Hypothesis
Published adult statistical word segmentation experiments, including the moderators said to shape this learning, would reproduce when re-run, with effect sizes close to the originals.
What failed
Across eight experiments replicating six studies, above-chance segmentation was confirmed every time, but many theoretically important moderations of learning did not replicate and learning was generally much weaker than originally reported. Several original papers also lacked key method details or described them wrongly.
Broken assumption
The moderator effects reported in the statistical word segmentation literature are robust rather than products of small, underpowered studies.
Why
The authors weigh differences in participant populations, low power in the original studies, or a mix of these and other factors.
From the published paper: Hartshorne et al. Collabra Psychology, 2019. ↗CC BY