HRI Author Response (Rebuttal)

HRI gives you one rebuttal, and it matters: because reviewers discuss and can revise scores after it, a focused rebuttal routinely changes decisions. But it is a conference rebuttal, not a journal revise-and-resubmit — you cannot run a new study before the deadline, and there is no in-cycle major-revision round. The job is to make the reviewers who read your paper feel their concerns were heard and, where they are genuinely wrong, corrected. Timing for the HRI 2026 cycle: rebuttal due ~12 Nov 2025 (see resources/official-source-map.md); confirm the length/format limit in the current instructions (待核实).

Before writing: triage every point

For each concern across the three externals and two ACs, sort into:

• Misreading — the reviewer misunderstood something already in the paper. Correct it crisply, quoting where it is (section/figure), without blaming the reviewer.
• Fixable in camera-ready — a clarification, an added analysis you already have, a missing citation, a reframed claim. Promise the specific change and, where possible, give the result now.
• Real limitation you cannot fix now — a design choice, a sample bound, a construct-validity concern. Acknowledge it honestly, argue why it does not sink the contribution, and scope the claim accordingly. Do not pretend it away.

Then rank by decisiveness: which concern, if resolved, actually flips a reviewer or wins the discussion. Spend your limited space there, not on the easy points.

Answer the HRI-typical critiques

HRI reviews cluster on study and interaction issues; have crisp answers ready:

• "Underpowered / N too small." Report your power basis, effect sizes with CIs already in the paper, and — if applicable — that the observed effect is large enough to matter. Do not promise "we will collect more data" (you cannot before camera-ready); argue the evidence you have.
• "Confound / demand characteristics." Point to the manipulation check, counterbalancing, or controls already present; if a real confound exists, bound its plausible effect honestly.
• "Wizard-of-Oz not convincing / undisclosed." Clarify what was autonomous vs. wizard-controlled and the wizard constraints/error logging; if you buried it, promise to foreground it.
• "Just a liking score, not behavior." Point to any behavioral/task outcome you measured; if the claim overreached the measure, narrow the claim rather than defend the overreach.
• "Statistics questionable." Address multiple-comparison correction, the confirmatory/exploratory split, and assumption checks; commit to a re-analysis in camera-ready if warranted.
• "Novelty / positioning." State the delta against the specific prior work cited, without breaking anonymity (third person).
• "Ethics unclear." Confirm IRB/consent and (anonymized) deception/debriefing details.

Promise only what the camera-ready can deliver

The rebuttal's currency is credibility. Every promise should be something you can actually do in the final version within the 8-page budget: rephrase a claim, add an already-computed analysis or effect size, foreground a disclosure, add a citation, clarify a figure. Never promise a new study, new data collection, or a result you have not computed — reviewers discount vague promises and remember broken ones.

Structure and constraints

• Group by concern, not by reviewer, when multiple reviewers raise the same issue — answer it once, well, and note it addresses R1/R3.
• Lead with the most decisive points; a tight limit means the last third may be skimmed.
• Be concrete and quantitative: quote the section, give the number, name the exact change.
• Stay within the length limit and any format rules; over-length rebuttals can be truncated.
• Stay anonymous — no author/institution names, no identifying links, no "as we showed in [our other paper]" in first person.
• Professional tone. Thank reviewers briefly, never argue with their competence or tone; give the sympathetic reviewer material to defend you in the discussion.

What not to do

• Do not concede your central contribution to sound agreeable — if the core claim is sound, defend it.
• Do not sprawl across every minor point and run out of room for the decisive one.
• Do not introduce a brand-new claim or result the reviewers had no chance to vet.
• Do not break anonymity to prove a point.

After the rebuttal

The 1AC opens an online discussion; you will not see it. Your rebuttal is the last input you control before the PC meeting, so make it self-sufficient. If accepted (possibly with shepherding), carry every promise into hri-camera-ready; if rejected, mine the reviews and re-route with hri-topic-selection.

Output format

[Triage] concerns sorted: misreading / camera-ready-fixable / real-limitation
[Decisive points] the 2-3 objections the rebuttal must win, per reviewer
[Responses] each: correction or promised change, concrete + quantitative
[Promises] all deliverable in camera-ready within 8 pages? no new-study promises?
[Anonymity + length] third-person, within limit?
[Draft] <the rebuttal, grouped by concern, decisive-first>