Non-Verbal Cognitive Tasks for Autism Spectrum Disorder

ASD deep-dive companion to nonverbal-cognitive-tasks: which language-free cognitive tasks best distinguish autism in humans, and how the same task families translate to mouse and monkey models — in particular SHANK3 models of Phelan–McDermid syndrome (PMS). Batteries covered here: cantab (CANTAB-ASD), the NIH Toolbox Cognition Battery, and the ABC-CT biomarker battery. Animal test platforms: automated-cognitive-testing-devices (Monkey CANTAB, Bussey-Saksida, CageLab and neighbours).

Overview

ASD is defined behaviourally — social-communication differences plus restricted/repetitive behaviour — and it has no single cognitive signature. Non-verbal (language-free) tasks are used for three reasons: they remove the language confound (many autistic people are minimally verbal, and toddlers cannot be tested verbally); the same task logic can be implemented in rodents and non-human primates (NHPs); and they yield objective, quantitative endpoints for intervention trials. The best-validated group-level signals are social-attention eye-tracking and modest, domain-general executive-function differences.¹²³

Non-verbal task families with ASD evidence and cross-species reach:

Domain	Best human tasks	Cross-species analogs
Visuospatial working memory	[[spatial-working-memory	CANTAB SWM]]
Cognitive flexibility (set-shifting, reversal)	[[intradimensional-extradimensional-shift	CANTAB IED]]; [[reversal-learning
Associative / episodic memory	CANTAB PAL; NIH-TCB Picture Sequence Memory	paired-association learning in macaques 4
Social attention & gaze	eye-tracking batteries (ABC-CT OMI; social visual engagement)	eye contact in SHANK3 monkeys 5; gaze in VPA marmosets 6 — no direct rodent analog
Inhibitory control, attention, speed	NIH-TCB Flanker, Pattern Comparison	not yet standard in ASD models — rodent test harmonisation is still being called for 7

Why ASD resists a simple cognitive signature

Three lines of evidence set the ceiling on “tasks that distinguish ASD”:

• Meta-analytic EF differences are real but moderate and unspecific. Demetriou et al. (2018) pooled 235 studies (14,081 participants) and found g = 0.48 overall, similar across EF subdomains, with no support for fractionation into distinct subdomain deficits — and “only a small number of EF measures achieved clinical sensitivity”.¹
• Diagnostic specificity is the weak point. Kofler et al. (2024) conclude that most studies fail to account for ASD–ADHD co-occurrence, that traditional neuropsychological tests and rating scales have poor specificity and construct validity, and that the most parsimonious reading of the field is that children with ADHD and/or ASD perform moderately worse than neurotypical children on a broad range of tests, with unique-vs-shared subprofiles still largely unknown.²
• Group effects have shrunk over time. Across 11 meta-analyses of autism–control comparisons, effect sizes for 7 distinct differences decreased over publication year, 5 of them significantly — consistent with broadened (more heterogeneous) diagnostic boundaries rather than simply better measurement.³

Human task evidence, by domain

Visuospatial working memory

The classic finding: high-functioning individuals with autism made more errors on the CANTAB SWM self-ordered search and were less likely to use a consistent search strategy, with deficits emerging when memory load exceeded a limited capacity.⁸ Cambridge Cognition highlights SWM as “highly sensitive to cognitive dysfunction in autism” in its ASD battery.⁹ Meta-analytic summaries put autistic WM differences at moderate size (school-age children larger than adolescents), with visuospatial tasks yielding larger effects than verbal ones.²

