Language
Statistical Learning Task
Also called: visual statistical learning task, VSL task, triplet learning task, embedded triplet task, statistical learning paradigm
Measure learning from exposure alone: participants watch a stream of shapes that hides fixed groups of three, then pick the real groups over mixed ones.
8-10 minUntimed
Free to start: 10 free participants per account, then $0.99 per participant for academic research, or any yearly plan. Preview any test free after you sign up.
- Papers, 2021 to 2025
- 210
- Evidence grade
- C
- Registered clinical studies
- 1
What it measures
Measures learning of regularities from exposure alone. Participants watch a stream of shapes and press when a shape repeats; unknown to them, the shapes come in fixed groups of three. A surprise test pairs each group with shapes that never appeared together. The main score is the share of choices that pick the real group.
Constructs
- Statistical learning
- Implicit learning
- Learning of transitional probabilities
- Sequence segmentation
Research uses
- Research on language learning, reading and dyslexia at the group level
- Developmental research in children and adults
- Implicit learning and attention research
How the paradigm works
The Statistical Learning Task measures how people pick up regularities just by watching. In the visual version of the method (Fiser and Aslin 2002; Turk-Browne, Junge and Scholl 2005), shapes appear one at a time in a stream that looks random, but the shapes come in fixed groups of three, called triplets: within a triplet, each shape always follows the one before it. Nothing marks where a triplet starts or ends, so the only cue is how often one shape follows another. After the stream, a surprise test shows a triplet and a group of the same shapes that never appeared together, and asks which looks more familiar. Choosing the triplets more often than chance shows that the regularities were learned. The method grew out of studies of syllable streams (Saffran, Aslin and Newport 1996), and visual statistical learning has been linked to reading ability in children and adults (Arciuli and Simpson 2012).
More about the paradigm
This version follows the procedure of Arciuli and Simpson (2012). Twelve shapes drawn by ConductScience's code are grouped at random into 4 triplets for each session. Each triplet appears 24 times (16 to 32 in the settings) in a seeded order with no triplet twice in a row and no pair of triplets twice in a row. Each shape is shown for 400 ms (400 to 800 in the settings), followed by a 200 ms blank. In one presentation in four, one shape of the triplet is shown twice in a row, and participants press the space bar or tap the screen when that happens. This cover task keeps attention on the shapes and says nothing about triplets. A practice stream of plain shapes comes first, with feedback, and runs again until both of its repeats are caught, 3 times at most.
In the test, participants learn that the shapes came in groups of three. Each of 32 trials shows one triplet and one foil, one after the other with a 1-second pause, at the stream's pace. A foil takes the first shape of one triplet, the second of another and the third of a third, so each shape keeps its place but its pairs never appeared in the stream (Turk-Browne, Junge and Scholl 2005). Every triplet meets every foil twice, once first and once second, so triplets and foils are seen equally often in the test. There is no feedback and no time limit.
The main score is the share of trials on which the triplet was chosen; 0.5 is chance (Turk-Browne, Isola, Scholl and Treat 2008). The task also reports the number of correct choices and whether it is above chance for that person by an exact binomial test (22 or more of 32), the median choice time, the share of repeats caught in the stream, presses when nothing repeated and the time taken to catch a repeat. Choices under 100 ms are anticipations and are left out. Scores support research and do not diagnose.
Scores
- Triplet Choice Rate%
- Share of test trials on which the triplet was chosen over the foil. Choices under 100 ms are left out.
- Higher means more of the regularities were learned; 0.5 is chance
- Correct Choicescount
- Test trials on which the triplet was chosen, out of the choices scored.
- Higher is better
- Above Chancecount
- 1 when the correct choices are above chance for this person by a one-sided exact binomial test at p < .05 (22 or more of 32), otherwise 0.
- 1 shows learning in this session
- Median Choice Timems
- Median time from the question to the choice, leaving out choices under 100 ms.
- Informational
- Repeats Caught%
- Share of shapes shown twice in a row that were caught by a press from 100 ms after the repeat until the next shape ended.
- Quality check; under 0.5 is flagged
- Presses Without a Repeatcount
- Presses during the stream that answered no repeat.
- Quality check
- Time to Catch a Repeatms
- Median time from a repeated shape's onset to the press that caught it.
- Informational
- Anticipationscount
- Test choices under 100 ms, left out of every score.
- Quality check
- Choices Scoredcount
- Test trials with a choice of 100 ms or more.
