Year 3 of the psychology (experimental) course of the University of Oxford (Part B; terms 7-9)

Advanced options, practical training and optional project or dissertation combinations; BA exit after three years.

Part II advanced option in the held 2022 papers. The source examination is shared by the psychology (experimental) course of the University of Oxford and the psychology, philosophy and linguistics course of the University of Oxford.

Solutions to all six options in this undergraduate Part II paper. The cover identifies the Second Public Examination; the Part II classification follows the audited paper record.

Source: Released original paper, archive member a13359w1.pdf.

Phonology and syntax usually show a left-hemisphere bias, whereas semantic processing depends on a more bilateral system with task-dependent asymmetries. None is confined absolutely to one cerebral hemisphere, and lateralisation of an experimental contrast is not equivalent to localisation of a representation. Phonology concerns structured sublexical units, semantics concerns meaning, and syntax concerns relations that organise words into sentences. These processes interact, so tasks must isolate the relevant contrast rather than merely require language.

Historical lesion observations motivated left language dominance, but their anatomical interpretation needs care. Modern MRI of Paul Broca's preserved cases showed damage extending beyond the conventional cortical region named after him. Severe impaired speech after a left lesion therefore supports lateralisation without proving that a small superficial patch alone constitutes the speech system. Large vascular lesions can also damage connections between surviving regions.^[ref]

For phonological processing, Robert J. Zatorre and colleagues used PET while participants judged different attributes of the same speech material. Phonetic discrimination increased left inferior-frontal activity, whereas pitch discrimination recruited right prefrontal regions; acoustically matched speech engaged secondary auditory cortices bilaterally. Holding the signal relatively constant while changing the decision supports functional differentiation beyond a contrast between unrelated stimuli. It does not show that all phonological representations reside in inferior frontal cortex: articulatory recoding, attention and decision demands can contribute to the phonetic contrast.^[ref]

The distinction matters because speech perception begins with bilateral auditory analysis, whereas selecting, sequencing or rehearsing phonological forms can have a stronger left bias. Impairment of repetition after damage to a dorsal language pathway and comprehension after other damage constrain different transformations. A deficit in producing a phoneme is not automatically a deficit in recognising it. Familiarity with a sound pattern and its linguistic role can also alter the task, so acoustic speed or frequency alone cannot define phonology.

Semantic processing illustrates why a leftward verbal-task result can conceal bilateral representational resources. In Gorana Pobric's rTMS experiment, perturbing either temporal pole slowed semantic associations for words and pictures while a perceptual comparison was spared; stimulation of a control site did not produce the same semantic effect. This supports a contribution from both anterior temporal lobes across input formats. It does not require identical response profiles throughout both lobes or imply that semantic retrieval and selection are entirely bilateral.^[ref]

Verbal naming places additional demands on lexical retrieval and speech production. Consequently, stronger left activity in naming than in passive picture viewing can reflect access and output as well as conceptual knowledge. Conversely, right-sided involvement in nonverbal meaning does not establish a separate nonlinguistic semantic store. Comparisons should separate semantic association from perceptual similarity and control both response requirements and difficulty. Lesion and perturbation results help adjudicate whether an activated region contributes to the task.

For syntax, lesion studies and functional imaging converge on a predominantly left frontotemporal network, rather than a single grammar centre. Evelina Fedorenko's individual-localisation study showed language-selective and domain-general regions next to one another within the anatomical territory called Broca's area. Averaging across people can merge them. Thus, an inferior-frontal response to a hard sentence may reflect language processing, domain-general control or both; independent functional localizers and nonlinguistic controls improve interpretation.^[ref]

Corianne Rogalsky's lesion study compared sentence-picture matching and plausibility judgments in 66 patients. The distribution of syntactic-comprehension impairment implicated a wider left-hemisphere system and challenged a simple equation between Broca's area damage and agrammatic comprehension. Convergence across tasks is useful because picture matching also demands visual analysis and response selection, while plausibility can recruit semantic expectations. Neither behavioural label makes the task syntactically pure.^[ref]

Sentence comprehension in Japanese and Japanese Sign Language also showed left-dominant frontal and temporoparietal responses across sensory conditions. This directly addresses hemispheric comparisons, although comprehension versus nonword detection is not a pure assay of syntax.^[ref]

A quantitative laterality index can summarise an explicitly defined response. For nonnegative left and right measurements $L,R$ with $L+R>0$,It ranges from $-1$ to $1$, but depends on baseline, threshold and region selection. Responses $(20,10)$ and $(10,5)$ have the same index $1/3$ despite different total responses. If left activity falls from $20$ to $10$ while right activity remains $10$, the index becomes zero without any right recruitment. These are illustrative values, not imaging observations. Negative contrasts need a separately justified convention.

The most convincing evidence therefore combines within-person process contrasts, unilateral perturbations or lesions, and converging measurements of behaviour and neural responses. It also describes individual variation instead of assuming that every participant has the population-average pattern. The justified conclusion is graded functional asymmetry within interacting networks: strongest left biases for many phonological and syntactic operations, bilateral conceptual contributions, and substantial dependence on task and representation level.

Basal ganglia abnormalities plausibly contribute to some developmental speech and language disorders, especially through motor sequencing and learning, but the evidence does not establish a single basal-ganglia cause across diagnoses. A causal account must specify the disorder, the affected computation and how an alteration in a circuit produces that impairment. Childhood apraxia of speech, developmental language disorder and stuttering are different phenotypes; difficulty with speech output does not by itself identify a grammatical deficit.

The basal ganglia participate in loops with cerebral cortex and thalamus. Such circuits could support selecting, initiating and learning sequences, including the rapid articulatory transitions needed for speech. A plausible mechanism is impaired acquisition or reliable initiation of a learned motor sequence, rather than storage of every word or syntactic rule in the striatum. This gives testable predictions about skill learning, nonword repetition and sensitivity to external timing cues. General plausibility, however, does not establish which circuit is responsible in an individual child.

The KE family supplied a particularly informative genetic case. Cecilia S. L. Lai and colleagues identified an FOXP2 alteration segregating with severe speech and language impairment, together with an independent case in which a chromosome translocation disrupted the same gene. The phenotype included major problems in coordinated orofacial sequencing alongside broader language difficulties. This strengthens a causal connection between the gene and development of the phenotype. It does not isolate the anatomical route: a transcription factor can alter several developing systems, and the family is not a representative sample of every developmental language disorder.^[ref]

Kate Watkins's MRI study compared affected family members, unaffected relatives and controls. Voxel-based morphometry identified abnormalities in multiple motor- and speech-related regions; targeted volumetry confirmed reduced caudate volume bilaterally. Relationships with oral praxis and nonword repetition made the caudate result functionally relevant. The unaffected relatives provide a valuable comparison, but an anatomical correlation cannot establish that caudate reduction alone mediates the gene's effects. Multiple abnormalities and correlated behavioural measures make that inference underdetermined.^[ref]

Animal experiments move beyond an association between a human scan and a symptom. Matthias Groszer and colleagues studied mice with a corresponding alteration of the FOXP2 orthologue. Heterozygotes had relatively preserved baseline motor abilities but impaired motor-skill learning, with abnormal plasticity in striatal and cerebellar circuits. The distinction between baseline performance and improvement with practice supports a learning-related mechanism. Yet a mouse motor task is not human syntax or learned speech, and the cerebellar findings themselves argue against an exclusively basal-ganglia interpretation.^[ref]

Stuttering offers a partially independent line of evidence. Soo-Eun Chang's study combined resting-state fMRI and probabilistic tractography in children aged 3–9. Differences involved basal-ganglia–thalamocortical and auditory–motor networks, including weaker connectivity between putamen and supplementary motor regions. Studying young children reduces the extent to which decades of compensatory speaking strategies can explain the result. Nevertheless, the measurements were taken after symptoms began. Correlated activity at rest is not direct synaptic coupling, and tractography does not demonstrate the direction or necessity of a physiological connection.^[ref]

Improved fluency with an external pace would be compatible with compensation for impaired internal timing, but it would not uniquely identify a basal-ganglia abnormality. External cues also change attention, anticipation, auditory feedback and task structure. Similarly, a group difference in a procedural-learning task can result from poor motor execution, working memory or comprehension of instructions. The critical comparison separates learning rate from initial accuracy and uses nonverbal tasks, matched feedback and retained skill after a delay.

Several causal alternatives remain. A shared developmental influence may alter both caudate structure and cortical language function. Reduced language practice may itself change network measurements. A cortical or white-matter abnormality may produce downstream striatal differences. An experimentally supported gene-to-phenotype relationship does not automatically identify every intermediate step in a proposed gene-to-basal-ganglia-to-language chain.

A stronger programme would combine prospective measurements before or close to onset, repeated follow-up of persistence and recovery, detailed speech versus language phenotyping, and independent motor-learning measures. Analyses should control age, hearing, motion, general motor ability and broader development, and assess multiple network regions. Selective perturbations in appropriate models can test mechanisms, while their behavioural scope must remain explicit. The strongest claim is that distributed corticostriatal learning circuits are one important contributor in some phenotypes; neither a universal basal-ganglia lesion model nor a language-specific function of FOXP2 follows from the available evidence.

Primary progressive aphasia has helped dissociate syntactic impairment from loss of word meaning, phonological memory and motor speech, and has implicated a left frontotemporal network and its dorsal connections. Its contribution is substantial, but degeneration is distributed and evolving, so it does not yield a one-region map of syntax. PPA is a syndrome in which progressive language impairment is initially the principal clinical problem; the variant label should follow a detailed assessment rather than substitute for one.

In Maria Luisa Gorno-Tempini's 31-patient study, the nonfluent group combined speech apraxia and complex-syntax difficulties with left inferior-frontal and insular atrophy. Semantic impairment was associated with anterior temporal damage, while a logopenic pattern involved left posterior temporal and inferior parietal regions. Comparing variants showed that slow speech, reduced sentence comprehension and impaired naming need not share a single cause. The study established useful contrasts, not perfectly isolated syndromes or a unique pathology for each language deficit.^[ref]

The central methodological gain is to test syntax rather than infer it from fluency. A person who cannot execute a well-planned utterance may still comprehend grammatical relations. Conversely, fluent production may conceal impaired thematic-role assignment. Compare short, reversible sentences whose meaning cannot be guessed from ordinary plausibility, vary grammatical construction while matching word knowledge, and measure both comprehension and production. Long sentences alone are insufficient because phonological working memory and attention also determine performance.

For example, an illustrative active/passive pair can reverse which noun is the agent while preserving familiar words. A sentence-picture matching task can offer pictures with the same participants but reversed roles. Failure on the noncanonical construction after single-word comprehension is verified is more informative about syntactic dependencies than failure to repeat a long unfamiliar sentence. These proposed controls do not make visual choice, memory demands or response selection disappear; they make the alternatives measurable.

Stephen M. Wilson's 2010 fMRI study examined eight patients with nonfluent variant primary progressive aphasia during sentence comprehension. Unlike controls, patients did not show the usual complex-versus-simple modulation in atrophic posterior inferior-frontal cortex, while a mid-superior-temporal region retained complexity modulation. A more anterior frontal region was recruited, but that recruitment did not support successful syntactic processing. This separates the presence of activation from normal computational responsiveness. The small, heterogeneous sample and altered structure limit inferences from a missing contrast.^[ref]

The subsequent tract study added a network constraint. In 27 patients, damage measured with diffusion imaging in the left superior longitudinal/arcuate system was associated with impaired syntactic comprehension and production, beyond the cortical-atrophy contribution included in the analysis. Comparable associations were not found for the ventral tracts examined. This supports the importance of dorsal frontotemporal communication under those tasks. It does not show that ventral connections never contribute to sentence meaning or that a diffusion-derived tract is exclusively syntactic.^[ref]

These findings help distinguish a network-disconnection account from a narrowly local account: surviving frontal cortex may fail because temporal inputs or long-range interactions are disrupted. They also motivate looking for separable grammatical computations rather than a single complexity variable. Hierarchical dependencies, maintaining a noun's features, assigning roles and selecting an overt response need not have identical neural requirements.