Cognitive flexibility: set-shifting and reversal

• An early landmark, Ozonoff et al. (2004), found deficits in planning efficiency (Stockings of Cambridge) and extradimensional set-shifting (IED) in 79 autistic vs 70 matched controls across ages 6–47.¹⁰
• The 2024 meta-analysis of cognitive flexibility (59 studies; 2,122 autistic participants without intellectual disability) found a small-to-moderate overall effect, with perseverative errors producing the largest effects, and substantial heterogeneity across task/age/outcome choices.¹¹
• Probabilistic reversal learning (PRL; reversal-learning) is the sharpest human probe: ASD participants performed normally during acquisition but at reversal made more regressive errors — reverting to the previously preferred response — and regressive-error counts correlated with independently rated restricted/repetitive behaviours, implicating frontostriatal flexibility circuitry.¹² PRL has also passed feasibility, test–retest, and intervention-sensitivity checks as a cognitive-flexibility outcome measure in a within-subjects ASD trial.¹³
• Nuance: the NIH-Toolbox DCCS (a set-shifting task) did not differentiate autistic children after age/IQ control,¹⁴ while in adolescents/young adults the flexibility composite did.¹⁵ Task demand and age matter.

Inhibitory control, attention, and processing speed

In autistic children aged 3–17 the NIH-TCB Flanker (inhibitory control/attention) and Pattern Comparison (processing speed) showed small but significant diagnosis-related differences after controlling age and IQ; picture-sequence memory and DCCS did not.¹⁴ In an IQ-matched adolescent/young-adult cohort (12–22), ASD showed poorer inhibitory control, cognitive flexibility, episodic memory, and processing speed — all Fluid composite components — while vocabulary and word reading (Crystallized) were spared: a fluid-over-crystallized profile that persisted across ages, with latent-profile analysis splitting autistic participants into a broad-impairment group and a crystallized-stronger group.¹⁵

Construct note: these measures index interference control and prepotent-response inhibition — meta-analysis of 41 inhibition studies found a moderate ASD effect on prepotent response inhibition (ES 0.55, moderated by age) and a smaller effect on interference control (ES 0.31), with substantial heterogeneity,¹⁶ while the narrower action-cancellation construct measured by the stop-signal task (stop-signal-task) shows little SSRT difference in ASD.¹⁷

Social attention, face processing, and emotion

This is where non-verbal tasks achieve their strongest discrimination:

• ABC-CT eye-tracking battery (280 autistic, 119 TD children, 6–11; Activity Monitoring, Social Interactive, Static Scenes, Biological Motion, Pupillary Light Reflex). Gaze-to-faces carried the signal: Activity-Monitoring %Face d = −1.04; the aggregated Oculomotor Index (OMI) d = −0.79; six-week stability (ICC) mostly moderate-to-high. The consortium explicitly framed these as group-asymmetry markers for mechanistic trials, “not … individual diagnostic precision”.¹⁸
• Social visual engagement vs expert diagnosis (475 children, 16–30 months): a single eye-tracking index of preferential social looking reached 71.0% sensitivity / 80.7% specificity, rising to 78.0%/85.4% in children with certain diagnoses — the strongest individual-level discriminative performance of any non-verbal battery to date.¹⁹
• GeoPref Test (1,863 toddlers; 1-minute social-vs-geometric movie): 98% specificity but only 17% sensitivity (33% with saccades) at a 69% fixation threshold — i.e. a subtype marker of an ASD segment, highly heritable (twin concordance).²⁰
• Meta-analysis 2025 (57 studies, ≤36 months): pooled social-fixation difference g ≈ 0.65; machine-learning pipelines reached sensitivity up to 89% and specificity up to 86%; heterogeneity and protocol standardisation remain the blockers to clinical translation.²¹
• The Selective Social Attention task (ABC-CT-adjacent) replicated toddler effects in 4–12-year-olds: reduced face looking during socially engaging conditions, with %Face correlating with SRS symptom scores.²²
• Emotion recognition (CANTAB ERT is inside the vendor’s ASD battery): meta-analysis shows non-selective impairment across all basic emotions, relatively specific to ASD vs other clinical groups but not to emotion-specific or face-specific processing — a softer discriminator than social attention.²³⁹

Associative learning and episodic memory

CANTAB PAL (paired associates learning) is a member of the vendor’s ASD battery (episodic-memory endpoint).⁹ On the NIH-TCB, picture-sequence (episodic) memory differentiated IQ-matched autistic adolescents,¹⁵ but not children in the feasibility study.¹⁴ The strongest cross-species motif here is the paired-association learning deficit in SHANK3-monkey models (see below),⁴ which maps onto PAL-like tasks in all three species.

