Students rarely rewrite specification comments: within the first 10 attempts, Unchanged (48.4%) and New (41.3%) dominate while substantial revisions total only 4.4%¶

Claim · All claims
Evidence · 1 study · associational r2 · q2
Currency · no dated study recorded · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Most comment writing and modification occurred early; within the first 10 attempts, Unchanged comments were 48.4%, New 41.3%, Deleted 6.0%, and Light Modification, Heavy Modification, and Full Rewrite together only 4.4%. → rq2-mod

Evidence¶

rq2-mod¶

Nasser Giacaman, Valerio Terragni, Paul Denny, and Viraj Kumar. 2026. Commenting with Copilot: A Taxonomy and Multi-Year Analysis of Student Code-Generation Specifications. Proceedings of the 2nd ACM Virtual Global Computing Education Conference (SIGCSE Virtual 2026). https://doi.org/10.1145/3795867.3830978

q2 · i? · associational · r2

Pooled modification analysis across all three exercises and four yearly cohorts (Figure 1), covering 124,201 extracted comments (91.0% overall) in the first 10 attempts, with modification levels assigned automatically using Levenshtein similarity thresholds. No effect size is printed.

"Within this window,Unchanged(48.4%) andNew(41.3%) dominate, followed byDeleted(6.0%);Light Modification,Heavy Modification, andFull Rewritetogether account for only 4.4%."

Discussion¶

Related Claims¶

• Diagnostic Review revisions contained larger shares of meaning-change, organization, and content/argument work, while Control revisions remained more dependent on surface-level editing — the general claim this finding bears on, with the studies behind it
• Comment revision effort concentrates on Multi-Step-Block and other procedural constructs, with the widest flows ending in Heavy Modification — a narrower finding that bears on this claim
• Low comment rewriting does not imply a smooth process: students describe the main effort as verifying generated code rather than revising prompts — related
• Student comments directing AI code generation are dominated by What comments, with How comments second and Why comments rare — related