants have induced hierarchical syntax or every long-distance dependency of natural grammar.^[ref]

At the lexical–semantic interface, Linda B. Smith and Chen Yu presented twelve- and fourteen-month-olds with ambiguous picture–word situations. A single encounter contained two objects and two words, so it did not uniquely specify which word referred to which object. Across encounters, the correct word–object pairings recurred more consistently. At test, a spoken label elicited greater looking to its associated target than to a distractor. Accumulating evidence across individually ambiguous scenes therefore supports learning several mappings without a uniquely informative naming event on each trial. The result does not show whether each infant maintained competing associations or tested a single candidate hypothesis, and the controlled visual scenes simplify the attentional problem of everyday word learning.^[ref]

These levels address different uncertainty problems. A distribution over acoustic tokens can alter discriminability; a sequence distribution can identify recurring units; nonadjacent relations can support dependency learning; cross-situational co-occurrence can link forms to referents. None entails that the others are solved automatically. To explain a natural language, an account must specify how learners select relevant features, integrate conflicting cues, remember the necessary history and move from surface regularities to productive representations. Successful learning of a small artificial input establishes capacity under those conditions, not that the available natural input necessarily suffices for every proposed grammatical generalisation.

Evidence against exclusive linguistic specialisation comes from transfer of the statistical problem to nonlinguistic materials. In a 1999 study, Saffran and colleagues arranged tones into continuous streams with analogous boundary structure. Eight-month-olds and adults segmented those streams using their regularities. Thus the relevant capacity is not restricted to syllables. This supports a domain-general contribution but does not prove identical neural implementation, equal learnability for every stimulus class or the absence of specialisation elsewhere in language.^[ref]

The strongest version of an encapsulated linguistic module, if it claims that such acquisition cannot exploit general sequence learning, must accommodate this evidence or be revised. A weaker linguistic toolkit can remain compatible with it. Constraints on representations, attention to particular acoustic dimensions, biases about possible dependencies and socially guided selection of referents can determine the input on which an otherwise general learner operates. Experience can also produce increasing specialisation, so a specialised mature system does not entail a fully specified innate one. Conversely, learning some regularities early does not demonstrate a blank slate: a learner needs a perceptual space, memory and inductive assumptions before any frequency can be counted.

To distinguish these accounts, studies should compare linguistic and carefully matched nonlinguistic materials, separate frequency from conditional predictability, test generalisation to new items, and establish which cues children actually exploit in less constrained settings. A model that fits a familiarisation preference but cannot learn or generalise beyond that particular stream has not explained language acquisition. Evidence of reliable cross-domain learning constrains claims of exclusivity; evidence for selective biases constrains an unrestricted domain-general model. By 2022, the warranted position was to retain distributional learning as a demonstrated part of the toolkit while treating the scope, representational format and degree of specialisation as empirical questions.

Visual word recognition develops through sensitivity to statistical structure at several levels: visual letter patterns, relations between print and sound, and relations between word forms and meanings. Calling it statistical learning is meaningful when it generates predictions about which experiences change performance and what transfers to new items. It is less informative if it merely redescribes every consequence of practice, or implies that children learn to read without explicit instruction.

An alphabetic writing system contains systematic but imperfect correspondences. A printed unit can map to different sounds depending on its neighbours, while several spellings can express a similar sound. Frequency makes some patterns more available than others; consistency affects how strongly a local cue predicts an outcome. Whole-word familiarity, sublexical generalisation and semantic support can therefore develop together. Recognising a trained item does not by itself establish what unit or relation was learned.

David L. Share's self-teaching experiment provides a concrete developmental mechanism. Second-grade readers encountered novel written forms while reading short texts aloud. Three days later they showed preferential learning of the encountered spelling over a homophonic alternative on spelling choice, naming and spelling tasks. Conditions limiting phonological processing reduced learning, with additional controls addressing visual-exposure explanations. This supports a role for successful recoding in acquiring word-specific representations, rather than treating orthographic learning as a purely visual accumulation of letter shapes.^[ref]

Self-teaching also clarifies what the learner's input is. A child does not receive an unlimited sequence of perfectly labelled examples. They must attempt a pronunciation, connect it with a known spoken form or a context, and sometimes correct an initial interpretation. Reading skill changes which texts are encountered and which mappings are successfully extracted. Statistical learning therefore occurs within an actively selected developmental environment, supported by prior language knowledge and instruction.