Batteries

CANTAB (and the CANTAB-ASD battery)

The vendor’s ASD battery packages: Emotion Recognition Task (ERT), Match to Sample Visual Search (MTS), Multitasking Test (MTT), One Touch Stockings of Cambridge (OTS), Paired Associates Learning (PAL), Reaction Time (RTI), Spatial Working Memory (SWM) — covering executive function, episodic memory, planning, and processing speed, with the claim that the tests discriminate ASD from comorbid ADHD and depression and are sensitive across severity and age.⁹ Underlying literature: OTS + IED deficits (Ozonoff 2004),¹⁰ SWM deficits load-dependent (Steele 2007),⁸ and ASD–ADHD differentiation by EF profile — interestingly, in one carefully matched study ADHD (not ASD) was the group with reduced CANTAB SWM, while ASD was distinguished by WISC perceptual-reasoning relative weakness, illustrating that EF tasks alone rarely separate diagnoses cleanly.²⁴ The CANTAB lineage is the most cross-species-extended test family in existence: human CANTAB → Monkey CANTAB IntelliStation → Bussey-Saksida rodent touchscreens (see automated-cognitive-testing-devices, lafayette-instrument, campden-instruments).

NIH Toolbox Cognition Battery (NIHTB-CB)

A standardised iPad battery; Fluid composite = DCCS, Flanker, Picture Sequence Memory, List Sorting, Pattern Comparison; Crystallized composite = Picture Vocabulary + Oral Reading Recognition (years 7+),²⁵ plus early-childhood versions for ages 3–6.²⁵ In ASD:

• Feasibility (116 autistic vs 80 TD, ages 3–17): 57% of autistic children completed all four tested tasks vs 88% of TD; those with IQ ≤ 70 completed on average 2.4 of 4 tasks, with task-completion attrition highest for processing-speed and memory tasks; performance was strongly IQ-dependent; only Flanker and processing speed carried diagnosis effects.¹⁴
• Profile (IQ-matched 12–22-year-olds): fluid-over-crystallized impairment with LPA-derived subgroups;¹⁵ both papers also note that mental-age-based administration (from the ID literature) improves feasibility in intellectual disability,¹⁴ and the battery has been formally validated in ID populations.¹⁴

ABC-CT (Autism Biomarkers Consortium for Clinical Trials)

A multi-site effort to qualify objective biomarkers for ASD trials: an eye-tracking battery (OMI; above),¹⁸ an EEG battery analysed with the same psychometric rigour, and video-tracking — deliberately built as trial-readiness instruments rather than diagnostic tests.²⁶

Mouse models: Shank3 and beyond

The standard behavioural battery

The field’s canonical framing tests the three diagnostic domains — social interaction, communication, and repetitive behaviour — with an expanding cognitive add-on layer.²⁷ A 2024 scoping review of rodent ASD models (genetic: Shank3, Fmr1, 16p11.2, Cntnap2, Mecp2, BTBR; environmental: VPA, poly(I:C)/MIA, propionic acid) calls for a comprehensive scoring system so models can be compared by how many neurobehavioral domains they capture rather than single assays.⁷ Cross-model comparison for SHANK3 specifically shows phenotypes that track construct and residual-isoform differences, mirroring clinical heterogeneity.²⁸

Cognitive tasks in Shank3 and other ASD models

• Shank3B mice (+/−, PDZ-domain deletion): slower to reach criterion in the touchscreen pairwise visual discrimination task (with trends toward more errors), while open-field activity and three-chamber sociability were normal — i.e. an early, cognitive phenotype in a model without gross social/activity confounds.²⁹ (The task was explicitly chosen for its similarity to CANTAB/NIH Toolbox assays — the translation argument in miniature.)
• Shank3 rats (Long-Evans; WT/HET/KO longitudinal): the most prominent adult deficit was touchscreen visual discrimination and reversal learning — rapid, error-prone responding — against a background of mostly normal social behaviour; female knockouts showed reduced distress ultrasonic vocalisations early; subtle juvenile memory/coordination abnormalities preceded the adult executive dysfunction.³⁰
• BTBR mice (idiopathic model): impairments in operant probabilistic reversal learning in both sexes — connecting the mouse assay to the human PRL findings.³¹
• A Shank3 transgenic line showed pronounced repetitive/stereotypic behaviour and anxiety-related phenotypes with only mild social change and slightly impaired cognitive flexibility across development.³²
• Platforms for these tasks: Bussey-Saksida touchscreen chambers (Campden/Lafayette) and open-source alternatives; rodent touchscreen batteries were designed explicitly for cross-species task homology (see automated-cognitive-testing-devices).³³³⁴