Converging evidence outside PPA is necessary. Corianne Rogalsky's stroke lesion work relates comprehension to a distributed left system, and her concurrent-articulation experiments show how working-memory demands can contribute to inferior-frontal syntactic contrasts. These observations caution against treating frontal disease, reduced rehearsal and damaged syntax as interchangeable. They do not erase the PPA findings; they sharpen the processes the imaging contrast must explain.^[ref][ref]

PPA also has characteristic limitations. Atrophy covaries across connected areas and worsens with disease severity; a regional correlation can reflect a broader disease gradient. Clinical groups may be defined partly by the impairment under investigation, so rediscovering that difference is not independent confirmation. Grey-matter volume and fractional anisotropy are indirect measurements, while altered perfusion or tissue loss affects the observed fMRI signal. Statistical adjustment for one anatomical variable cannot guarantee complete separation of these causes.

Longitudinal within-person assessments, repeated controls, individual functional localisation and analyses of both structural and functional connections would strengthen the account. A decisive interpretation should predict which construction becomes impaired, which simpler operations remain intact and how the deficit changes with progression. PPA has made syntax a tractable network-level target by dissociating components and revealing dorsal connectivity requirements; it supports process-specific, convergent explanations rather than a diagnostic label or isolated grammar centre.

There is no secure, universal bilingual advantage in domain-general executive function. There are important benefits of knowing and using two languages, some positive task findings and suggestive associations with cognitive reserve, but these are different claims with different evidential requirements. Bilingualism varies with proficiency, acquisition age, frequency of use and the situations in which languages are switched. A binary label can obscure the experience most relevant to a proposed mechanism.

The executive-control hypothesis begins with competition: using one language while another is available may repeatedly exercise selection and interference control. Transfer to nonlinguistic tasks is plausible, but it is a far-transfer claim, not an automatic consequence of speaking two languages. The account should predict which control operation improves and under which switching experiences, rather than count every faster response as evidence of inhibition.

Ellen Bialystok's three studies of middle-aged and older adults reported smaller Simon interference costs in bilingual participants, with stronger effects in older groups and demanding conditions. In the Simon task, the required response depends on a relevant stimulus feature while an irrelevant spatial location can favour the competing response. Comparing congruent and incongruent trials operationalises conflict. These positive results support a conditional association; nonrandom language histories leave socioeconomic, educational and other selection explanations to be tested.^[ref]

The relevant estimand must be stated. Let $C_g={\overline{RT}}_{g,\mathrm{incongruent}}-{\overline{RT}}_{g,\mathrm{congruent}}$ for group $g$. A conflict advantage requires $C_{\mathrm{bilingual}}<C_{\mathrm{monolingual}}$, which is a group-by-congruency interaction. Illustratively, monolingual means of $(500,550)$ ms and bilingual means of $(450,500)$ ms for congruent/incongruent trials give equal costs of $50$ ms. The bilingual group is faster overall, but this example supplies no selective conflict advantage. Accuracy and speed-accuracy trade-offs must also be examined. These numbers are a calculation, not observed study data.

Negative studies are therefore informative rather than disposable exceptions. Kenneth R. Paap and colleagues tested 15 executive-processing indicators across three studies, including Simon, flanker, antisaccade and switching tasks, and did not find a coherent bilingual benefit. Inconsistent cross-task correlations also challenged treating every task's interference score as a stable measure of one general control faculty. This does not prove an exactly zero effect for every bilingual population, but it contradicts a broad, easily expressed advantage.^[ref]

Child data also constrain generality. Eneko Antón's study used the Attention Network Test in bilingual and monolingual children with careful attention to age and socioeconomic variables, and found no bilingual advantage in the principal attentional measures. A reliable conflict effect within participants verifies that a task manipulation worked; it does not guarantee reliable individual differences or a sensitive group contrast. Measurement reliability and adequate statistical power matter to both positive and negative interpretations.^[ref]

Selective reporting can make the literature appear more consistent than the completed research. Angela de Bruin and colleagues followed conference abstracts into publication and found that results supporting an advantage were more likely to be published than null or opposing results. This is direct evidence about the research pipeline. It does not invalidate every positive experiment or by itself estimate the true average effect. Preregistration, shared outcomes and publication of null results address the identified bias.^[ref]

A different question concerns later clinical symptoms. Suvarna Alladi's retrospective clinic series contained 648 people with dementia, including 391 bilingual participants, and reported symptom onset about 4.5 years later in the bilingual group after considering several covariates. The result is compatible with cognitive reserve. It is not evidence that bilingualism reduced dementia incidence: all participants already had dementia, and referral, survival, recalled onset and unmeasured life experiences can affect the comparison. Later symptoms also do not necessarily imply less underlying pathology.^[ref]

A stronger test would measure language experience continuously, recruit groups from the same communities, prespecify a small set of well-measured outcomes and analyse multiple tasks with an explicit measurement model. Longitudinal studies with baseline cognitive performance reduce selection ambiguity; randomised language-learning interventions with active controls test narrower training effects, though short courses do not reproduce lifelong bilingual experience.

Finally, linguistic communication and cultural access are substantive benefits regardless of executive-task results. Metalinguistic awareness and language-specific proficiency are closer to the practised skill than general inhibition, and should be evaluated as separate outcomes. Bilingualism is valuable, but the evidence warrants specific, experience- and task-dependent claims rather than a universal cognitive enhancement or a promise of dementia prevention.

Sign languages recruit much of the left-dominant frontotemporal language network while also engaging modality-dependent visual, parietal and motor systems. The evidence supports shared organisation across modalities more strongly than it proves that the same neural populations contain fully amodal phonological, semantic and syntactic codes. Natural sign languages have conventional sublexical structure and grammar; they are not simply gestures illustrating spoken words.

Brain-lesion evidence provides an important starting point. In Gregory Hickok's study, 19 lifelong signers with unilateral lesions completed single-sign and simple/complex sentence comprehension tasks. Left temporal involvement was associated with substantial impairment across tasks; right-lesion and other groups performed well on the simpler conditions, with more difficulty on complex material. The result argues against assigning language to the right hemisphere merely because signs are visual and spatial. Complex-command performance can also reflect memory, spatial processing or general comprehension demands.^[ref]

Functional imaging identifies both common and distinct components. Karen Emmorey and colleagues compared sign and speech within native hearing ASL–English bilinguals using PET for picture naming and fMRI for sentence comprehension. Shared language-related responses coexisted with greater parietal recruitment for signing and different temporal responses for speech. The within-person comparison reduces confounding between hearing status and language modality. It also makes clear that motor output, visual movement and auditory feedback contribute to the total activation pattern.^[ref]

An amodal representation is stronger than a common anatomical location: it should capture relevant information independently of how that information enters or leaves the system. A shared inferior-frontal mean response can arise from common control demands, different intermingled populations or the same computation over differently formatted representations. This is the distinction expressed by regional activation does not identify a neural code. Conjunction of significant maps is useful evidence for shared territory, but cannot by itself establish identity of the represented information.

For phonology, the relevant units need not be sounds. A sign can be distinguished by handshape, location and movement, just as speech forms can differ in sublexical features. Mairéad MacSweeney's study compared BSL location judgments with English rhyme judgments using picture-based tasks. A partly shared left frontoparietal network was modulated by language, hearing status and age of first acquisition. Thus sublexical form analysis can share network resources across modalities, while the differences argue against an entirely invariant implementation. Delayed access to a first language must be separated from deafness itself.^[ref]

A stronger phonological test would match response selection and difficulty, contrast legal with carefully controlled sublexical combinations, and include nonlinguistic hand-movement judgments. Even successful cross-modal decoding would need interpretation: sign handshape and spoken sound have no automatic one-to-one physical mapping. An abstract combinatorial relation could be shared without every modality-specific feature having an amodal counterpart.

For semantics, signs and spoken words can access the same conceptual relations while having different forms. Emmorey's within-bilingual comprehension experiment used detection of semantic anomalies and identified overlapping frontotemporal responses with modality-dependent components. This supports common meaning-related processing, although anomaly detection also requires monitoring and a decision. A more specific test would train a decoder on one modality and test concept identity in the other, holding out stimuli and separating meaning from the picture, response or decision used to measure it.^[ref]

For syntax, Kuniyoshi L. Sakai's comparison of Japanese Sign Language and Japanese sentences found common left-dominant frontal and temporoparietal recruitment for comprehension. Comparing sentence comprehension with sentential nonword detection improved specificity beyond passive sign viewing. Different participant groups and potentially different demands nevertheless prevent inferring identical neuronal syntax codes. Spatial referent tracking, prosody and grammatical relations may draw on partly overlapping resources rather than a single modality-free operation.^[ref]

The best design combines native Deaf signers, hearing native signers and hearing nonsigners where appropriate, with careful measurement of first-language access and proficiency. Nonlinguistic biological motion, matched visual complexity, meaningful versus meaningless sequences and matched response demands help identify what each contrast removes. Individual functional localisation, lesion convergence and cross-modal multivariate tests are stronger together than any activation map alone. A failed decoder does not establish separate representations if measurement noise or anatomical alignment prevents generalisation.

Right-hemisphere activity after left-hemisphere damage can be compensatory, inefficiently helpful, maladaptive or incidental. Its role depends on the region, task, residual left network and recovery stage; a group activation map cannot decide among those possibilities. Functional compensation means that the recruitment supports performance in the damaged system. It need not restore normal performance or reproduce the computation formerly performed on the left.

Longitudinal evidence immediately challenges a uniform interpretation. Dorothee Saur and colleagues repeatedly imaged auditory comprehension in 14 patients after left middle-cerebral-artery stroke. Initially weak activation was followed by increased bilateral activity with a prominent right inferior-frontal homologue response, and later by a shift towards left language areas alongside further recovery. The trajectory makes time after stroke central to interpretation. Correlation between activation and improvement is compatible with a contribution to recovery, but can also reflect restored capacity to engage with the task.^[ref]

A decrease in left dominance is particularly ambiguous. The laterality index $(L-R)/(L+R)$ can fall because left activity falls, without right activity increasing. Acute ischemia, diaschisis and altered neurovascular coupling also change the fMRI signal. Matching the task instruction is insufficient if one patient comprehends the sentences and another only attempts them. Accuracy, response time and an independent physiological assessment should therefore accompany the activation contrast.

A maladaptive account proposes that increased activity in a right homologue interferes with useful remaining left function or promotes an inefficient strategy. Interhemispheric inhibition is one candidate mechanism, but cannot be assumed from the fact that the two cerebral hemispheres change differently. Greater right activation among more severely impaired patients could mean that right activity causes impairment; equally, larger left lesions may both worsen performance and make right support more necessary. Statistical adjustment must account for lesion location and disconnection, not simply lesion volume.

