Verb Selectivity as Informative Context: Testing the Propose-but-Verify Mechanism in Cross-Situational Word Learning Open Access
Zeng, Xiaoyi (Spring 2026)
Abstract
Cross-situational word learning (CSWL) demonstrates that learners can extract stable word–referent mappings from ambiguous input across multiple encounters. However, the mechanism underlying this phenomenon remains debated. The predominant assumption—that learners gradually accumulate co-occurrence statistics across trials—has been challenged by evidence that highly informative (HI) moments disproportionately shape learning (Medina et al., 2011; Trueswell et al., 2013). The propose-but-verify (PBV) model offers an alternative account in which learners generate single candidate hypotheses that are either confirmed or discarded based on subsequent evidence. Prior work has manipulated informativeness exclusively in the visual domain; whether timing-sensitive informative moments generalize to linguistic context has not been tested.
The present study addresses this gap by introducing verb selectivity as a novel manipulation of informativeness in CSWL. Twenty-one adult participants learned 20 nonce noun–referent mappings across five learning blocks. Each mapping appeared with either a high-selectivity (HI) or low-selectivity (LI) verb, and the timing of the HI verb was varied within-subject across four conditions: HI-first, HI-middle, HI-last, and No-HI. Visual ambiguity was held constant throughout, isolating verb selectivity as the sole source of variation in informativeness.
Results provided no evidence for statistical accrual: overall accuracy did not increase monotonically, and accuracy following incorrect guesses did not exceed baseline. By contrast, verb informativity significantly shaped learning trajectories. A mixed-effects logistic regression revealed a significant Learning Instance and Condition interaction, with accuracy peaks aligning with the position of the HI verb across conditions. However, condition differences did not remain significant at the final cumulative test. Exploratory analyses further showed that the diagnostic value of first-trial correctness differed depending on whether the first exposure was HI or LI, suggesting that linguistic informativity may change not only whether learners succeed but what that success reflects.
These findings argue against statistical accrual as a necessary mechanism for CSWL and are more consistent with a process-oriented account of propose-but-verify, in which informative moments disproportionately shape when successful hypotheses emerge and how strongly they carry forward. The study extends prior work by demonstrating that linguistic context—not only visual context—can serve as a meaningful source of informativity in word learning, and that similar final outcomes may conceal fundamentally different learning dynamics.
Table of Contents
Abstract
Table of Contents
1 Introduction
1.1 Referential indeterminacy, CSWL as a phenomenon, and the problem of mechanism
1.2 Statistical accrual as the predominant assumption
1.3 Limits of statistical accrual and PBV as an alternative
1.4 Why linguistic informativity? Verb selectivity as a manipulation
1.5 The current study
1.6 Research questions and predictions
2 Method
2.1 Participants
2.2 Materials and Study Design
2.3 Trial Structure, Competitor Rotation, and Filler Balancing
2.3.1 Competitor rotation (block-wise shift)
2.3.2 Filler balancing
3 Results
3.1 Is there evidence of statistical accrual?
3.1.1 Overall performance did not show gradual improvement across learning
3.1.2 Accuracy following previous incorrect guesses did not exceed baseline
3.1.3 Follow-up: no clear evidence that previous HI trials improved later responding
3.2 Do HI moments defined by verb selectivity matter?
3.2.1 Verb informativity affected accuracy in the moment
3.2.2 Condition differences did not remain significant at the final cumulative test
3.3 Exploratory analysis: later accuracy as a function of first-trial selectivity and correctness
4 General Discussion
4.1 Is there evidence of statistical accrual?
4.2 Does verb informativity matter, and how?
4.3 Broader implications, limitations, and future directions
5 Conclusion
References
Method Appendix
Appendices
About this Honors Thesis
| School | |
|---|---|
| Department | |
| Degree | |
| Submission | |
| Language |
|
| Research Field | |
| Keyword | |
| Committee Chair / Thesis Advisor | |
| Committee Members |
Primary PDF
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
|
Verb Selectivity as Informative Context: Testing the Propose-but-Verify Mechanism in Cross-Situational Word Learning () | 2026-04-22 18:04:39 -0400 |
|
Supplemental Files
| Thumbnail | Title | Date Uploaded | Actions |
|---|---|---|---|
|
Data Analysis (R code for data analysis) | 2026-04-08 00:52:19 -0400 |
|
|
|
Memorability Scores (Memorability scores for object pictures) | 2026-04-08 00:52:29 -0400 |
|
|
|
Schedule HI-first (Schedule for HI-first condition experiment) | 2026-04-08 00:52:40 -0400 |
|
|
|
Schedule HI-last (Schedule for HI-last condition experiment) | 2026-04-08 00:52:51 -0400 |
|
|
|
Schedule HI-middle (Schedule for HI-middle condition experiment) | 2026-04-08 00:53:00 -0400 |
|
|
|
Schedule no-HI (Schedule for no-HI condition experiment) | 2026-04-08 00:53:13 -0400 |
|
|
|
Table A1 (Table A1 for appendix) | 2026-04-08 00:53:22 -0400 |
|
|
|
Table A2 (Table A2 for appendix) | 2026-04-08 00:53:32 -0400 |
|
|
|
Table B3 (Table B3 for appendix) | 2026-04-08 00:53:41 -0400 |
|
|
|
Table C3 (Table C3 for appendix) | 2026-04-08 00:53:50 -0400 |
|
|
|
Experiment Data (Raw data from experiment) | 2026-04-08 00:54:10 -0400 |
|