- Informational
Use and evidence
210 papers from 2021 to 2025 name this task in their title or abstract, up 21% compared with 2016 to 2020. That ranks it 120 of 395 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 58 |
| 2024 | 46 |
| 2023 | 36 |
| 2022 | 40 |
| 2021 | 30 |
| 2020 | 49 |
| 2019 | 36 |
| 2018 | 35 |
| 2017 | 32 |
| 2016 | 22 |
| 2015 | 31 |
| 2014 | 17 |
| 2013 | 19 |
| 2012 | 17 |
| 2011 | 20 |
| 2010 | 29 |
| 2009 | 7 |
| 2008 | 3 |
| 2007 | 5 |
| 2006 | 3 |
| 2005 | 2 |
| 2004 | 2 |
| 2003 | 4 |
| 2002 | 1 |
| 2001 | 2 |
| 2000 | 0 |
Use grade C
100 to 299 papers from 2021 to 2025.
Evidence grade C
At least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies.
- Papers on test-retest reliability
- 5
- Papers on reliability or validity
- 15
- Registered clinical studies using it as an outcome
- 1
Note on the counts: may include machine-learning papers.
How the grades work
Use: A is 1,000 or more papers from 2021 to 2025, B 300 to 999, C 100 to 299, D 30 to 99 and E fewer than 30. Evidence: A means at least 25 papers report its test-retest reliability and at least 25 registered clinical studies use it as an outcome; B, at least 8 test-retest papers, and either 5 registered clinical studies or 150 reliability or validity papers; C, at least 2 test-retest papers, 30 reliability or validity papers or 3 registered clinical studies; D, little published measurement evidence found. The counts measure use. They include every paper that names the task in its title or abstract. Grades describe the published research on this task, not ConductCognition's version of it.
Source: ConductCognition test census 0.14.0, counted 31 October 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Shapes
- 12 shapes drawn by ConductScience's code, grouped at random into 4 triplets for each session. Practice uses a circle, a square and a ring.
- Stream
- Each triplet 24 times (standard; 16 to 32), never the same triplet twice in a row and never a pair of triplets twice in a row. Each shape is shown for 400 ms (standard; 400 to 800), then a 200 ms blank.
- Repeat check
- In one presentation in 4, one shape of the triplet comes twice in a row, spread over the three places, and participants press the space bar or tap the screen. A press counts from 100 ms after the repeat until the next shape ends.
- Test
- 32 trials, each a triplet and a foil one after the other with a 1,000 ms pause, at the stream's pace, then a choice of the more familiar group with no time limit and no feedback. No more than 3 trials in a row with the triplet in the same place.
- Foils
- The first shape of one triplet, the second of the next and the third of the one after, so each shape keeps its place but its pairs never appeared in the stream. Every triplet meets every foil twice, once first and once second.
- Practice
- 12 plain shapes with 2 repeats, followed by how many were caught. It runs again until both are caught, 3 times at most.
The procedure follows Arciuli and Simpson (2012), after Fiser and Aslin (2002) and Turk-Browne, Junge and Scholl (2005); the shapes are ConductScience's own. Many people do not score above chance on tests of this kind, and a visual version was retested at r = .58 (Siegelman, Bogaerts and Frost, 2017), so compare groups rather than individuals. The scores support research and do not diagnose.
Quality checks
Each session is marked valid, needs review or invalid, with the reason, before you see its scores.
Scored rows: statistical_learning_test from the scored run, one choice of the first or second group per trial, and the stream of the same run. Accuracy significantly below chance (one in two) invalidates and accuracy not above chance is flagged. Choices under 200 ms on at least half of the trials invalidate. The same answer on 90% of at least 12 choices, with accuracy not above chance, invalidates. Fewer than half the repeated shapes caught in the stream, or no stream, is flagged and withholds the percentile. A hidden tab withholds the percentile.
Why these rules
Accuracy significantly below chance (one-sided exact test, p < .05) cannot come from guessing; it means the response mapping was reversed or the task was not followed (Slick, Sherman and Iverson 1999 treat below-chance forced-choice performance as definite response bias). Chance-level accuracy can be a real result in impaired participants, so it is only flagged. Statistical learning is often slight, and many people do not score above chance, so chance-level choices are only flagged. One option on at least 90% of answered trials, when the correct options are balanced and accuracy is not above chance, means the answers did not depend on the items (the task counterpart of long-string responding in surveys; Curran 2016). The triplet comes first on half the trials, so one answer for nearly every trial cannot follow the shapes. An answer under 200 ms is faster than a simple visual reaction, so it cannot reflect the item. Following response-time effort screening for untimed tests (Wise and Kong 2005), the session is invalid only when at least half the answers are that fast. The choices rest on watching the stream, so a stream in which most repeats were missed leaves them uninterpretable (Arciuli and Simpson 2012 excluded participants who caught fewer than half). A hidden tab lets the study, delay or response timers run without the participant, so a percentile is withheld after one.