Perturbation supplies a more direct test. Lutz Winhuisen and colleagues used PET to select inferior-frontal targets and then disrupted their activity during a semantic task in 11 patients examined within two weeks of stroke. Right-sided stimulation impaired performance in a subset of patients with right IFG activation. This supports a necessary contribution to residual performance in those individuals, even though their language remained less effective than in patients relying more on recovered left function. The result is not evidence that every activated right region is beneficial.^[ref]

Evidence sometimes presented for maladaptation also needs qualification. Margaret A. Naeser's open study applied repeated right inferior-frontal stimulation to four people with chronic aphasia and reported later picture-naming improvement. Suppression-associated improvement is compatible with a counterproductive component of right activity. Without a randomised sham comparator, however, practice, expectancy and nonspecific intervention effects remain possible; the result is preliminary, not a general treatment rule. An anterior target in chronic aphasia also need not have the same function as a different inferior-frontal site early after stroke.^[ref]

A proposed mechanistic study would recruit a sufficiently characterised stroke cohort across prespecified stages, quantify left cortical and tract damage, and measure baseline naming, comprehension and connected speech. Use individual structural imaging and independent functional localisation to identify a right language-related target and a control target. Because an activated voxel can contain mixed functions, site selection should use a relevant contrast, not just the maximum response to any difficult task.

Within each participant, counterbalance a transient perturbation condition with credible sham and an active control, use masked assessment, parallel stimulus sets and a prespecified primary language outcome. Include nonlinguistic reaction and motor tasks to detect general distraction or response slowing. Session spacing should address carryover, and the analysis should model repeated measurements and interactions with stage and residual left-network integrity. Safety screening and consent are integral to a clinical stimulation study; a sample size should be justified from variability and the smallest informative effect, not invented from the exam prompt.

If perturbing the right target worsens language selectively relative to both controls, that supports a currently compensatory role. If it improves language selectively and reproducibly, that supports a maladaptive contribution under those conditions. No behavioural change is inconclusive unless the perturbation was effective and the outcome sensitive. The intervention could spread through the network, so neither sign of effect automatically identifies a local code or proves inhibition of the left hemisphere.

Finally, a treatment trial asks a different question from transient necessity. Randomise a longer intervention against sham with matched speech-language therapy, assess untreated items and everyday communication, and follow durability. Repeated imaging can test whether behavioural improvement accompanies a predicted network change, while avoiding circular selection of patients or outcomes on the same data. The evidence favours conditional, stage- and site-specific right-hemisphere contributions; perturbation with appropriate controls, followed by a separate clinical trial, can distinguish useful functional compensation from harmful recruitment more convincingly than activation alone.

Part II advanced option in the held 2022 papers. The source examination is shared by the psychology (experimental) course of the University of Oxford and the psychology, philosophy and linguistics course of the University of Oxford.

Solutions to all six essay options in the sensing and perceiving light, colour and materials course of the University of Oxford examination of 26 May 2022. Both printed parts of question 5 have separate solutions; its combined answer and each other essay are below 1,800 words. The scientific evidence used predates the examination. Source release: released examination archive, member a16861w1.pdf. Original figures illustrate stated mathematical models and do not reproduce examination diagrams or participant data.

Post-receptoral processing compares cone signals, organizes those comparisons across space and time, and uses them to represent surfaces and objects. Three receptor classes do not imply three neurons labelled with three color names. A complete answer must distinguish the information available at the receptors from its neural transformation and its perceptual interpretation.

The principle of univariance supplies the starting constraint. For one cone class, an absorption-dependent response does not identify wavelength independently of intensity. Comparing differently sensitive classes makes chromatic information available. Relative to an adapted background, cone contrasts can be written $c_j=(Q_j-Q_{j0})/Q_{j0}$, where $Q_{j0}>0$ is the background catch. This representation separates a change relative to the prevailing input from the raw number of absorbed photons. It is a useful coordinate system, not a claim that every neuron implements this exact formula.

Retinal circuitry immediately transforms the signals. Light hyperpolarizes cone cells and changes their transmitter release. ON and OFF bipolar pathways respond differently to changes in cone output, while horizontal and amacrine circuitry contribute spatial and temporal interactions. Retinal ganglion cells transmit spike trains rather than simply forwarding a cone's voltage. Their center–surround organization compares regions, so chromatic and spatial structure are already intertwined. Near the central retina, the midget pathway can oppose an L- or M-dominated center to a surround containing other cone input. The same circuitry can carry information useful for fine spatial structure and for chromatic contrast; these are not necessarily separate wires.

A distinct S-cone route illustrates the anatomical specificity. Dennis M. Dacey and colleagues combined physiological characterization with identification of ganglion-cell morphology in primate retina, finding that the S-ON opponent response belonged to a small bistratified ganglion-cell type rather than simply another midget cell.^[ref] This supports parallel circuits with different cone combinations. Calling the pathway “blue-on” is shorthand for its stimulus response, not proof that the cell itself produces a unique blue experience.

At a simplified computational level, an achromatic-like combination and two chromatic combinations can be represented asThis invertible three-channel opponent transform reorganizes three coordinates rather than creating additional independent spectral information. Their matrix has determinant $-2$, so the transform is invertible: $c_L=(y_0+y_1)/2$, $c_M=(y_0-y_1)/2$ and $c_S=y_2+y_0/2$. Actual weights depend on cell, location, adaptation and stimulus conditions. This algebra is an illustrative opponent transform, not a fitted retinal wiring diagram or a complete perceptual color space.

The lateral geniculate nucleus preserves several parallel retinal pathways while also participating in cortical interactions. A. M. Derrington and colleagues studied macaque geniculate cells with gratings whose luminance and chromatic modulation could be varied independently on a calibrated color television display. Finding directions that changed cone stimulation without driving a cell constrained its linear cone weights.^[ref] For a nonzero linear weight vector, the zero-response contrasts obey $r={\mathbf{w}}^{\mathsf{T}}\mathbf{c}=0$, defining a chromatic null plane. A null response therefore tests a relation between inputs, rather than merely assigning the cell a preferred color name. The resulting opponent directions need not coincide with human unique hues.

Cortical processing adds more complex spatial comparisons. Bevil R. Conway mapped cone inputs to color-responsive neurons in alert macaque V1 using cone-isolating spots and eye-position-corrected reverse correlation. Some cells combined chromatic opponency with opposite tuning in adjacent spatial subregions.^[ref] Such double-opponent cells can respond strongly to a chromatic boundary while responding less to a spatially uniform change. This offers a mechanism useful for separating surface differences from shared illumination changes. It does not establish that one V1 population alone solves color constancy under arbitrary scenes.

Further processing must relate color to form, illumination, material and the current task. The same cone contrasts can be interpreted differently in a shadow, highlight or translucent object. Surface color judgments, discrimination thresholds and color naming therefore constrain different computations. Feedback and attention can influence which information is selected, while learning can alter how a signal is interpreted or normalized. A lesion or imaging response can constrain contributions to those tasks, but an area responsive to color is not automatically an isolated “color center.”

The strongest picture is consequently a hierarchy of interacting computations: cone-specific signals, opponent and spatial filtering, combinations tuned to richer image structure, and object-level judgments. Trichromatic sampling limits spectral information; later processing determines how that information supports boundaries, stable surface appearance and behavior.

Sensitivity adjustments occur at several stages and over several timescales. They reveal adaptive gain control and flexible reference points, but no single aftereffect uniquely identifies one anatomical site. Changes in threshold, apparent hue, an achromatic setting and a receptor current must be distinguished before explaining them with one adaptation mechanism.

At the receptor level, a finite photocurrent range has to represent environments with very different mean light levels. Julie L. Schnapf and colleagues measured membrane currents from individual macaque cone outer segments. Responses to a maintained light step had an initial component followed by a smaller steady response, and a steady background reduced sensitivity to an added weak flash. This desensitization was distinguishable from loss of pigment through bleaching.^[ref] Direct recording establishes that some gain adjustment occurs before any color name, decision or cortical object representation. The preparation is macaque tissue, and a measured cone current is not identical to a human appearance judgment.

An idealized normalization model illustrates the benefit. If a mechanism's incremental gain is proportional to $1/(I+I_0)$, maintaining a fixed response increment requires $\Delta I\propto I+I_0$. At high backgrounds this gives approximately constant fractional contrast $\Delta I/I$, a Weber-like regime. It should not be confused with a universal law across all intensities. Under a known-background Poisson counting model with mean $N$, the noise standard deviation is $\sqrt{N}$; a small increment has signal-to-noise ratio approximately $\Delta N/\sqrt{N}$. A fixed signal-to-noise criterion then gives fractional threshold proportional to $N^{-1/2}$ instead. The two predictions derive from different assumptions about what limits performance.

Mean adaptation and contrast adaptation also differ. A diagonal cone-gain model replaces $Q_j$ by $g_j{Q}_j$, with gains related to recent stimulation. It can reduce the effect of a persistent chromatic shift in the illuminant. If the signal changes approximately as $Q_j^{\prime }=k_j{Q}_j$, choosing $g_j^{\prime }=g_j/k_j$ cancels that change. This is exact only for the assumed channelwise scaling; real spectral illuminant changes need not transform every surface's catches by the same three factors.

To test adjustments beyond mean cone excitation, experiments can modulate color around an unchanged time average. John Krauskopf and colleagues found direction-selective threshold elevation after temporal chromatic or luminance modulation, supporting separable adaptable mechanisms for particular directions near threshold.^[ref] Michael A. Webster and colleagues then measured appearance after adaptation to modulation along different color-space directions. A color monitor, a modulated adapting field and asymmetric appearance matches allowed testing effects beyond detection thresholds. Appearance shifted away from the adapting direction, including directions intermediate to the simplest cardinal color directions.^[ref] Thus a three-independent-channel account sufficient for some threshold results was not sufficient for all suprathreshold appearance changes. Constant mean catches constrain a simple mean-receptor explanation, but do not prove that all receptor nonlinearities have been removed.

Longer experience changes reference points too. Jay Neitz and colleagues altered chromatic experience with colored filters over days and measured unique yellow settings. Shifts persisted for one to two weeks after filter use ended.^[ref] Persistence after removing the immediate adapting stimulus demonstrates more than an instantaneous receptor response. It supports experience-dependent normalization without showing that a color category or verbal label has been rewritten. Any interocular transfer must also be measured explicitly; duration alone does not identify cortex.

A natural perturbation extends the range to months. Peter B. Delahunt and colleagues followed achromatic settings before and after removal of cataractous lenses. Removal greatly changed short-wavelength input; settings continued to readjust over roughly three months, whereas chromatic discrimination recovered much faster.^[ref] An achromatic setting compensates for a perceived tint, so its physical shift should not be mistaken for a same-direction tint in appearance. The findings indicate distinct adjustment timescales. They are observational evidence with a changing optical input, not a randomized isolation of one neural site or a general treatment recommendation.

These studies support a multilevel experimental strategy. Measure receptor responses, detection thresholds and appearance separately; manipulate mean and contrast independently; test spatial specificity, interocular transfer and recovery. Control exposure, eye position, stimulus calibration, response criterion and learning. A loss of sensitivity can reflect adaptation that preserves dynamic range rather than damage, and stable appearance can coexist with changed receptor inputs. Conversely, perfect compensation is not guaranteed.

Photoreceptors determine which optical differences are available for vision, with what noise and spatial–temporal precision. They strongly constrain light, color and material perception without uniquely determining the eventual percept. Later processing can exploit context and action, but cannot recover arbitrary information already lost in receptor sampling.

For cone class $j$, a simplified expected quantum catch over one sampling interval iswhere $N$ is the incident photon-number spectrum at the receptor, including ocular filtering, and $a_j$ effective absorption probability. An energy spectrum requires conversion by photon energy $hc/\lambda$ before this counting interpretation. The integral is many-to-one: equal catches do not imply equal spectra. The principle of univariance means that an individual receptor does not independently report photon wavelength and number.

