Skip to main content

28. A Final Checklist

Before you submit, walk through this. Some of it overlaps with the official Responsible NLP Checklist; treat that as the canonical version.

A standalone, printable copy is at checklists/submission-checklist.md.

Content​

  • One-sentence research question is clear and stated explicitly somewhere in the introduction.
  • Contributions list in the introduction is concrete and honest.
  • Every section serves the central argument; nothing is there because you "had it lying around."
  • Figures and tables are readable in grayscale.
  • All claims are supported by experiments or citations.
  • Results are reported with variance, not single seeds.
  • Strongest available baselines are included; missing ones are justified.

Writing​

  • Read the full paper aloud at least once. Sentences that you stumble over need rewriting.
  • No undefined acronyms.
  • Notation is consistent throughout.
  • No "novel" or "we are the first" without verification.
  • Captions are paragraphs, not labels.

Structure​

  • Limitations section is present, substantive, and honest.
  • Ethical Considerations section is present if relevant.
  • References use the ACL Anthology version of each paper; every citation has been verified to exist.
  • Appendices contain supplementary material only; main claims are in the body.

Submission Logistics​

  • Anonymized: no author names or affiliations, and self-citations in the third person ("Smith (2023) showed," not "we showed in [Smith 2023]").
  • Acknowledgments and identifying funding/grant numbers removed from the review version (they go back in the camera-ready).
  • PDF metadata clean — your name is not in the file's document properties (check the Author field; this catches many otherwise-anonymous submissions).
  • No links to non-anonymous repositories; use an anonymized mirror (e.g. anonymous.4open.science) for code and data.
  • Page limits respected (Limitations, Ethics, References, Appendix do not count).
  • Responsible NLP Checklist filled out.
  • PDF compiled with the correct year's ACL style file.
  • Ran aclpubcheck on the PDF and fixed what it flags — it catches the margin, font, page-limit, and citation errors that trigger desk rejects (aclpubcheck --paper_type long paper.pdf). Ignore its line-number warnings on a review-version PDF.

The reviewer test​

Read the paper one final time pretending you are a tired reviewer with eight more papers to read. Where would you stop reading? That is where you have more work to do.

A closing word​

Your first ACL paper will not be perfect. It will not be the best paper in your career. That is okay — it does not need to be. It needs to be honest, clear, and complete. A paper that says one true thing clearly will earn you more respect than a paper that promises everything and delivers a moving target.

Show your draft to people who will tell you the truth. Revise more than you think you need to. Submit, and if it gets rejected, take the reviews seriously, improve the paper, and resubmit. Almost every senior researcher you admire has a stack of rejection letters that they have, eventually, learned more from than from their acceptances.

Good luck.