Running this test
Settings you can change
| Setting | Standard | Allowed |
|---|---|---|
| Presentations of each triplet | 24 | 16 to 32 |
| Time each shape is shown (ms) | 400 | 400 to 800 |
Yearly plans can change settings within these limits. Free and pay-as-you-go studies use the standard settings.
Need a setting this test does not have?
Devices and languages
- Untimed test: runs on desktops, laptops, tablets and phones. Participant devices
- Languages: English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese.
- Available on: Every plan, including the free plan.
- Recruiting online? See the setup guides for Prolific, Sona Systems, CloudResearch Connect and Qualtrics.
Cite and describe
Participants completed the Statistical Learning Task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Fiser & Aslin, 2002; Turk-Browne, Junge & Scholl, 2005), with the standard settings (presentations of each triplet: 24; time each shape is shown (ms): 400). The main score was triplet choice rate (share of test trials on which the triplet was chosen over the foil).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Fiser, J., & Aslin, R. N. (2002). Statistical learning of higher-order temporal structure from visual shape sequences. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(3), 458-467. https://doi.org/10.1037/0278-7393.28.3.458
- Turk-Browne, N. B., Jungé, J. A., & Scholl, B. J. (2005). The automaticity of visual statistical learning. Journal of Experimental Psychology: General, 134(4), 552-564. https://doi.org/10.1037/0096-3445.134.4.552The repeat check during the stream, and foils that keep each shape's place.
- Turk-Browne, N. B., Isola, P. J., Scholl, B. J., & Treat, T. A. (2008). Multidimensional visual statistical learning. Journal of Experimental Psychology: Learning, Memory, and Cognition, 34(2), 399-407. https://doi.org/10.1037/0278-7393.34.2.399The share of triplets chosen over foils as the score.
- Arciuli, J., & Simpson, I. C. (2012). Statistical learning is related to reading ability in children and adults. Cognitive Science, 36(2), 286-304. https://doi.org/10.1111/j.1551-6709.2011.01200.xThe procedure: timing, repeats and the test.
- Siegelman, N., Bogaerts, L., & Frost, R. (2017). Measuring individual differences in statistical learning: Current pitfalls and possible solutions. Behavior Research Methods, 49(2), 418-432. https://doi.org/10.3758/s13428-016-0719-zHow reliable the scores are, and when one person is above chance.
Source papers
- Arciuli J, Simpson IC (2012). Statistical learning is related to reading ability in children and adults. Cognitive Science, 36(2):286-304.Paradigm source, Children and adults
- Siegelman N, Bogaerts L, Frost R (2017). Measuring individual differences in statistical learning: current pitfalls and possible solutions. Behavior Research Methods, 49(2):418-432.Paradigm source, Adults
- ConductCognition platform normsParadigm source, N/A
Questions
How long does the Statistical Learning Task take?
About 8 to 10 minutes, as listed in the test library.
What device does a participant need?
Untimed test: runs on desktops, laptops, tablets and phones. The browser must be Chrome, Edge, Firefox or Safari.
What scores does it produce?
Triplet Choice Rate, Correct Choices, Above Chance, Median Choice Time, Repeats Caught, Presses Without a Repeat, Time to Catch a Repeat, Anticipations and Choices Scored. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Statistical Learning Task used in research?
210 papers from 2021 to 2025 name it in their title or abstract, up 21% compared with 2016 to 2020. That ranks it 120 of 395 test families in the ConductCognition test census (use grade C). Source: ConductCognition test census 0.14.0, counted 31 October 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Statistical Learning Task have?
In the published research, 5 papers report its test-retest reliability and 1 registered clinical study uses it as an outcome (evidence grade C in the ConductCognition test census).
Which settings can a researcher change?
Presentations of each triplet (16 to 32) and time each shape is shown (ms) (400 to 800). Settings can be changed on yearly plans; free and pay-as-you-go studies use the standard settings.
Which languages can participants use?
Participants can take this test in English, Spanish, French, German, Italian, Portuguese (Brazil), Chinese (Simplified) and Japanese. English is on every plan, and the other languages come with yearly plans. Have a native speaker read the machine-translated text before launch.
Is this a clinical test?
No. ConductCognition is for research use only. It runs, scores and exports the test; clinical interpretation is outside the platform.
Every test: the full test list.
Run this test in your next study
Start free with 10 free participants on any test, then use participant credits or a yearly plan.