Multiple receptor classes provide comparisons, but ordinary cone-mediated trichromacy still samples a spectral distribution with only three class responses at a specified location and time. In a discretized model, $\mathbf{Q}=A\mathbf{n}$ with three rows and more than three spectral bins. The matrix has a nonzero kernel. Around a strictly positive spectrum, a sufficiently small perturbation in that kernel preserves nonnegative intensities while leaving all three catches unchanged. This derives the possibility of metamers; it does not mean all real-world spectral differences are unobservable under every condition. Altering illumination, observer or spatial–temporal context can break a particular match.

Sensitivity also has a photon budget. In an ideal Poisson model, mean catch $N$ has variance $N$, so relative counting uncertainty is $1/\sqrt{N}$. Gathering more photons reduces this uncertainty but requires a larger effective collecting area or more integration time. Temporal averaging can obscure rapid change, while spatial pooling can sacrifice fine detail. Receptor and downstream noise add limitations beyond shot noise. Rods support low-light detection with different sensitivity and pooling from cone cells; if only one receptor class provides usable signals, wavelength–intensity ambiguity cannot be resolved by comparisons within that class. Mesopic vision, where rods and cone cells can both contribute, is not captured by declaring all dim-light vision achromatic.

The receptor array is also uneven. Christine A. Curcio and colleagues mapped rods and cone cells in whole-mounted human retinas using microscopic counts. Cone density was highest centrally and fell strongly with eccentricity, while the rod distribution differed and the central foveal region was rod-free.^[ref] Thus acuity and sensitivity should not be treated as uniform over the visual field. Optical blur and downstream sampling constrain performance alongside receptor spacing. The S-cone mosaic is sparser than the combined L/M mosaic, so equal spatial detail cannot simply be assumed for all chromatic signals.

A regular one-dimensional model of photoreceptor sampling makes one limitation explicit. Sampling a sinusoid at spacing $d$ gives identical samples for frequencies $f$ and $f+k/d$, because the added phase at sample $n$ is $2\pi kn$. Restricting a real signal to frequencies below $1/(2d)$ avoids this basic aliasing ambiguity. Real retinal mosaics are two-dimensional, irregular and optically filtered, so this is a benchmark rather than a literal universal retinal cutoff. Fine material texture can consequently disappear or alias even when coarser color and shape remain visible.

Individual variation changes the sampling system itself. Heidi Hofer and colleagues' adaptive-optics measurements showed large L:M cone-ratio variation among observers passing normal color tests.^[ref] Gabriele Jordan and colleagues found behavioral evidence consistent with an extra chromatic dimension in one of a selected set of carriers with potential extra pigment diversity, rather than in every carrier.^[ref] These findings separate receptor availability from downstream use: receptor counts and pigment variants constrain possibilities without directly specifying appearance or functional dimensionality.

Material perception adds an inverse problem. For a matte surface under a single simplified illuminant, the reflected photon spectrum is proportional to $E(\lambda )R(\lambda )$, so catches mix illumination and surface reflectance. The same catches can therefore arise from different surfaces, even before considering three-channel spectral compression. Specular reflection, subsurface scattering, translucency and fluorescence introduce further dependencies on direction, geometry or wavelength conversion. Three cone values cannot independently label every physical material property.

Nevertheless, humans can exploit cues across space, movement and time. Highlights change with viewpoint; texture and shading constrain shape; surrounding surfaces provide illumination evidence. Such information can reduce ambiguity under useful assumptions, while metamerism and sampling limits remain. Receptor properties define an information bottleneck and a set of trade-offs. Successful material recognition depends on combining what survives that bottleneck with contextual constraints, not on receptors directly measuring “wood,” “metal” or intrinsic surface color.

L, M and S describe the relative spectral sensitivities of cone classes. Calling them red, green and blue confuses receptor sampling with the color appearance constructed from comparisons and context. The shorthand also wrongly suggests three narrow, nonoverlapping detectors or the primaries of a display.

First, each class has a broad sensitivity curve. A long-wavelength-sensitive cone responds to many wavelengths, and its sensitivity maximum is not a unique red stimulus. L- and M-cone sensitivities overlap extensively, while S cone cells are relatively more sensitive at shorter wavelengths. Light at one wavelength can therefore excite more than one class. Andrew Stockman and colleagues derived L- and M-cone sensitivity estimates using color-matching data and measurements from observers of known pigment genotype. Their analysis emphasizes broad overlap and the need for isolation procedures to measure one class without contamination from others.^[ref] Exact curves depend on pigment variant and optical filtering, so the letters denote relative classes rather than three invariant color labels.

Second, the principle of univariance prevents one cone from identifying hue. For a monochromatic stimulus, let mean absorption be $Q=Ia(\lambda )$. Whenever $a({\lambda }_1)$ and $a({\lambda }_2)$ are positive, intensities satisfying $I_2=I_{1}a({\lambda }_1)/a({\lambda }_2)$ give the same $Q$. The single receptor cannot distinguish those two inputs on absorption count alone. Calling that receptor a red detector would assign a color judgment to a signal that cannot even uniquely distinguish the two wavelengths. It remains true that spectral sensitivity constrains the comparison; univariance does not mean all receptor types are identical.

Third, perceptual color requires combinations. Comparing L and M signals and comparing S with L+M supplies chromatic information; summing appropriately weighted signals contributes to luminance-related processing. An L-cone increment can be part of several stimulus changes depending on what happens to M and S and on the adapted background. Neither an L response nor a raw L:M ratio by itself specifies a context-independent color word. The retinal and cortical opponent circuitry demonstrated by Dennis M. Dacey and Bevil R. Conway provides mechanisms for comparisons, rather than a relabelling of three cone outputs as three sensations.^[ref][ref]

Fourth, “blue” or “red” need not identify a single wavelength. Spectrally different lights can be metamers, and a purple appearance can arise from a mixture rather than a monochromatic wavelength. Conversely, the same physical spectrum can look different against another surround or after adaptation. Cone labels cannot replace the distinction between emitted or reflected spectrum, receptor catch, opponent representation and reported hue.

Fifth, the display analogy is backwards. An RGB screen emits mixtures from three selected primary spectra. The cone cells receive and integrate the resulting spectrum; their response curves are not identical to those primary spectra. If primary intensities are $p_k$ and cone sensitivities to each primary form matrix $B$, then $\mathbf{Q}=B\mathbf{p}$. Off-diagonal terms are generally nonzero because each primary can excite several cone classes. If $B$ is nonsingular, an intended catch change can be computed through $B^{-1}$ within the display's feasible gamut. It does not follow that every requested catch vector is realizable with nonnegative bounded primary intensities. Calibration and observer differences therefore matter in cone-isolating experiments.

Finally, color terms themselves vary across languages and tasks, whereas receptor class is a biological classification. Receptor genes, color matches, discrimination and linguistic categorization are related but distinct measurements. The labels L, M and S prevent an unwarranted assumption that one anatomical cell class directly instantiates one learned word or unique subjective hue.

The research challenge is to retain causal control while measuring appearance under the coupled spectral, spatial and temporal variations of actual scenes. Simply replacing a flat patch with a photograph does not guarantee that either the stimulus or the task is ecologically informative.

A display's RGB values do not specify the spectrum reaching a participant's cone cells. Primary spectra, viewing geometry, calibration, luminance range and the observer's lens and pigment sensitivities matter. A photograph can clip highlights, alter white balance and compress dynamic range; a display cannot necessarily reproduce the original spectrum, brightness or angular light field. A researcher studying spectral constancy should measure radiance and illuminants or specify the limits of a display simulation, rather than treating an uncalibrated image as physical ground truth.

Real materials also vary in more than diffuse reflectance. Shape, roughness, specularity, translucency and surrounding illumination jointly determine an image. Changing one rendering parameter can unintentionally change highlight area, contrast and apparent shape. Conversely, holding an image statistic constant can produce an unnatural stimulus. Experiments need independent manipulation of candidate causes and a clear statement of which physical scene properties are represented by the apparatus.

James M. Kraft and David H. Brainard demonstrated a productive compromise with a chamber of actual surfaces. They independently controlled illumination and surface reflectances and used a projection colorimeter to let observers adjust a test patch to appear achromatic. Manipulations could hold candidate image statistics approximately fixed while changing the illumination. The results rejected explanations based solely on local adaptation, the image mean or the most intense region.^[ref] The achievement was not realism alone: it was enough realism combined with manipulations that made competing mechanisms disagree. These results do not show that local adaptation never contributes.

For material appearance, Roland W. Fleming and colleagues used rendered objects illuminated by natural illumination distributions and systematically altered those distributions. Reflectance matching was more reliable under realistic illumination statistics than under some altered statistics.^[ref] This makes illumination an experimental variable, not merely a nuisance background. It also warns that a model tested only under a small set of studio lights can exploit cues unavailable or misleading elsewhere.

The dependent variable needs equal care. Selecting the same paint, matching apparent color, adjusting an achromatic point, detecting an illuminant change and naming a material do not measure the same outcome. A person may correctly infer that two patches are the same material while still noticing an appearance difference. Instructions, response format, adaptation time and available context can alter the task. Eye and head movements change retinal exposure, so fixation constraints should be chosen deliberately and their cost to natural viewing acknowledged.

A strong program therefore combines calibrated simplified stimuli, controlled real scenes, validated rendering and tests on previously unseen objects and illumination. It should sample more than one observer and record relevant visual variation, report both accuracy and uncertainty, and test model generalization across scenes rather than only across random frames of the same scene. Proposed interventions can vary geometry while preserving spectral input, or alter illumination while suppressing a particular cue. Their outcomes must be measured, not assumed. The goal is interpretable generalization: experimental control sufficient to distinguish mechanisms and scene diversity sufficient to test whether those mechanisms remain useful.

The visual system must infer relatively stable object properties from images jointly determined by illumination, shape, material and viewpoint. The problem is intrinsically ambiguous; robust performance exploits several constraints and tolerates uncertainty rather than perfectly recovering a unique physical scene from every image.

Even a diffuse model contains a surface-illuminant ambiguity. Ignoring a fixed geometric factor, $N(\lambda )=E(\lambda )R(\lambda )$. For any positive function $g$ that leaves the resulting reflectance physically admissible, define $E^{\prime }=Eg$ and $R^{\prime }=R/g$. Then $E^{\prime }{R}^{\prime }=ER$, so even an observer measuring the complete outgoing spectrum could not distinguish those two explanations from that surface alone. Three cone catches introduce further spectral metamerism. Thus perfect color constancy cannot be obtained by a universal local inversion of three numbers.

For directional optical reflection, an image sample instead depends on an integral such asThe expression sums contributions from incident directions, weighted by the bidirectional reflectance distribution function and projected area. It assumes a local, nonfluorescent optical reflection model and omits emission. A glossy surface therefore reflects aspects of its environment, and moving the observer changes the sampled outgoing direction even if the material is unchanged. Subsurface transport and transmission require further terms or models.

The system can constrain illumination by comparing surfaces, spatial boundaries and changes over time, while using shape and scene layout to distinguish a shadow from a pigment boundary. Cone-gain adjustment can approximately compensate shared spectral changes but cannot solve every mixture of differently colored lights, interreflections and specular components. A spatial average is unreliable if a scene is genuinely dominated by one colored material. A bright region might be a lamp, a highlight or a high-reflectance object, so “assume the brightest point is white” is not a universal solution.

