Language
Cross-Situational Word Learning
Also called: CSWL, cross-situational learning, cross-situational statistical learning, statistical word learning, observational word learning
Measure statistical word learning: participants see groups of new shapes with made-up words and work out which word names which shape from what comes together.
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
- 101
- Evidence grade
- D
- Registered clinical studies
- 0
What it measures
Measures how people learn what words mean from what goes together across many moments. Participants see groups of new shapes while made-up words appear one at a time, with nothing to say which word names which shape; each word always comes with its own shape. Then they pick each of the 18 words' shapes from four. The main score is the percent of words matched to their shapes.
Constructs
- Statistical word learning
- Word learning
- Learning under ambiguity
- Associative learning
Research uses
- Language development and language disorder research
- Statistical learning research
- Bilingualism and second-language research
How the paradigm works
Cross-situational word learning shows how people learn what words mean when no single moment says which word names which thing. In the method of Yu and Smith (2007), people see several new objects and meet several new words at once, with nothing to link a word to an object. Across many such moments, each word comes with its own object every time and with the others only now and then, and people learn many of the pairings from that pattern alone; infants do it too (Smith and Yu, 2008). The method is used to study the mechanisms of statistical word learning, how long what is learned lasts (Vlach and Sandhofer, 2014) and how learning differs across learners, including children with developmental language disorder, adults with aphasia and bilingual speakers (Roembke and colleagues, 2023).
More about the paradigm
This version follows the design of Yu and Smith (2007): 18 made-up words and 18 new shapes drawn by ConductScience's code, each word in 6 groups, 3 seconds per word. In the standard version each group shows 4 shapes and 4 words, 27 groups in about 5.5 minutes; researchers can set 2 or 3 per group instead, which makes each group less ambiguous. Most studies speak the words. This version writes them, one at a time between the rows of shapes, in Latin letters in seven languages and in katakana in Japanese, so it needs no sound; in a study that compared written and spoken words, online and in the laboratory, adults recognized the written ones more accurately (Escudero, Smit and Angwin, 2023). No two words share more than 3 groups, so each word's own shape is the shape it comes with most often. Participants are told that each word names one of the shapes on the screen, not how to find which. After the groups, each word is shown with four shapes, its own and three of other words, each shape a wrong option equally often (Suanda and Namy, 2012), and participants click or tap one, with no feedback and no time limit. A practice of three groups and two questions with feedback comes first, and which word goes with which shape is drawn for each session.
The main score is the percent of the 18 words matched to their shapes; 25% is chance. The task also reports the number of words matched, the median choice time and choices under 100 ms (anticipations, left out of the scores). Each test row records how many groups every option shared with the word, for analyses of how those shared groups draw wrong choices (Suanda and Namy, 2012). Scores support research and do not diagnose.
Scores
- Percent Correct%
- Percent of the 18 words matched to their shapes at the test, leaving out choices under 100 ms; 25% is chance.
- Higher is better
- Words Matchedcount
- Words matched to their shapes at the test, of 18.
- Higher is better
- Median Choice Timems
- Median time of the test choices, leaving out those under 100 ms.
- Lower is faster
- Anticipationscount
- Test choices under 100 ms, left out of every score.
- Quality check
- Choices Recordedcount
- Test choices recorded in the run.
- Fewer than planned invalidates the session
- Choices Plannedcount
- Test choices the session planned: every word once.
- Context for the scores
Use and evidence
101 papers from 2021 to 2025 name this task in their title or abstract, down 4% compared with 2016 to 2020. That ranks it 185 of 395 test families in our census. See the counts for every test.
Show the numbers
| Year | Papers |
|---|---|
| 2025 | 26 |
| 2024 | 23 |
| 2023 | 18 |
| 2022 | 19 |
| 2021 | 15 |
| 2020 | 25 |
| 2019 | 18 |
| 2018 | 19 |
| 2017 | 23 |
| 2016 | 20 |
| 2015 | 12 |
| 2014 | 14 |
| 2013 | 16 |
| 2012 | 19 |
| 2011 | 13 |
| 2010 | 16 |
| 2009 | 11 |
| 2008 | 2 |
| 2007 | 3 |
| 2006 | 4 |
| 2005 | 2 |
| 2004 | 1 |
| 2003 | 0 |
| 2002 | 0 |
| 2001 | 0 |
| 2000 | 0 |
Use grade C
100 to 299 papers from 2021 to 2025.
Evidence grade D
Little published measurement evidence found.
- Papers on test-retest reliability
- 0
- Papers on reliability or validity
- 9
- Registered clinical studies using it as an outcome
- 0
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 2 November 2026 from OpenAlex and ClinicalTrials.gov.
Task design
The trial sequence as the test runs in ConductCognition.
- Words and shapes
- 18 made-up words of three syllables and 18 shapes drawn by code in 6 colors, plain, striped or dotted. The words are written: in Latin letters, and in katakana in Japanese sessions. Which word names which shape is drawn for each session.
- Groups
- 4 shapes and 4 words a group (standard; 2 to 4). Each word comes in 6 groups, always with its own shape, and no two words share more than 3 groups: 27 groups at the standard, 54 with 2 and 36 with 3.