Monkey models: SHANK3 and other NHP models

SHANK3 macaques

• Founder generation (CRISPR-Cas9, germline-transmissible; Zhou et al. 2019, Nature): sleep disturbances, motor deficits, increased repetitive behaviours, and social and learning impairments, with altered local/global functional connectivity.³⁵
• A single SHANK3-mutant cynomolgus (SHANK3M3) followed longitudinally against matched controls: delayed growth, increased stereotyped behaviour, reduced exploration, anxiety-related crook-tail posture, reduced social initiation, and reduced eye contact; chronic fluoxetine (2.5 mg/kg/day) improved social interaction and partially normalised brain glucose metabolism — the first drug-response evidence in a SHANK3 primate.⁵
• F1 generation cohort (Jiang et al. 2026, Neuron; larger, non-mosaic): sleep disturbances, diminished exploration, atypical social interactions, stereotypical behaviours, altered functional connectivity, and a markedly diminished auditory ERP response. Cognitively, the F1 monkeys showed no major working-memory deficit but impaired paired-association learning — the cleanest evidence that the SHANK3 haploinsufficiency signature in primates is selective, and that explicit memory tasks (PAL-like) are more sensitive than WM tasks. A multi-task behavioural array was developed to score autism-related phenotypes systematically, and revealed pronounced inter-individual heterogeneity.⁴

Other NHP models and the monkey-task landscape

• MECP2-overexpressing macaques: autism-like behaviours (repetitive locomotion, increased stress responses, reduced social interaction) with germline transmission — the proof-of-principle for transgenic NHP models of syndromic autism.³⁶
• Maternal VPA macaques: neurogenesis defects and autism-like behaviours in offspring.³⁷
• VPA marmosets: reduced early synaptogenesis (then increased at juvenile ages, as in human tissue), transiently increased plasticity in infancy associated with altered vocalisations, and gene-expression changes resembling human ASD;³⁸ a separate cohort showed reduced gaze to conspecifics in childhood that predicted adult social deficits and perseverative reversal-learning behaviour — the strongest longitudinal childhood-marker→adult-phenotype chain in any animal model.⁶
• Task availability is the bottleneck: a review of macaque screening tools concluded that limited phenotype-screening batteries constrain the translational utility of NHP models and called for specialised tests and rodent–macaque collaboration.³⁹ New instruments continue to be developed, e.g. a scalable touchscreen delayed non-match-to-position task validated in 12 marmosets and 71 humans that parametrically manipulates delay and spatial interference and works identically in both species.⁴⁰

Cross-species task map

Task family	Human evidence (ASD)	Mouse	Monkey
Visuospatial WM (self-ordered / DNMTP)	SWM errors ↑, strategy ↓ 8	(no strong Shank3 WM phenotype reported)	DNMTP validated marmoset↔human 40; WM intact in SHANK3 F1 macaques 4
Discrimination & reversal / PRL	regressive errors ↔ RRB 12; PRL outcome-ready 13	discrimination & reversal deficits in Shank3 rats 30; PRL deficits in BTBR 31	reversal perseveration in VPA marmosets 6
Paired-association / associative memory	PAL in CANTAB-ASD battery 9	PAL tasks available on rodent touchscreen platforms	impaired in SHANK3 F1 macaques 4
Social attention / gaze	OMI d≈−0.8; 71/81% sens/spec 1819	no direct analog (sociability assays as proxies 27)	eye contact ↓ (SHANK3) 5; gaze ↓ predicts inflexibility (marmoset) 6
Repetitive behaviour	PRL/EF correlates; RRB scales 12	grooming/stereotypies (reported across Shank3 lines 3228)	stereotypies in SHANK3 founders 35 and MECP2 monkeys 36
Emotion recognition	ERT; non-selective impairment 23	absent	—