Material judgments require distinguishing diffuse shading from highlights and transparency or translucency cues. Isamu Motoyoshi and colleagues showed that manipulating image statistics, including luminance-distribution skewness, could influence apparent gloss and lightness in their surface-image conditions.^[ref] Such regularities offer efficient cues. They cannot uniquely specify a material: a histogram-preserving spatial permutation changes spatial organization while retaining the luminance histogram and all its moments. That mathematical ambiguity motivates experiments separating histogram cues from geometric consistency; it does not, by itself, predict every appearance judgment after scrambling.

Juno Kim and colleagues directly tested the relation between highlights and diffuse shading. Manipulating their positional and orientation congruence reduced perceived gloss, including when simple proximity to broad luminance maxima was controlled.^[ref] The visual system therefore uses relationships among image structures, not merely the presence of a few bright pixels. The conclusion concerns the tested images and judgments, rather than a claim that one cue explains every shiny or transparent material.

Temporal information and active viewing help further. A highlight can move differently from a surface marking; multiple views constrain geometry, while familiar object structure and expected illumination supply priors. These strategies can fail under unusual lighting or deliberately misleading images. The system must also decide which differences matter: recognizing a material category, judging gloss and estimating spectral reflectance are related but separable goals.

Language is not necessary for the basic capacity to discriminate or organize colors, but it can influence color judgments, attention and memory in particular tasks. The evidence supports neither the claim that words create all color experience nor the claim that language is irrelevant whenever the stimuli are visible.

The starting distinctions are essential. A receptor sensitivity, a color-matching relation, a discrimination threshold, a category boundary and a color name are different measurements. Molecular differences can change available cone signals, while linguistic experience can change how available information is grouped or used. Finding different words for two samples does not show that one group cannot discriminate them. Finding the same matching relation does not establish identical subjective appearance or identical naming.

The biological basis imposes constraints independent of a person's vocabulary. Three cone classes provide overlapping spectral measurements, and later circuits compare those measurements. Altered receptor signals or adaptation can change performance without requiring a new word. Conversely, changing a response strategy can change performance while leaving receptor input intact. The central empirical question is therefore which stage and task language influences, not whether one can name a color without knowing its name.

Prelinguistic evidence rules out a strict requirement for color words. Alice E. Skelton and colleagues used infants' recognition-memory responses to sample relations among colors across the hue circle. Infants' categorical distinctions showed organization related to common divisions in adult color lexicons and to early visual mechanisms.^[ref] Infants did not need to produce the relevant color terms to show those distinctions. The inference is not that newborns possess an immutable adult lexicon: the infants already had visual experience, and looking-time measures require controls for discriminability, novelty and attention. The title's biological interpretation should not be strengthened into proof that experience plays no role.

Cross-language studies also challenge a rigid universal partition. Debi Roberson and colleagues compared English speakers with Berinmo speakers using naming, recognition and category-learning tasks. Effects of linguistic category boundaries appeared for the categories used by each group, and response-bias controls weakened a simple claim that universally focal colors always enjoy a memory advantage.^[ref] The article's historical title is retained as a citation, not adopted as a description of people. The comparison shows that category use and memory cannot be reduced to one vocabulary-independent set of preferred labels. Because language group covaries with cultural and visual experience, it does not isolate vocabulary as the sole cause.

A more direct manipulation was used by Jonathan Winawer and colleagues. Russian has conventional basic terms distinguishing lighter and darker blues that English usually groups under “blue.” Participants saw a target square and two alternatives together and selected the matching alternative. Russian speakers showed a speed advantage for comparisons crossing their linguistic boundary. A concurrent verbal task removed that advantage, while a spatial interference task did not produce the same pattern.^[ref] The simultaneous display reduces the long-term color-memory explanation, and the interference manipulation implicates access to language-related processes during performance. It does not establish a new receptor limit, a complete inability to discriminate within a category, or a necessary change in private color qualia. Concurrent tasks can affect more than one processing stage.

These results motivate separating sensitivity from response facilitation. A faster match can reflect attentional selection, category-assisted comparison or decision formation. To claim changed sensory resolution, an experiment should measure psychometric functions over controlled stimulus separations, fit location, slope and lapse parameters, and test whether any effect persists with nonverbal responses and reversed response mappings. Equal physical distance in a standard color space does not automatically ensure equal discriminability for every participant; individual thresholds and adaptation should be measured where they matter.

A stronger causal program can combine within-person interference, language-learning interventions and longitudinal or bilingual comparisons, while matching exposure and task difficulty. Testing transfer to new samples distinguishes memorized stimulus labels from a changed category rule. Comparing immediate discrimination, delayed recognition, naming and appearance matches identifies whether an effect generalizes across measurement levels. These are proposed strategies rather than invented experimental outcomes. No test based only on verbal descriptions can directly establish equality or inequality of another observer's subjective experience.

The seemingly opposing findings can therefore coexist. Biology and early experience can produce color organization before fluent naming, while learned labels subsequently guide the use of that organization. Color vision does not depend on language for its existence; linguistic resources can nevertheless alter how people attend to, classify, remember and rapidly compare colors under specified conditions.

Part II advanced option in the held 2022 papers. The source examination is shared by the psychology (experimental) course of the University of Oxford and the psychology, philosophy and linguistics course of the University of Oxford.

Reading and language: development and disorder, Trinity Term 2022, 1 June at 09:30 BST. This undergraduate Part II advanced option has a three-hour writing period; the cover identifies the Second Public Examination without printing a Part number. Every essay option is treated below using evidence available by 1 June 2022.

Course: Reading and language: development and disorder course of the University of Oxford. Public source: Oxford-supplied archive, a10986w1.pdf.^[ref]

Reading comprehension is measurable as performance on specified tasks and as a theoretically organised construct, but it is not a context-free quantity revealed equally well by every test. A score reflects a reader, a text, a question and an assessment format. The important question is whether the intended interpretation is supported: screening for difficulty, explaining an individual profile and evaluating an intervention can require different measurements.

A useful framework is the simple view of reading, which distinguishes decoding from linguistic comprehension. Wesley A. Hoover and Philip B. Gough tested predictions from a product relation in a longitudinal bilingual sample. The model captures the idea that understanding spoken language cannot compensate for completely inaccessible print, while accurate word reading alone cannot supply the meanings of the words. It does not imply that all raw test scores can simply be multiplied, or that each broad component is psychologically indivisible.^[ref]

For an illustration, suppose decoding $D$ and linguistic comprehension $C$ are represented on idealised scales from zero to one and $R=DC$. Profiles $(D,C)=(0.3,0.9)$ and $(0.9,0.3)$ both give $R=0.27$. The same comprehension outcome can therefore arise from different bottlenecks. These are hypothetical model values, not diagnostic cutoffs or observations about children. The distinction motivates measuring component skills rather than assigning one intervention from an aggregate score alone.

Direct comparisons show that tests are not interchangeable. Janice M. Keenan and colleagues compared five reading comprehension measures. Decoding explained more variation in the PIAT and Woodcock–Johnson passage task, whereas listening comprehension contributed more strongly to the Gray Oral Reading Test and the two Qualitative Reading Inventory measures. The demands of some tests also changed with developmental level. Modest agreement therefore need not mean that comprehension is unreal: it can reveal different mixtures of processes and content, alongside measurement error.^[ref]

Assessment design determines what evidence is available. A short cloze item may be solved through local word knowledge and decoding, while questions about a passage can require integrating information across sentences. Retelling additionally requires organising and producing an answer. Removing the text increases memory demands; keeping it visible permits rereading and strategic search. Oral answers may expose expressive-language difficulties that a recognition format does not. None is automatically superior: the format must match the intended inference.

Reliability and validity are distinct. Repeated passages and sufficiently sampled items can improve precision, but a highly consistent score can still overemphasise decoding when the goal is discourse understanding. Convergence across texts and formats strengthens a construct interpretation; disagreement can identify task-specific demands. Measurement invariance matters when comparing ages or groups, because the same item format can depend on different skills as word recognition improves. A change in a standard score is not automatically an equal change in the underlying process at every age.

The first pressure point is efficient access to printed words. Slow or inaccurate decoding limits which propositions become available and can consume resources needed for integration. Assessment should distinguish familiar-word accuracy, unfamiliar-word decoding and fluency rather than assume that one word-reading score captures all three. A reader can identify isolated words accurately yet struggle to coordinate them in connected text.

The second pressure point is linguistic knowledge. Vocabulary breadth and depth determine whether word meanings are accessible and appropriate to context; grammar helps assign relations between entities and events. A text can be decoded accurately while remaining opaque because its meanings or sentence structures are unfamiliar. Background knowledge can support interpretation, but can also produce confident answers unsupported by the passage. Text selection should therefore vary topics rather than make one knowledge domain stand for general comprehension.

Higher-level processes provide further pressure points: generating necessary inferences, building a coherent representation, identifying relevant information and noticing inconsistencies. Kate Cain and colleagues assessed children at eight, nine and eleven and related comprehension to working memory and component skills. The study supports a multicomponent analysis, while its concurrent prediction models do not establish that training any correlated component will improve reading. Working memory demands are also partly determined by the text and response format.^[ref]

A useful assessment combines a comprehension measure with listening comprehension, vocabulary, grammar, decoding and fluency, then investigates the child's actual errors. Can they explain a referent, connect two statements or notice a contradiction when word-reading demands are reduced? Contrasting written and spoken versions can test a proposed bottleneck, although listening itself has memory and attention demands. A profile is a hypothesis about performance, not proof of a separate disorder or a unique cause.

Intervention supplies complementary evidence. Paula J. Clarke and colleagues randomised children with accurate reading but poor comprehension to text-focused, oral language, combined or untreated conditions. Comprehension improved in the intervention groups, with continued gains after oral language training. This demonstrates that difficulties can be modified through more than word decoding, but an improvement in a package does not isolate one underlying process. Maintenance, untrained texts and independent assessment are important when distinguishing learning the assessment from broader transfer.^[ref]

The defensible conclusion is that comprehension is a measurable but task-dependent achievement supported by interacting skills. Its pressure points can lie in print access, language knowledge, integration, inference or monitoring, and their relative importance changes with the reader and text. Good measurement makes those dependencies explicit and samples enough situations to support the intended decision, rather than treating a single score as a complete explanation of failure.

Longitudinal data are critical for several developmental questions because they show whether an earlier characteristic precedes later reading, how individuals change and whether apparently similar outcomes have different histories. They are not sufficient to identify causes, and they are not the only informative evidence. The strongest understanding combines repeated observation with experiments, intervention and explicit models of measurement and development.

A cross-sectional age comparison confounds age with cohort, instruction and selection. Older and younger children may have received different teaching or differ in who remains in a sample. Following the same children can establish continuity and change, show whether an early weakness persists, and locate periods when a skill becomes more predictive. It also permits assessment before conventional reading instruction, reducing the risk of describing a consequence of reading experience as its original cause.

Valerie Muter and colleagues followed ninety school entrants over two years. Earlier letter knowledge and phoneme sensitivity predicted word recognition, whereas comprehension drew on word recognition, vocabulary and grammar. The temporally ordered pattern supports distinguishing developmental pathways rather than using a single readiness score. It does not prove that a predictor is sufficient, immutable or uniquely causal: teaching, family experience and other correlated skills may contribute to both the early measure and later outcome.^[ref]

Developmental relations can change as constraints shift. When decoding is effortful, individual differences in print access may dominate a comprehension measure; later, oral language differences can become more visible. In Arne Lervåg and colleagues' five-year study beginning at about seven and a half, shared variation across vocabulary, grammar, verbal working memory and inference predicted listening and reading comprehension. Their modelling included interactions and nonlinearities. This shows why a fixed linear contribution estimated at one age may not describe the developmental process.^[ref]

Repeated observation can distinguish trajectories concealed by an endpoint. Two children with similar reading at nine may differ because one began with weak decoding and improved, while another decoded adequately but developed a widening comprehension difficulty as texts became linguistically demanding. These histories generate different hypotheses about support and assessment. Conversely, using the later outcome to define groups can bias retrospective descriptions of their predictors, especially when classifications depend on noisy scores.

