ijcai-writing-style

IJCAI Writing Style

Use this when revising the main paper. IJCAI rewards clear, significant, original AI work

that a broad program committee can evaluate quickly.

Revision rules

• Put the AI contribution in the first page: problem, gap, core idea, evidence, and why it

matters beyond a narrow benchmark.

• Keep the submission self-contained; supplementary material should support, not rescue, the

main argument.

• Spend the 7-page body on claims reviewers can verify: formal setup, method, theorem or

algorithm, baselines, ablations, and limitations.

• Use third-person citation language for prior work by the authors and remove

acknowledgements, contribution statements, and identity-revealing phrasing from the review

version.

• Include an ethics or broader-impact statement when sensitive data, high-stakes deployment,

safety, discrimination, privacy, consent, copyright, or societal harm is plausible.

• Treat LLM assistance conservatively. Style polishing may be acceptable in some cycles, but

LLM-written manuscript text, fake citations, or unverified claims can trigger rejection.

Writing for a cross-subfield committee

IJCAI's PC mixes symbolic AI, search, planning, KR, constraint reasoning, multi-agent, game

theory, ML, NLP, and vision. Write so a reviewer outside your niche reaches the contribution

fast, because the out-of-area reviewer often drives summary-reject risk.

| Reader risk | Symptom in the draft | Page-discipline fix |

| --- | --- | --- |

| Out-of-area reviewer lost early | Contribution buried after a long formal preamble | State problem, gap, and claim on page one |

| Notation overload | New symbols before motivation | Introduce notation only when the claim needs it |

| Niche jargon | Subfield acronyms undefined | Define on first use; assume a broad AI reader |

| Body overflow | Method spills toward the reference pages | Move proofs and grids to the appendix, keep claims in the body |

Worked vignette: framing a KR paper for breadth

A knowledge-representation paper opens with three paragraphs of description-logic preliminaries

before naming its contribution. A planning or learning reviewer may summary-reject for unclear

significance. The fix: open with the reasoning task, why current encodings fall short, and the

new tractable fragment, then defer the heavy formalism. The technical depth stays; only the

ordering changes so a broad IJCAI reader sees the payoff within two reads.

Reviewer pushback and the venue-specific fix

• "Could not find the contribution." Rewrite the first page as problem, gap, idea, evidence,

significance; this is the single highest-leverage IJCAI edit.

• "Reads like a journal extension." Compress to the body limit and cite the longer work in

third person rather than reproducing it.

• "Significance only within one benchmark." Add a sentence on why the result matters across AI,

not just on the chosen suite.

Output format

[Writing diagnosis] clear / overloaded / underspecified / policy-risky
[First-page fix] <new framing>
[Compression cuts] <what to move or delete>
[Anonymity edits] <phrases to rewrite>
[Ethics/reproducibility edits] <specific additions>