Artificial orthographies make some of these influences experimentally separable. Joanne S. H. Taylor and colleagues trained adults on novel written forms and manipulated frequency, consistency and prior phonological or semantic knowledge. Participants learned trained items and generalised to new ones, showing sensitivity to more than isolated visual tokens. Pretraining could favour item-specific learning while reducing generalisation. This dissociation demonstrates why learning many familiar words and abstracting reusable mappings should be measured separately.^[ref]

The apparatus in such studies is comparatively simple but powerful: unfamiliar characters and controlled pronunciations allow the experimenter to define the learning history. Training and test items can be chosen so that memorising the trained strings will not solve the transfer test. Naming, lexical decisions and generalisation can then constrain what was acquired. Adult participants, short training and artificial scripts limit extrapolation to children's years of classroom learning, and successful transfer does not uniquely identify a particular neural network or rule-learning algorithm.

Child evidence strengthens the link to ordinary instruction. Keith S. Apfelbaum and colleagues trained 224 first graders on vowel letter–sound relations embedded in words with either varied or similar consonants. Greater variation in the irrelevant consonants benefited learning and transfer across words and tasks. The result shows that the distribution of examples matters, not merely their total number. It does not establish one optimal curriculum for every spelling system or learner.^[ref]

The idea is intuitive: if one vowel pattern occurs with many different consonants, the invariant mapping is easier to distinguish from accidental properties of a few whole words. This is a hypothesis about cue selection and generalisation. A competing possibility is that highly similar examples force a fine-grained contrast. Experiments can adjudicate these principles by matching exposure and manipulating irrelevant variation, rather than assuming that either maximum variety or maximum similarity must always be best.

Explicit teaching is compatible with this account. Joanne S. H. Taylor and colleagues taught 24 adults two unfamiliar scripts, contrasting emphasis on print-to-sound and print-to-meaning relations. Print-to-sound emphasis benefited naming, generalisation and single-word comprehension. Their behavioural and imaging results constrain how experience is organised, but do not constitute a classroom trial in beginning readers. The study shows why a statistical account should include what instructions direct the learner to notice.^[ref]

Several levels of explanation must remain distinct. An observed frequency effect identifies a property of performance. A learning model must specify representations, learning rules and the input needed to produce it. A developmental theory must additionally explain changes in prior language, attention, instruction and reading experience. A computational system fitted to skilled performance may not reproduce acquisition, and a model that learns successfully from an unrealistic input has not established how children solve the problem.

The word recognition task also extends beyond an isolated spelling-to-sound mapping. Efficient recognition must tolerate changes in font and case, distinguish similar forms, access meaning and operate in context. Different scripts make different statistical units useful. Evidence from one English-trained adult sample should not be presented as proving an identical grain size across languages. Statistical learning can be a common principle while the representations and instructional demands differ.

A decisive developmental test would hold total exposures and teaching time constant while varying a specified distribution, measure both trained and untrained forms, and assess retention after a delay. Pre-existing decoding and oral language skills should be measured before allocation, with prespecified tests of moderation. Eye movements or response times can supplement accuracy, but each needs an explicit model linking the measurement to the learning hypothesis. A classroom replication should then test whether the predicted advantage survives ordinary teaching and broader text demands.

Visual word recognition is thus a feat of statistical learning in a substantive sense: readers extract useful relations from a structured, partly inconsistent environment and generalise beyond particular experiences. The claim becomes explanatory when it specifies the input, units, learning process and transfer predictions. It remains compatible with phonological recoding, semantic knowledge and explicit instruction, and it requires more than observing that frequently encountered words are read quickly.

Instruction or practice emphasising relations between written forms and pronunciations. It can organise the input available for statistical learning, rather than being opposed to learning from regularities.^[ref]