Longitudinal data are also needed to investigate reciprocal influence. Oral language knowledge can facilitate reading, while reading supplies new words and makes phonological structure more explicit. A path from an earlier reading score to later vocabulary is compatible with feedback, but it can also reflect stable differences in home experience or language ability. A conventional cross-lagged model may mix between-child differences with within-child change. Temporal ordering and statistical adjustment constrain explanations without automatically identifying the effect of changing a particular child's experience.

The home-literacy study by Monique Sénéchal and Jo-Anne LeFevre illustrates both the value and limits. Their five-year sample linked book exposure with language outcomes and direct parental teaching with early literacy, followed by later reading. Measuring activities and skills at different stages helps specify candidate routes that a single concurrent correlation would conceal. The observational design and relatively advantaged sample, however, do not establish that prescribing the measured activity would reproduce the association across all families.^[ref]

Several methodological conditions are essential. The measures must remain interpretable across time: replacing an easy early task with a harder later one can be sensible, but common anchors or defensible latent models are needed for growth comparisons. Ceiling and floor effects can create apparent changes in prediction. A reliable group-average effect does not guarantee reliable ranking of individuals. Multiple measures can reduce task impurity, although a fitted latent factor still depends on its indicator and model assumptions.

Attrition can distort a trajectory if children with greater difficulty or disadvantage are less likely to remain. Recruitment and retention should be reported alongside the handling of missing observations. Statistical methods cannot recover information without assumptions, so sensitivity analyses should examine plausible departures from them. More waves do not automatically solve this problem, and closely spaced assessments can introduce practice effects while widely spaced assessments miss the timing of change.

Intervention provides additional causal leverage. Peter J. Hatcher and colleagues compared matched groups of seven-year-old poor readers receiving reading, phonological work, their integration or a control condition. Phonology-only work most improved the phonological tasks, while the integrated programme produced the greatest reading progress. This demonstrates why a longitudinal predictor cannot simply be translated into an isolated training prescription. The design supports the tested contrast, while allocation, contact and the content of each teaching package constrain finer causal claims.^[ref]

An informative future programme would begin before instruction, repeatedly measure oral language, phonological skills, letter knowledge, teaching and reading, and retain an inclusive range of children. Embedded randomisation could vary a specified teaching component while keeping adequate instruction available to everyone. Assess near-term target engagement, generalisation to untrained material and later comprehension. This combines developmental timing with an intervention contrast instead of treating observational paths as effects of manipulation.

Cross-linguistic and artificial orthography experiments can answer questions that a single natural cohort cannot. They can hold learning opportunities constant while changing the consistency or distribution of print-to-sound mappings. Longitudinal classroom studies then test whether the proposed process survives ordinary instruction and changing language demands. Agreement across methods is more persuasive than extending the duration of one observational design indefinitely.

Longitudinal data are therefore indispensable when the claim concerns ordering, change, persistence or reciprocal development. They are not a substitute for credible measurement or causal design. Their greatest contribution is to make developmental explanations more precise and testable, while experiments and interventions establish which changes in experience actually alter the path.

A meaningful distinction can be made between developmental language disorder and reading disorders, provided it is expressed as a distinction between dimensions and profiles rather than a claim that every child falls into one isolated biological category. Spoken-language difficulty and impaired word reading can occur separately or together. Their overlap is informative, and the definition of 'reading disorder' matters: dyslexia centred on word-level reading is not synonymous with every form of poor reading comprehension.

The CATALISE consensus endorsed developmental language disorder for persistent language difficulties associated with functional impairment and poor prognosis, without a known associated biomedical aetiology. Risk factors do not automatically exclude the diagnosis, and a verbal–nonverbal discrepancy is not required. This is a research and clinical terminology agreement, not an experiment proving a natural boundary. Earlier studies of specific language impairment used criteria that must be examined before their groups are equated with contemporary DLD.^[ref]

Dyslexia typically concerns persistent difficulties in accurate or fluent word reading and spelling. A child can have these difficulties while understanding spoken language relatively well. Another child can decode words adequately but struggle with vocabulary, grammar or discourse understanding, limiting comprehension in both spoken and written form. A third can have both profiles. The simple view of reading provides a useful two-dimensional description, while actual assessment must also consider fluency, instruction, language experience and the demands of the text.

Hugh W. Catts and colleagues tested the relation prospectively in a representative 527-child sample, comparing kindergarten language impairment with later dyslexia classifications. They found overlap but not identity. In a separate profile analysis, dyslexia was more closely associated with pronounced phonological-processing weaknesses, whereas language impairment without dyslexia showed milder phonological difficulties. The evidence supports distinguishable profiles, but depends on the diagnostic rules, measures and developmental occasions used.^[ref]

Margaret J. Snowling and colleagues followed a cohort enriched for familial reading risk or preschool language difficulty. At eight, they identified dyslexia-only, DLD-only and comorbid groups; comprehension was measured at eight and nine. All groups had difficulties relative to controls, but the dyslexia-only group's comprehension weakness was milder than that of the groups with DLD. This is consistent with different routes to a shared outcome. Comorbidity rates in an enriched sample should not be treated as population prevalence.^[ref]

The contrast between studies is itself useful. A representative cohort and a risk-enriched cohort can produce different overlap estimates without one necessarily invalidating the other. Selecting families with early language problems increases the chance of observing both difficulties. Restricting one group to normal decoding or oral language also creates cleaner profiles than may occur in routine assessment. Group definitions can therefore sharpen a theoretical contrast while understating the continuum of mixed presentations.

There are plausible shared and distinct developmental processes. Weak phonological representations or awareness can hinder learning letter–sound relations, while broader lexical, grammatical and discourse weaknesses affect understanding beyond word identification. These processes can covary because language systems interact and because experiences support several skills together. A phonological difficulty does not have to account for every aspect of DLD, and broader language risk does not guarantee dyslexia. A probabilistic multiple-risk account fits overlap more naturally than a necessary-and-sufficient deficit assigned to every child in a label.

Assessment should therefore specify the difficulty rather than infer it from the category. Include unfamiliar-word decoding, familiar-word accuracy and fluency, spelling, vocabulary, grammar, listening comprehension and discourse tasks. Nonword repetition and phoneme-awareness measures can contribute, but each has perceptual, memory, motor or metalinguistic demands. Hearing, opportunity to learn, multilingual experience and functional impact also require consideration. Low performance in the language of schooling is not, by itself, evidence of an intrinsic language disorder.

The distinction matters when choosing an intervention target. A child whose main barrier is acquiring written word forms needs teaching that addresses that barrier; accurate decoding cannot be assumed to resolve a broader comprehension weakness. Paula J. Clarke and colleagues' trial of children with specific comprehension difficulties showed benefit from oral language and text-focused interventions, beyond the children's existing accurate reading. It supports treating the identified language/comprehension problem, while not proving that a diagnostic label uniquely determines the best programme.^[ref]

The combined profile may require support on both dimensions. Improvement should be monitored separately for word reading, oral language and comprehension, because a gain in one can leave the other limiting performance. A programme's effect on a trained task is not enough: transfer to untrained passages and everyday educational participation are relevant outcomes. Categories can help communication and access to services, but should not prevent a child with mixed needs from receiving both forms of support.

A stronger test of separability would recruit children before reading instruction, use reliable repeated measures and examine profiles without selecting only extreme, nonoverlapping groups. Dimensional models can be compared with categorical models, and competing predictions preregistered. An intervention study could test whether targeted decoding and oral language components have different effects across baseline profiles, with adequate power for interactions. Mere significance in one group and nonsignificance in another would not establish a differential response.

The meaningful distinction is consequently between patterns of spoken-language and written-word difficulty, with implications for mechanisms, assessment and support. The evidence rejects a simple equation of DLD with dyslexia and also rejects treating reading comprehension failure as one homogeneous condition. It does not require sharp, immutable boundaries or deny shared risks, comorbidity and developmental change.

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.

By June 2022, important principles of reading acquisition were established, including the need to learn written word forms and to understand the language they express. The major unanswered questions concerned how these skills become efficient and integrated, why children respond differently to experience and instruction, and how experimental gains become sustained comprehension and participation. Useful future research should target those uncertainties with explicit competing predictions rather than reopen every instructional question as if no evidence existed.

One question concerns what is learned from a successful encounter with a new written word. Does the child acquire an item-specific spelling, a reusable mapping, a contextual guess, or some combination? David L. Share's delayed learning experiments motivate the distinction, while artificial orthography work shows that strong item learning and generalisation need not move together. Future studies should therefore distinguish recognition of practised items from transfer to novel combinations and later retention, rather than use one immediate naming test as a complete index of learning.^{[ref] [ref]}

A specific experiment could randomise unfamiliar but pronounceable items to matched exposure schedules varying the consistency of a spelling–sound relation. Hold the number of encounters, visual complexity and oral vocabulary knowledge constant. Assess trained spellings against homophonic alternatives after a delay, and test untrained recombinations that require the relevant mapping. Eye tracking during text reading could establish where the new form attracts attention, while comprehension questions verify meaningful engagement. The experiment should specify whether its prediction concerns initial learning, retention or generalisation.

A second question is how the distribution of teaching examples interacts with the learner's existing knowledge. Keith S. Apfelbaum and colleagues found that varied irrelevant consonants helped first graders learn vowel mappings. It remains necessary to identify when variation helps, whether its effects depend on prior decoding and how long the advantage lasts. A multisite classroom experiment could independently vary relevant consistency and irrelevant variation while matching total teaching time and opportunities to respond.^[ref]

Pretraining measures should be collected before randomisation, and interaction tests powered prospectively. A significant benefit in one subgroup and a nonsignificant effect in another does not demonstrate different treatment effects. Instructional fidelity and actual exposure need observation because a teacher may adapt a nominally identical activity in different ways. Broader spelling, unfamiliar-word reading and later passage comprehension would distinguish a useful instructional principle from a narrow training-task effect.

A third question is how best to coordinate word-level and language-level support. Margaret J. Snowling and colleagues' longitudinal profiles show that dyslexia, DLD and their co-occurrence can lead to comprehension difficulty through different constraints. The unresolved question is not whether either dimension matters, but which combination, timing and intensity of support helps a specified child. Diagnostic categories are too coarse to substitute for a measured profile.^[ref]

A factorial teaching trial could retain adequate core reading instruction for everyone and independently randomise additional decoding and oral language components, with contact and resource use measured. Include children across the skill distribution rather than only extreme, mutually exclusive categories. Prespecify primary outcomes for word reading, language and comprehension, with delayed follow-up. A subsequent adaptive stage could randomise a well-defined change for children not responding, instead of attributing improvement after an uncontrolled adjustment to personalisation itself.

