ACM MM Author Response

Use this to write the ACM Multimedia rebuttal. It is optional and anonymous, it must not add new external links, and its real audience is the area chair who will weigh it against the reviews in discussion.

Triage before writing

Sort every review point into one of three buckets:

Bucket	Example	Response
Factual error	"No ablation of the fusion" (there is one, in §4.3)	Point to it precisely; this is your strongest move
Fixable gap	"Missing a late-fusion baseline"	Add the small result if you can run it in time
Judgment call	"The delta is incremental"	Reframe the significance briefly, or concede narrowly

Lead with the factual corrections — they are the points an AC can verify and act on fastest.

Structure for the AC

Write for a reader who is skimming several rebuttals. Make it navigable:

[R1, R2 shared concern: fusion not shown to matter]
  -> Table A (new): leave-one-modality-out; audio-text term = +X of the +Y total gain.
[R1 factual: "no user study"]
  -> §5 reports a user study, N raters, agreement kappa; see supplement S3.
[R3 judgment: "incremental over late fusion"]
  -> Our gain over the late-fusion baseline is in Table 1; the mechanism (Table A) is the novelty.
[R2 request: clarify synchronization assumption]
  -> Conceded; we will state it explicitly in §3 (one sentence).

Group shared concerns once; do not repeat the same answer per reviewer.

Small confirmatory results

• Add only results you can actually produce before the deadline — a leave-one-modality-out ablation, one requested baseline, or a small user-study extension.
• Describe new numbers in text/tables inside the rebuttal; you cannot add a link to an external file, and you cannot add pages to the paper.
• Do not promise experiments you cannot show; an AC discounts unbacked promises.

Calibrated concessions

Conceding a fair minor point raises credibility on the points you contest. State clearly what you will change (a sentence, a clarified assumption, a moved figure) and why the core claim survives it.

Policy guardrails

Anonymous: no author names, institutions, or identifying links.
No new external links: results go in the rebuttal text/tables, not a URL.
No scope change: you cannot rewrite the contribution or add pages.
No reviewer-targeting: never argue a reviewer is unqualified or try to identify them.

Length and format discipline

• Keep the rebuttal tight and scannable; an AC reading many responses rewards structure over volume, and a wall of text buries your best point.
• Put new evidence in a small in-line table, not prose — a reviewer can check a table in seconds.
• Respect any length limit the cycle sets, and do not try to smuggle in a link to sidestep it.

Rebuttal skeleton by review type

Strong-reject-on-a-misreading:  correct the fact FIRST, with an exact section/figure pointer.
Borderline-wants-a-baseline:    add the baseline in a mini-table; state it is now in the paper plan.
"Fusion is incremental":        show the mechanism ablation (the term that carries the gain).
"Subjective claim unproven":    report the user study's N, agreement, and result inline.
Fair-minor-point:               concede in one sentence and say exactly what changes.

Tone

Precise and non-defensive. Address the reviewer's underlying concern, not their wording; a rebuttal that reads as annoyed helps the negative reviewer more than you.

Output format

[Factual corrections] <ordered, with exact pointers>
[New results] <ablation/baseline/study added in-rebuttal>
[Concessions] <narrow, with the fix>
[Contested] <judgment points, reframed>
[Policy check] anonymous / no new links / no scope change — pass or fix