Error correction, revision, and learning

Small controlled classroom study of adult L2 writers (exact n not verified from the paper)

Adult second-language English learners

The canonical demonstration that the revision task and the new-writing task are different outcomes, and that a study measuring only the first will overstate what correction achieves. Learners wrote a narrative; one group had errors underlined and the other did not; both revised; a week later both wrote a NEW narrative. The design is simple and its point is structural rather than statistical. Recorded from the abstract and from secondary reports; the numbers behind the comparison were not verified against the paper, and the sample is adult L2, so it bounds nothing about children directly - its value to this archive is the dissociation it names.

The underlining group revised significantly better than the group whose errors were not marked - correction works on the text being corrected. On the new narrative written a week later, the two groups' change in error rate was "virtually identical". Error correction improved the corrected text and did not improve the next one. This is the sharpest available statement of the gap between fixing conventions in a piece of writing and learning conventions, and it is the direct counterpart to Van Beuningen et al. (2012), which found the opposite in a larger school-age sample - the two are the reason the written-corrective-feedback question is recorded here as contested rather than settled.

Not applicable in the usual sense - both groups wrote and revised, and the contrast is the presence or absence of underlining. The limitation is power and sample rather than selection.