Translation strategy and platforms

• The identical-test strategy is the field’s ideal: the first direct demonstration ran the same object-location paired-associates task in Dlg2 knockout mice and in humans carrying DLG2 CNV deletions, showing genotype-linked PAL impairment in both species.³³ (DLG2 is a postsynaptic-density scaffold gene — the same protein family logic as SHANK3, making this the template for SHANK3 co-clinical studies.)
• Co-clinical trials: computer-automated cognitive batteries (CANTAB-family) are argued to be the practical bridge for mouse→human trials.³⁴ A 2025 systematic review of actual human–rodent touchscreen co-clinical studies found only 6 eligible studies; behavioural flexibility and visuospatial cognition showed the best comparability, with methodological diversity the main gap.⁴¹
• For implementing these tasks in monkeys and mice at scale, see the platform landscape in automated-cognitive-testing-devices — including Monkey CANTAB, Bussey-Saksida chambers and the self-adapting in-cage systems (CageLab and neighbours).

Recent literature (2023–2026 highlights)

• Jiang et al. 2026 (Neuron) — F1 SHANK3 macaques: selective cognitive signature (PAL-impaired, WM-spared); multi-task phenotype array.⁴
• Pearson et al. 2026 (bioRxiv, NIMH) — Shank3 Long-Evans rats: late-emerging, touchscreen-defined executive dysfunction as the leading adult phenotype.³⁰
• Vanderlip et al. 2026 (Neurobiol Learn Mem) — cross-species delayed non-match-to-position (marmosets + humans) with parametric delay/interference.⁴⁰
• Tecar et al. 2025 (J Clin Med) — eye-tracking early-screening meta-analysis: 57 studies, g ≈ 0.65, ML pipelines up to 89%/86%.²¹
• Martins et al. 2025 (J Neurochem) — systematic review of human–rodent co-clinical touchscreen trials; flexibility + visuospatial most comparable; field thin (6 studies).⁴¹
• Lage et al. 2024 (Neurosci Biobehav Rev) — cognitive-flexibility meta-analysis; perseverative errors largest effect.¹¹
• Kofler et al. 2024 (Nat Rev Psychol) — ADHD/ASD EF review; specificity and co-occurrence as the core methodological problems.²
• Ongoing: ABC-CT-derived measures are the current template for trial-ready non-verbal endpoints in ASD,¹⁸²⁶ and SHANK3-macaque platforms are moving toward drug testing.⁵⁴

Open questions

• Which task, for what purpose? Screening/subtype detection (eye-tracking GeoPref-type, high specificity/low sensitivity) and mechanistic trial endpoints (OMI, PRL, PAL) are different products with different psychometrics.¹⁸²⁰
• Human–animal discordance: humans show visuospatial WM weaknesses; SHANK3 macaques show intact WM but PAL deficits.⁸⁴ Is that model-specific, task-specific (self-ordered search vs DNMTP), or a species difference in what the gene does?
• Specificity vs transdiagnostic value: EF and flexibility differences overlap heavily with ADHD; the ASD-unique cognitive core remains contested.²
• Standardisation: from mouse batteries (call for a comprehensive scoring system⁷) to eye-tracking protocols²¹ and co-clinical task harmonisation,⁴¹ the field lacks shared standards.
• Predictive validity: PRL¹³ and eye-tracking¹⁸ are the first non-verbal measures with published trial-feasibility data; none has yet been shown to track a successful disease-modifying intervention.

