| name | emnlp |
| description | Writing and formatting guidance for EMNLP (Conference on Empirical Methods in Natural Language Processing) papers. Use when drafting, formatting, or submitting to EMNLP 2026. EMNLP shares the acl.sty template with ACL. Covers review/final mode switching, double-blind anonymity, EMNLP-specific submission policies, camera-ready formatting, and NLP-specific writing conventions including datasets, baselines, and error analysis. See the ACL skill for shared formatting rules. |
EMNLP 2026 Conference Writing Skill
Overview
This skill covers LaTeX formatting, submission rules, and writing conventions for EMNLP (Conference on Empirical Methods in Natural Language Processing) — a CCF-B conference in the AI category, held annually at https://2026.emnlp.org. The publisher is ACL.
Prerequisite: EMNLP shares the same acl.sty template as ACL, NAACL, COLING, and CoNLL. Read the ACL skill first for shared formatting rules (documentclass, packages, citations, bibliography, figures, tables, equations, appendix). This skill covers EMNLP-specific rules and differences only.
Template path: ccf-latex-templates/EMNLP/
acl_latex.tex — main template (identical to ACL/NAACL)acl.sty — shared style file
EMNLP-Specific Rules
Document Class and Modes
EMNLP uses the same three-mode system as ACL:
% Submission (anonymized review)
\documentclass[11pt]{article}
\usepackage[review]{acl}
% Camera-ready (final)
\documentclass[11pt]{article}
\usepackage{acl}
% Non-anonymous preprint
\documentclass[11pt]{article}
\usepackage[preprint]{acl}
Standard Packages
Identical to ACL:
\usepackage{times}
\usepackage{latexsym}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{microtype}
\usepackage{inconsolata}
\usepackage{graphicx}
If the title/author block overflows:
\setlength\titlebox{6cm} % minimum is 5cm
Title and Author
\title{Your EMNLP Paper Title}
\author{First Author \\
Institution, City, Country \\
\texttt{email@domain} \\\And
Second Author \\
Institution, City, Country \\
\texttt{email@domain} \\}
In review mode, acl.sty automatically replaces the author block with "Anonymous ACL submission".
Page Limit
EMNLP 2026: 8 pages for the main paper, excluding references and appendix.
Confirm the exact limit in the official EMNLP 2026 Call for Papers, as policies can change year-to-year.
Submission Policies
Double-Blind Review
- Submission must be anonymized: no author names, affiliations, or acknowledgments that reveal identity.
- In review mode,
acl.sty automatically shows "Anonymous ACL submission" as the author block.
- Self-citations: Remove or anonymize citations to prior work by the authors. Replace with "[REMOVED FOR REVIEW]" or similar placeholder, then restore for the camera-ready.
- Do not post anonymized drafts to arXiv or public repositories before the review period ends.
- Do not anonymize citations to non-author prior work.
Originality and Reproducibility
- Submissions must be original, unpublished work not currently under review elsewhere.
- EMNLP increasingly emphasizes reproducibility. Include:
- Hyperparameters and training details in appendices.
- Links to code repositories (anonymized if submitted as supplementary).
- Results averaged over multiple random seeds where applicable.
Conflicts of Interest
- Declare all conflicts of interest in the submission system.
- Do not include conflict declarations in the PDF.
Multiple Submission Policy
- Do not submit the same work to EMNLP and another venue simultaneously.
- Withdrawal and desk-rejection policies follow EMNLP guidelines.
Camera-Ready Preparation
After acceptance, prepare the camera-ready version:
NLP Writing Conventions for EMNLP
EMNLP papers typically follow this structure:
- Abstract — 150–200 words, state problem/method/results
- Introduction — motivation, task, challenge, contribution list (4–5 bullet points)
- Related Work — prior work categorized by approach or task
- Background / Preliminaries — formal notation, task definition
- Method / Model — core contribution with enough detail to reproduce
- Experimental Setup — datasets, metrics, baselines, implementation details
- Results and Analysis — quantitative results, ablation studies, error analysis, qualitative examples
- Conclusion — summary, limitations, broader impact
- Acknowledgments
- References
- Appendix
Recommended EMNLP Introduction Structure
\section{Introduction}
% Paragraph 1: Problem and motivation
Natural language processing has seen significant advances with...
However, the task of X remains challenging due to Y...
% Paragraph 2: Gap in prior work
Prior work has addressed this problem through approaches A, B, and C.
While effective, these approaches suffer from limitations:
(1) They require large amounts of labeled data...
(2) They fail to capture semantic nuances...
(3) They are not generalizable to new domains...
% Paragraph 3: Our approach
In this paper, we propose a novel approach that...
Our key insight is that...