The trial would also address a fourth question: what constitutes meaningful transfer? Janice M. Keenan and colleagues showed that comprehension assessments differ in their reliance on component skills. An intervention can look more successful simply because its outcome task resembles training. Future evaluations need independent passages, several assessment formats and checks that the construct retains a comparable meaning across time and groups. Function in the classroom and the child's experience should accompany test scores, without treating either as a substitute for accurate measurement.^[ref]

A fifth uncertainty concerns how findings generalise across writing systems, language backgrounds and educational environments. Artificial-script work offers experimental control, but Joanne S. H. Taylor and colleagues' instruction study used adults learning small novel vocabularies. It cannot establish the full trajectory of a multilingual child encountering an unfamiliar language of instruction. Coordinated cross-language studies should define the mappings and language experiences that differ, rather than merely compare national averages.^[ref]

Recruitment should include varied socioeconomic circumstances and children whose language development differs from the majority sample. Assessment needs to separate opportunity to learn from persistent difficulty and to examine the validity of the measures in each context. Shared protocols and culturally appropriate materials can coexist. The aim is to test which mechanisms generalise and which depend on script, vocabulary or teaching, not presume that every observed group difference reveals an intrinsic deficit.

Computational models can sharpen all these questions. A model should learn from a specified developmental input, make predictions for held-out items and trajectories, and face comparison with alternatives that use different representations or learning rules. Matching average skilled-reader accuracy is weaker than predicting the effect of an experimental change in input. Parameters should not be allowed to absorb every individual difference without independent constraints; failure should inform revision rather than invite unrestricted refitting.

Finally, the timescale of research must match the educational claim. Short experiments identify candidate processes, longitudinal cohorts reveal ordering and trajectories, and controlled implementation studies assess delivery and lasting benefit. Preregistration, transparent exclusions, appropriate handling of school clustering and missing follow-up, and independent replication make the chain more credible. The central unanswered problem is how to connect the child's particular learning history and profile with an instructionally manipulable process that produces durable, generalisable gains.

Oral language shapes written language development by supplying the phonological, lexical, grammatical and discourse knowledge that print must connect with. The influence is not unidirectional. Learning to read and write can make speech structure more explicit, expand vocabulary and alter the processing of spoken words. The strength and direction of a particular relation depend on age, writing system, instruction and the aspect of language being measured.

At the word level, alphabetic reading links graphemes with speech sounds and familiar spoken forms. Phonological awareness can help a child analyse a word and use letter–sound relations, but it is not identical to speech comprehension or to the ability to pronounce a word. A child may understand a spoken word without being able to segment its phonemes deliberately. Reading instruction and metalinguistic awareness can consequently develop together rather than one being wholly completed before the other starts.

Longitudinal evidence separates some routes. Valerie Muter and colleagues found that letter knowledge and phoneme sensitivity predicted early word recognition, while vocabulary and grammar contributed to comprehension alongside word recognition. The result supports a differentiated oral language foundation. It does not imply that vocabulary is irrelevant to every unfamiliar-word problem or that the observed predictors alone specify a complete teaching programme.^[ref]

For text comprehension, oral vocabulary supplies meanings and grammar supports interpretation of relations and reference. Inferencing and discourse knowledge connect information beyond a sentence. Listening comprehension offers a useful comparison when decoding is suspected to be the bottleneck, but it is not a demand-free measure of language. Speech unfolds over time and can tax attention and memory; written text can be reread. These differences should be considered before interpreting a written–spoken discrepancy as a pure decoding deficit.

Oral language also contributes to producing written language. Choosing words, constructing sentences and organising a narrative draw on language knowledge, while spelling and transcription add constraints. Julie E. Dockrell and colleagues followed 64 children with a history of language impairment, assessing them at eight and their written text at ten. The writing showed difficulties in sentence structure and organisation; concurrent vocabulary and reading helped explain variation. The associations support a broad profile, not proof that an isolated vocabulary treatment would resolve composition difficulties.^[ref]

Intervention gives stronger leverage on one direction. Paula J. Clarke and colleagues' oral language programme improved reading comprehension in children who could already read accurately, with gains extending beyond the intervention period. This supports oral language as a modifiable contributor, while the multicomponent programme and untreated comparator leave its precise ingredients partly open. An implication is that improving word recognition alone may leave a language-level limitation unresolved.^[ref]

Home experiences provide potentially different inputs. Monique Sénéchal and Jo-Anne LeFevre linked book exposure with vocabulary and listening comprehension, and direct parental teaching with early literacy in a five-year study. These routes help explain why 'home literacy' is not one uniform activity. The observational, relatively advantaged sample limits causal and population inference; shared reading can involve both spoken interaction and print exposure, making it difficult to assign its effects exclusively to either medium.^[ref]

Evidence for the reverse direction has a long history. José Morais and colleagues compared adults who remained illiterate with people who had learned to read in adulthood. The former had difficulty deliberately adding or deleting a speech phone, whereas adult readers performed the tasks more readily. The contrast suggests that explicit access to phonemic structure is not an automatic consequence of mature spoken language. Because literacy was not randomly assigned, schooling and selection remain alternative contributors; the result concerns metalinguistic performance rather than absence of ordinary speech perception.^[ref]

Written language also supplies new lexical knowledge. William E. Nagy and colleagues found small but reliable gains in knowledge of unfamiliar words encountered in natural text. Repeated reading can therefore contribute to vocabulary growth, rather than vocabulary always being an independent precursor. The amount learned from a passage depends on prior knowledge, context and assessment sensitivity, so tentative population extrapolations should not be confused with a guaranteed learning rate for every child.^[ref]

Neuroscience provides converging evidence of reorganisation. Stanislas Dehaene and colleagues compared adults with different literacy histories using fMRI responses to written and spoken material. Literacy was associated with changed visual responses to writing, stronger phonological responses to speech and activation of orthographic information from spoken input. The inclusion of adult learners shows that relevant differences are not restricted to childhood schooling, but the groups were not randomly assigned. The imaging does not by itself identify a unique instructional cause or show that larger activation is always better.^[ref]

The developmental relation can thus be recursive. Better language knowledge helps a child understand print; successful reading permits more independent encounters with words and structures; those encounters can support later oral language performance. Difficulty at either stage may restrict opportunities at the next. Such a loop is a hypothesis about mechanisms and experience, not a reason to assume that all observed growth differences are self-amplifying or irreversible.

A strong test would combine repeated measures before and after instruction with controlled changes to oral or written experience. One study could compare learning the same unfamiliar meanings through listening, independent reading or supported reading while matching content and exposure. Test both spoken understanding and written use after a delay, alongside decoding and prior vocabulary. Another could vary print support during phoneme instruction while assessing untrained speech-segmentation tasks. These designs can distinguish medium, content, instruction and transfer more clearly than a concurrent correlation.

Writing outcomes should remain part of that programme. Oral retelling, dictated composition and independent writing can help distinguish language formulation from spelling and transcription demands, though the tasks still differ in support and planning. Cross-language and multilingual studies can test whether reciprocity depends on script or experience rather than assuming that an English alphabetic result is universal.

Oral language is therefore a major foundation for reading and writing, but literacy is also an experience that reshapes language knowledge and access. The most useful account specifies the component and timescale in each direction. It treats experimental, longitudinal, intervention and imaging evidence as complementary constraints on a reciprocal developmental system.

Part II advanced option in the held 2022 papers. The source examination is shared by the psychology (experimental) course of the University of Oxford and the psychology, philosophy and linguistics course of the University of Oxford.

Source release: Oxford-released bundle, member a15915w1.pdf.

The cover requires three answers and gives a limit of 1,800 words without explicitly stating whether it applies per answer or per submission. All six options are developed here as separate study essays, each below 1,800 words; their aggregate is not presented as a three-answer submission. Scientific evidence is restricted to work available by 27 May 2022.

Arithmetic develops through learning what number words and symbols mean, coordinating counting with quantities, and selecting increasingly useful strategies. Teaching should connect these components and diagnose how a child obtained an answer, rather than treating fluent recitation or speed as sufficient evidence of understanding. Early abilities to notice quantity are relevant resources, but they do not by themselves supply exact symbolic arithmetic. The educational question concerns how children connect representations and procedures across tasks.

Counting requires more than producing a familiar sequence. The child must coordinate one number word with each object, preserve a stable word order and use the final count to answer how many objects there are. The cardinality principle links the counting procedure to the cardinality of the set. A child may recite farther than the quantities they can reliably produce on request. Comparing recitation, counting a displayed collection and giving a requested number of objects therefore reveals different aspects of knowledge rather than redundant measures of one skill.

Karen Wynn's longitudinal study followed two- and three-year-olds as they learned number-word meanings and the counting system. Children could recognize that number words referred to distinct quantities before knowing the exact quantity associated with each word. Their acquisition was gradual, with a substantial interval between early word-specific knowledge and understanding how counting represents quantity. A request to give a number of objects is particularly informative because repeating the word sequence alone cannot satisfy it. The result argues against assuming that a correct count routine establishes a complete number concept.^[ref]

This transition also explains why a single performance should not be overinterpreted. A child can misunderstand instructions, lose track of touched objects or know a quantity in one context without transferring the word to another. Assessment should vary the arrangement and objects while keeping the numerical question comparable. Success with a small collection need not show that the child understands arbitrary larger counts; failure on a lengthy verbal instruction need not show absence of every numerical insight. Number-word knower levels are useful descriptions of task performance, not fixed identities or universal timetable requirements.

Arithmetic then creates opportunities to coordinate these meanings. For $8+5$, a count-all strategy can represent both collections and count the combined objects. A counting-on strategy starts from the known eight and counts five further steps. A child can also use decomposition, $8+5=8+2+3=10+3=13$, drawing on a relation to ten. These are different routes to 13, with different demands on attention, working memory and available facts. They should not be inferred solely from a correct answer or a fast response.

Counting on from the larger addend reduces the number of required increments. For nonnegative natural numbers $a$ and $b$, commutativity permits starting from $\max (a,b)$ and making $\min (a,b)$ unit steps. This reduces counting effort without changing the sum. Understanding that relation is different from memorizing a rule to start with the larger written digit. When children explain or model why the shortcut works, the teacher can check whether it reflects a relation between quantities or only a rehearsed instruction.

Strategy development is not well described as every child abandoning one method on the same date. Children can use several procedures depending on number size, familiarity and confidence. Robert S. Siegler and colleagues used a microgenetic method to study second-graders discovering an arithmetic shortcut, with trial-by-trial response times and explicit reports. Evidence of the shortcut often appeared in performance before children could report it. The study shows why explanations and performance measures should be combined: failure to verbalize a strategy does not necessarily mean that no useful change has occurred.^[ref]

Increasing fluency can reduce the resources needed for familiar calculations, but rote speed and conceptual knowledge are not opposing educational goals. Reliable facts can support reasoning with less familiar numbers; understanding a relation can make facts easier to derive and check. Conversely, rapid retrieval of a wrong fact is not progress. Error patterns, use of inverses and flexible decomposition provide stronger evidence about developing arithmetic than a single timed total. Practice should therefore be evaluated through accuracy, efficiency, flexibility and retention, with task demands made explicit.

The teaching implications are specific. Counting activities should connect spoken numbers, objects and written numerals. Materials are useful when children understand what each object or movement represents, rather than merely manipulating attractive objects. Teachers can compare different correct solutions, ask children to predict how an answer changes when an addend changes, and connect a verbal story to its symbolic form. The relevant support depends on the child's current knowledge: supplying a more elaborate verbal explanation may burden someone who still needs to coordinate the count.