Related pages

nonverbal-cognitive-tasks · cantab · automated-cognitive-testing-devices · preclinical-drug-screening — task pages: intradimensional-extradimensional-shift, reversal-learning, stop-signal-task, delayed-match-to-sample, spatial-working-memory, emotion-recognition-task, one-touch-stockings-cambridge, delayed-response-tasks-working-memory — entities: lafayette-instrument, campden-instruments

References

Footnotes

1. raw/papers/demetriou-2018-ef-meta-analysis.md ↩ ↩²

2. raw/papers/kofler-2024-ef-adhd-asd-review.md ↩ ↩² ↩³ ↩⁴ ↩⁵

3. raw/papers/rodgaard-2019-temporal-effect-sizes.md ↩ ↩²

4. raw/papers/jiang-2026-f1-shank3-macaques-neuron.md ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷ ↩⁸

5. raw/papers/tu-2019-shank3-monkey-crispr.md ↩ ↩² ↩³ ↩⁴

6. raw/papers/nakagami-2022-marmoset-vpa-attention.md ↩ ↩² ↩³ ↩⁴

7. raw/papers/ornoy-2024-rodent-models-scoping.md ↩ ↩² ↩³

8. raw/papers/steele-2007-spatial-working-memory-autism.md ↩ ↩² ↩³ ↩⁴

9. raw/articles/cantab-autism-battery-2026.md ↩ ↩² ↩³ ↩⁴ ↩⁵

10. raw/papers/ozonoff-2004-cantab-autism.md ↩ ↩²

11. raw/papers/lage-2024-cognitive-flexibility-meta.md ↩ ↩²

12. raw/papers/dcruz-2013-prl-autism.md ↩ ↩² ↩³

13. raw/papers/schmitt-2021-prl-outcome-measure.md ↩ ↩² ↩³

14. raw/papers/jones-2021-nih-toolbox-feasibility-autism.md ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

15. raw/papers/solomon-2020-nih-toolbox-autism.md ↩ ↩² ↩³ ↩⁴

16. raw/papers/geurts-2014-asd-inhibition-meta.md ↩

17. raw/papers/lipszyc-2010-sst-meta-analysis.md ↩

18. raw/papers/shic-2022-abcct-eye-tracking-battery.md ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶

19. raw/papers/jones-2023-social-visual-engagement-jama.md ↩ ↩²

20. raw/papers/wen-2022-geopref-biomarker.md ↩ ↩²

21. raw/papers/tecar-2025-eyetracking-screening-meta.md ↩ ↩² ↩³

22. raw/papers/shic-2023-selective-social-attention.md ↩

23. raw/papers/yeung-2022-facial-emotion-meta.md ↩ ↩²

24. raw/papers/matsuura-2014-asd-adhd-differentiation.md ↩

25. raw/articles/nih-toolbox-cognition-battery-2026.md ↩ ↩²

26. McPartland JC et al. (2020). The Autism Biomarkers Consortium for Clinical Trials (ABC-CT): Scientific Context, Study Design, and Progress Toward Biomarker Qualification. Front Integr Neurosci 14:16. https://doi.org/10.3389/fnint.2020.00016 ↩ ↩²

27. raw/papers/silverman-2010-mouse-autism-assays.md ↩ ↩²

28. raw/papers/delling-2021-shank3-model-comparison.md ↩ ↩²

29. raw/papers/copping-2016-shank3b-touchscreen.md ↩

30. raw/papers/pearson-2026-shank3-rat-touchscreen.md ↩ ↩² ↩³

31. raw/papers/alvarez-2023-btbr-prl.md ↩ ↩²

32. raw/papers/bauer-2022-shank3-mouse-sex-genotype.md ↩ ↩²

33. raw/papers/nithianantharajah-2015-identical-touchscreen.md ↩ ↩²

34. raw/papers/palmer-2021-touchscreen-translation.md ↩ ↩²

35. raw/papers/zhou-2019-shank3-macaques-nature.md ↩ ↩²

36. raw/papers/liu-2016-mecp2-transgenic-monkeys.md ↩ ↩²

37. raw/papers/zhao-2019-vpa-macaque.md ↩

38. raw/papers/watanabe-2021-vpa-marmoset-model.md ↩

39. raw/papers/qin-2019-macaque-screening-tools.md ↩

40. raw/papers/vanderlip-2026-cross-species-swm.md ↩ ↩² ↩³

41. raw/papers/martins-2025-touchscreen-coclinical-review.md ↩ ↩² ↩³