% Paragraph 4: Contributions (numbered list)
Our main contributions are:
\begin{itemize}
\item We propose a new model architecture for X that achieves Y...
\item We introduce a benchmark dataset consisting of Z...
\item We demonstrate through extensive experiments that our method
outperforms prior approaches by X\% on standard benchmarks.
\end{itemize}
EMNLP Writing Conventions
Dataset Description (Required)
\section{Experimental Setup}
\subsection{Datasets}
We evaluate on four standard benchmarks:
\begin{table}[h]
\caption{Dataset statistics. All datasets are English unless noted.}
\label{tab:datasets}
\centering
\begin{tabular}{lrrrr}
\toprule
Dataset & Train & Dev & Test & Source \\
\midrule
CoNLL-2003 & 14,987 & 3,466 & 3,684 & Reuters \\
Ontonotes & 75,722 & 8,321 & 8,263 & Various \\
\bottomrule
\end{tabular}
\end{table}
\textbf{Preprocessing:} Tokenized using spaCy, lowercased.
Rare entities (<5 occurrences) replaced with <UNK>.
Baseline Selection
We compare our approach against the following baselines:
\textbf{Standard baselines:}
\begin{itemize}
\item \textbf{Baseline A}: The method of \citet{smith2023}.
\item \textbf{Baseline B}: Vanilla BERT fine-tuning \citep{devlin2019}.
\end{itemize}
\textbf{State-of-the-art:}
\begin{itemize}
\item \textbf{Method C}: Recent state-of-the-art from \citet{jones2024}.
\item \textbf{Method D}: The approach of \citet{brown2022}.
\end{itemize}
Metric Reporting
\subsection{Metrics}
We report standard NLP metrics for each task:
\textbf{Sequence labeling:} Token-level F1 score (CoNLL-style).
\textbf{Text classification:} Accuracy, macro F1 score.
\textbf{Generation:} BLEU \citep{papineni2002bleu}, ROUGE-L \citep{lin2004rouge},
METEOR \citep{banerjee2005meteor}.
All scores are computed using the official evaluation scripts.
Statistical Significance
\subsection{Statistical Significance}
We assess statistical significance using paired t-tests
(5 seeds, $\alpha = 0.05$) following \citet{berg2012empirical}.
Results marked with $\dagger$ indicate $p < 0.05$ vs. the best baseline.
Error Analysis
\subsection{Error Analysis}
We manually analyze 200 randomly sampled errors.
\textbf{Main error categories:}
\begin{itemize}
\item \textbf{Long-range dependencies} (34\%): The model struggles
with dependencies spanning more than 512 tokens.
\item \textbf{Rare tokens} (28\%): OOV words that appear fewer
than 5 times in training.
\item \textbf{Ambiguous cases} (22\%): Examples where even human
annotators disagree.
\item \textbf{Annotation noise} (16\%): Likely annotation errors in
the dataset.
\end{itemize}
Key EMNLP Expectations
- Datasets: Describe source, size, language, domain, preprocessing, licensing. Include data splits (train/dev/test).
- Baselines: Compare against strong, well-established baselines (not strawmen).
- Metrics: Report standard NLP metrics (accuracy, F1, BLEU, ROUGE-L, BERTScore, etc.). Include confidence intervals or statistical significance tests.
- Ablation: Include ablation studies to isolate the effect of each component.
- Error Analysis: Provide qualitative examples of successes and failures. Show representative errors.
- Limitations: Discuss limitations honestly (model assumptions, computational cost, generalizability, bias).
References
Same as ACL — use acl_natbib:
\bibliographystyle{acl_natbib}
\bibliography{anthology,your_file} % include Anthology bibliography
Download the ACL Anthology BibTeX: https://aclweb.org/anthology/anthology.bib.gz.
Figures and Tables
Same as ACL:
\begin{figure}[t]
\includegraphics[width=\linewidth]{figure}
\caption{Caption here.}
\label{fig:label}
\end{figure}
Use PDF, PNG, or EPS. Match fonts to document. Do not override caption sizes.
For tables, use booktabs-style rules (no vertical lines). See ACL skill for examples.
Supplementary Materials
Check the EMNLP 2026 CFP for supplementary material policy. Typical rules:
- Supplementary files (code, data, proofs, additional experiments) are not reviewed.
- Label supplementary files clearly and reference them from the main text.
- Do not exceed file size limits.
- Anonymize any supplementary code or data before uploading.
Supplementary Checklist
% In main paper:
Detailed proofs are provided in Appendix A.
Extended experimental results are in Appendix B.
Our code is available at [anonymous URL].
% In supplementary:
\section*{Appendix A: Proofs}
\section*{Appendix B: Extended Experiments}
\section*{Appendix C: Hyperparameters}
Checklist Summary