Thomas P. Carpenter and colleagues tested whether research knowledge about children's mathematical thinking could improve classroom instruction. Twenty first-grade teachers received a workshop and twenty were randomly assigned to a control condition. The workshop did not prescribe one script; intervention teachers paid more attention to children's problem-solving processes, and their pupils showed stronger outcomes on several mathematical measures. This is evidence that teacher understanding can matter beyond distributing a new worksheet. It does not identify one isolated workshop component as the unique cause or show that unguided discovery is always superior.^[ref]

Language, opportunities to participate, familiarity with symbols and classroom instruction all affect the observations used to judge arithmetic. A child learning the language of instruction may need a way to show a quantity without a lengthy explanation. Apparent success on a memorized format may fail on an unfamiliar representation. Formative assessment should therefore sample multiple tasks and use errors to select the next instructional step, while retaining ambitious goals rather than fixing a child permanently at an inferred ability level.

The strongest implication is to teach and assess connections: number words to quantities, counting to cardinality, procedures to relations, and practiced facts to flexible calculation. Developmental research identifies informative distinctions and plausible mechanisms; classroom experiments test whether using that knowledge improves learning. Neither a universal maturation schedule nor one preferred teaching method follows from the evidence.

Psychological research can improve mathematical intervention by identifying a child's component difficulties, translating a proposed mechanism into teachable activities and testing whether gains transfer to meaningful mathematics. Low attainment is an outcome to explain, not proof of one defective “number sense” system. A useful intervention must also specify its comparison condition, dose, intended learners and success criteria. The same program need not suit children whose errors have different origins.

Assessment can separate counting, number-word and numeral knowledge, symbolic magnitude comparison, number-line estimation, fact retrieval, calculation procedures and understanding of word problems. Language and working memory can affect these tasks, as can prior instruction and opportunity to practice. A broad low score may therefore conceal a relatively specific bottleneck. Developmental dyscalculia describes persistent mathematical learning difficulty in an appropriate assessment context; it should not be inferred from one poor session or used to assume that every affected child has the same cognitive profile.

Ann Dowker's Numeracy Recovery work illustrates a component-oriented approach: examine the child's performance in different arithmetical activities and provide targeted support rather than prescribing identical exercises from an overall score. Its early pilot status is important. Improvement after a program is compatible with benefit, but without a strong counterfactual it can also reflect ordinary teaching, repeated testing or regression toward the mean. The psychological formulation supplies a rationale that subsequently requires controlled evaluation; it is not itself an effect estimate.^[ref]

For a child who can recite numbers but cannot connect them to quantities, an intervention can make the relationship between number words, numerals and ordered quantities explicit. Geetha B. Ramani and colleagues tested a linear number board game against an otherwise similar color board game with low-income preschoolers. Roughly an hour of play improved several numerical outcomes, including magnitude comparison, counting, numeral identification and number-line estimation; gains were retained at a nine-week follow-up. The matched game comparison helps separate numerical content from merely receiving attention or playing with an adult.^[ref]

The result supports a mechanism-informed activity, but its scope needs care. Low-income preschoolers are not equivalent to children diagnosed with developmental dyscalculia. A number-board game combines several cues—counting words, movement, spatial order and numerals—so improvement does not uniquely identify which cue was necessary. Transfer to the assessed numerical tasks is informative without proving a lasting effect on every school arithmetic outcome. An extension should test classroom implementation and broader learning rather than equating a promising short intervention with complete remediation.

Practice can then be selected according to a learner's developing strategies. Lynn S. Fuchs and colleagues randomized at-risk first-graders to tutoring with speeded practice, tutoring with nonspeeded practice or no tutoring. Both tutoring conditions shared a substantial number-knowledge component and differed in the final five minutes of each session. Sixteen weeks of tutoring improved mathematical outcomes, while the practice contrast provided additional evidence about efficient counting and arithmetic learning. Because most instruction was shared, the comparison does not test timed practice as a stand-alone replacement for teaching concepts.^[ref]

This design is useful because it distinguishes package benefit from a narrower component question. A teacher observing that children improve after frequent arithmetic practice cannot tell whether time, feedback, explanation or practice format produced the gain. Randomizing a well-defined component within otherwise similar sessions provides a more interpretable answer. Yet a component can have different effects at different levels of prior knowledge. Testing those interactions requires prespecified measures and sufficient statistical power, not selecting a subgroup only after the average result is disappointing.

Computer delivery can adapt task difficulty and record detailed responses, but an adaptive interface is not evidence of efficacy. Pekka Räsänen and colleagues randomized children with low numeracy between two computer games emphasizing different numerical representations. The study found improvements in number comparison, with more limited generalization to other skills. It therefore supports asking which representation is trained and which outcome changes, rather than treating all improvement as a rise in general mathematical ability. The modest study and comparison structure also limit a universal ranking of software.^[ref]

A common alternative is to train a domain-general capacity thought to constrain mathematics. The crucial distinction is between association and intervention: working memory can predict performance without working-memory practice becoming an effective arithmetic treatment. Darren L. Dunning and colleagues randomized children with low working memory to adaptive training, low-level practice or a comparison condition. Improvements in related memory performance did not establish a corresponding generalized academic benefit. The study illustrates why near transfer should not be presented as demonstrated far transfer to mathematics.^[ref]

Intervention evaluation should include untaught examples, delayed tests and classroom-relevant performance. A program that only improves practiced items may still have value, but its claim is narrower than improving flexible calculation or word-problem understanding. Blinded assessors, comparable adult contact, monitoring of attendance and suitable analysis of class or school clustering strengthen inference. A no-treatment comparison estimates the package effect relative to ordinary opportunities; an active comparison is needed for stronger claims about distinctive instructional ingredients.

Researchers should also ask why an intervention fails. A child may not understand the vocabulary, may be overloaded by instructions or may have had too little opportunity to participate. Mathematical anxiety can affect engagement, but it should not become an unfalsifiable explanation for every weak response. Fidelity and acceptability matter alongside cognitive mechanism: an effective activity in a closely supported laboratory may be difficult to deliver in an ordinary classroom. Those constraints are part of the educational intervention, not distractions from its psychological theory.

The practical model is a cycle of assessment, targeted teaching and reassessment. Performance can determine the next task, while repeated review prevents a provisional description from becoming a permanent ceiling. A strong design measures proposed processes before outcomes, then tests whether the intervention changed the intended process and whether that change predicts relevant learning. Statistical mediation alone remains weaker than a selective manipulation of the mechanism, especially when motivation and extra teaching change together.

The contribution of psychology is not one universal cure but a disciplined way to match an instructional problem to a testable account of learning. Targeted number teaching has encouraging experimental support; the evidence also warns against mistaking commercial sophistication, trained-task improvement or a correlational cognitive predictor for demonstrated mathematical benefit. Evaluation should remain tied to the particular child population, content and transferable outcome claimed.

Separating pupils by attainment is not, by itself, a reliable solution to individual differences. Benefits depend on whether the arrangement changes the instruction pupils receive, while costs can arise from restricted opportunities, inaccurate placement and persistent labels. Evidence distinguishes whole-curriculum separation from subject-specific and flexible grouping; it does not justify treating all forms as one intervention. Current achievement should also not be mistaken for a fixed measure of learning potential.

Streaming by attainment assigns a pupil to a group across much of the curriculum. Subject setting regroups pupils for a particular subject, allowing different placements in, for example, reading and mathematics. Within-class grouping creates smaller instructional groups inside a shared classroom and can be revised by task. These arrangements differ in exposure to peers, teachers, curriculum and opportunities to move. A finding about flexible mathematics groups therefore cannot automatically defend an enduring all-subject stream.

The favorable hypothesis is instructional fit. A narrower range of prior knowledge may let a teacher select examples, pace and support more accurately, while advanced pupils avoid unnecessary repetition. But the group label must actually change the lesson in a useful way. Separating pupils and then delivering the same unsuitable instruction supplies little reason to expect benefit. Nor does an appropriate lesson logically require fixed separation: formative assessment and flexible task allocation can address differences within a mixed classroom.

Robert E. Slavin's best-evidence synthesis distinguished assignment to self-contained ability classes, which lacked a general achievement advantage, from some subject regrouping and within-class mathematics arrangements with more favorable results. Its conditions emphasized reassessment and adapting the level and pace of teaching. The historical importance is the distinction between organization and instruction, not a timeless verdict that any arrangement called “grouping” works. A synthesis also inherits variation in the quality of its constituent comparisons.^[ref]

Saiying Steenbergen-Hu and colleagues' later second-order meta-analysis likewise separated forms of grouping. Across thirteen meta-analyses, between-class grouping showed little overall achievement benefit, while within-class grouping, cross-grade subject grouping and specialized arrangements had more positive average results. Acceleration was analyzed separately and should not be collapsed into ordinary attainment sorting. The evidence spans school ages and implementations, so its averages are not estimates of one particular primary-school policy. Reviews that share primary studies are also not independent replications to be counted as separate votes.^[ref]

These conclusions do not establish that tracking must harm low-attaining pupils in every setting. Esther Duflo and colleagues conducted a randomized tracking experiment in 121 Kenyan primary schools. In that context, tracking benefited lower-achieving pupils through better instructional targeting, despite the potential advantages of exposure to higher-achieving peers. The experiment provides a causal counterexample to a universal claim that mixed-attainment classes are always best. Its teacher incentives, curriculum and wider school conditions also limit transfer to a different education system.^[ref]

The Kenyan result helps distinguish two effects often conflated in discussion. Peer effects concern how classmates' skills or behavior influence learning; instructional targeting concerns what a teacher teaches to whom. They can act in opposite directions. A lower-attaining group might lose some peer opportunities yet gain a lesson that it can use. A different system could concentrate less experienced teachers or narrow tasks in the same group, producing a different outcome. The causal question is therefore the effect of the whole implemented arrangement, not a social label in isolation.

Placement creates further challenges. An assessment contains measurement error and reflects prior opportunity, language and familiarity with the task. Selecting pupils at one time point can therefore turn temporary performance differences into apparently stable categories. Even frequent movement on paper is insufficient if movement rarely occurs in practice. Researchers need placement histories and the material actually taught, not only the school's description of its policy. Comparing final scores of high and low groups is especially uninformative because initial attainment helped define those groups.

Teacher expectations and children's interpretations of their placement are plausible additional pathways. A label can affect the challenge offered or the opportunities to demonstrate improvement; a child may interpret it as a statement of fixed worth. These are hypotheses requiring direct measurement rather than automatic explanations for every group difference. Self-concept, participation and belonging matter even when an average test score is unchanged. Conversely, a pupil may value appropriately demanding peers and tasks, so increased challenge should not be equated with social harm merely because it involves grouping.

A useful evaluation would randomize schools or classes to well-specified arrangements, measure baseline attainment before allocation and monitor curriculum, staffing and movement. It should retain an intention-to-treat comparison rather than excluding pupils whose placements changed. Multilevel analysis or appropriate uncertainty estimates are required when pupils share teachers or classrooms. Prespecified subgroup analyses can ask whether effects differ by starting attainment, while distributional outcomes show whether an average gain masks widening gaps. Longer follow-up is important because early curricular restriction may have consequences beyond an immediate assessment.

The relevant comparison should be equally credible. A well-resourced setting program compared with an under-supported mixed class does not identify separation alone as the active ingredient. Equally, a mixed class receiving no help to differentiate instruction is not a decisive test of responsive mixed teaching. Comparing arrangements under explicit resource and implementation conditions makes the evidence more useful for practice. It also permits testing whether flexibility and good assessment capture the benefit without committing pupils to permanent categories.

For primary education, where performance profiles and opportunities can change quickly, the most defensible principle is responsiveness. Grouping may serve a specific instructional purpose when it is revisable and tied to what the pupil needs to learn next. Its effects should be monitored rather than justified by a presumed innate hierarchy. Maintaining ambitious curriculum access and checking actual movement are safeguards whose value should be assessed alongside attainment, not assumed to follow from a policy name.