- Timing
- The shapes stay for 3 seconds per word, and each word appears between the rows of shapes 500 ms into its 3 seconds, one at a time, with a 500 ms blank between groups. No sound.
- Test
- Each word once, with 4 shapes: its own and 3 of other words, each shape a wrong option 3 times. A click or tap, with no time limit and no feedback.
- Practice
- 3 groups of 2 shapes, then 2 words, each with 3 shapes and the right shape shown for 2,000 ms.
- Anticipations
- A choice under 100 ms is an anticipation and is left out.
The method follows Yu and Smith (2007). Most studies speak the words; this version writes them, so it needs no sound. The words and shapes are ConductScience's own. A 2023 review found no data on how reliably a typical cross-situational task measures one person, so compare groups rather than individuals. 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: cross_situational_test from the scored run, one choice among four shapes for every word. Accuracy significantly below chance (one in four) invalidates and accuracy not above chance is flagged without withholding the percentile. Choices under 200 ms on at least half of the words invalidate. The same place on 90% of at least 12 choices, with accuracy not above chance, invalidates. Fewer test choices recorded than planned invalidate. 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. Learning from co-occurrence alone can be slight, and many adults learn only some of the words, 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 right shape takes each of the four places about equally often, so one place chosen for nearly every word cannot follow the words. 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. A session with missing test choices cannot be compared with complete ones. 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 |
|---|---|---|
| Words and shapes per group | 4 | 2 to 4 |
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 Cross-Situational Word Learning task in a web browser (ConductCognition task version 1.0.0, scoring version 1.0.0; Yu & Smith, 2007; Smith & Yu, 2008), with the standard settings (words and shapes per group: 4). The main score was percent correct (percent of the 18 words matched to their shapes at the test, leaving out choices under 100 ms; 25% is chance).
ConductCognition versions: task version 1.0.0, scoring version 1.0.0. Every export records the versions each session used, for a methods section.
- Yu, C., & Smith, L. B. (2007). Rapid word learning under uncertainty via cross-situational statistics. Psychological Science, 18(5), 414-420. https://doi.org/10.1111/j.1467-9280.2007.01915.xThe paradigm: 18 words in 6 groups each, 2 to 4 words a group, 3 seconds a word, and a four-choice test.
- Smith, L., & Yu, C. (2008). Infants rapidly learn word-referent mappings via cross-situational statistics. Cognition, 106(3), 1558-1568. https://doi.org/10.1016/j.cognition.2007.06.010Cross-situational learning in infants.
- Suanda, S. H., & Namy, L. L. (2012). Detailed behavioral analysis as a window into cross-situational word learning. Cognitive Science, 36(3), 545-559. https://doi.org/10.1111/j.1551-6709.2011.01218.xEvery picture a wrong option equally often, and how often each option shared a group with the word.
- Vlach, H. A., & Sandhofer, C. M. (2014). Retrieval dynamics and retention in cross-situational statistical word learning. Cognitive Science, 38(4), 757-774. https://doi.org/10.1111/cogs.12092The 2, 3 and 4 word groups at equal learning time, and retention a week later.
- Escudero, P., Smit, E. A., & Angwin, A. J. (2023). Investigating orthographic versus auditory cross-situational word learning with online and laboratory-based testing. Language Learning, 73(2), 543-577. https://doi.org/10.1111/lang.12550Written words recognized more accurately than spoken ones, in a study run online and in the laboratory.
- Kachergis, G., Yu, C., & Shiffrin, R. M. (2014). Cross-situational word learning is both implicit and strategic. Frontiers in Psychology, 5, 588. https://doi.org/10.3389/fpsyg.2014.00588Learning with and without the instruction to learn the words.
- Roembke, T. C., Simonetti, M. E., Koch, I., & Philipp, A. M. (2023). What have we learned from 15 years of research on cross-situational word learning? A focused review. Frontiers in Psychology, 14, 1175272. https://doi.org/10.3389/fpsyg.2023.1175272Review, including the open question of how reliably the task measures one person.
Source papers
- Yu C, Smith LB (2007). Rapid word learning under uncertainty via cross-situational statistics. Psychological Science, 18(5):414-420.Paradigm source, Adults
- ConductCognition platform normsParadigm source, N/A
Questions
How long does the Cross-Situational Word Learning 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?
Percent Correct, Words Matched, Median Choice Time, Anticipations, Choices Recorded and Choices Planned. Each session is also marked valid, needs review or invalid, with the reason.
How widely is the Cross-Situational Word Learning used in research?
101 papers from 2021 to 2025 name it in their title or abstract, down 4% compared with 2016 to 2020. That ranks it 185 of 395 test families in the ConductCognition test census (use grade C). Source: ConductCognition test census 0.14.0, counted 2 November 2026 from OpenAlex and ClinicalTrials.gov.
How much reliability evidence does the Cross-Situational Word Learning have?
In the published research, 0 papers report its test-retest reliability and 0 registered clinical studies use it as an outcome (evidence grade D in the ConductCognition test census).
Which settings can a researcher change?
Words and shapes per group (2 to 4). 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. Its words are made up and the same in every language, written in Latin letters, or in katakana in Japanese sessions. 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.