research-paper-writing

Follow an iterative pipeline: define a one-sentence contribution, map every experiment to a claim, run baselines with statistical significance tests, then draft using the official venue LaTeX template. Finish with a simulated self-review and the venue's mandatory checklist before submitting.

Never generate BibTeX from memory — fetch it programmatically via DOI content negotiation from CrossRef, and verify each paper exists in at least two sources such as Semantic Scholar and arXiv. Mark anything unverifiable as a placeholder citation for manual review.

Report error bars (stating standard deviation vs standard error), 95% confidence intervals via bootstrapping, McNemar's test for paired method comparisons, and effect sizes such as Cohen's h. Always state sample sizes and number of independent runs.

NeurIPS allows 9 pages, ICML 8 pages, ICLR 9 pages, AAAI 7 pages, and COLM 9 pages for main content, with references and appendices unlimited at all of these venues. Camera-ready versions typically gain one additional page.

Use human evaluation when automated metrics cannot capture your claims, such as fluency, helpfulness, or preference in generation tasks — ACL and EMNLP reviewers expect it for such work. Report annotator counts, inter-annotator agreement (Krippendorff's alpha), and compensation details.

Yes, if the analysis of why the hypothesis failed is rigorous and informative. Venues such as NeurIPS Datasets & Benchmarks, TMLR, and dedicated workshops explicitly welcome well-executed negative results with airtight methodology.