| name | acl |
| description | Writing and formatting guidance for ACL (Association for Computational Linguistics) conference papers. Use when drafting, formatting, or submitting papers to ACL 2026. Covers the acl.sty class with review/final/preprint modes, Times font and microtype setup, title box adjustment for oversized headers, natbib citation commands, appendix formatting, and double-blind anonymity rules. This skill applies equally to EMNLP, NAACL, COLING, and CoNLL which share the same acl.sty template. |
ACL Conference Writing Skill
Overview
This skill covers LaTeX formatting, submission rules, and writing conventions for ACL (Association for Computational Linguistics) — a CCF-A conference in the AI category, held annually at https://2026.aclweb.org. The publisher is ACL.
Note: EMNLP, NAACL, COLING, and CoNLL all share the same acl.sty template. Consult this skill for shared formatting rules, then check the EMNLP or NAACL skill for venue-specific policies.
Template path: ccf-latex-templates/ACL/
acl_latex.tex — main template
acl.sty — style file (sourced from acl-style-files)
Document Class and Packages
Modes
The acl.sty supports three modes, controlled via the acl package option (not a class option):
| Mode | Option | Effect |
|---|
| Submission (anonymized) | \usepackage[review]{acl} | Anonymous author block, line numbers, page numbers |
| Camera-ready | \usepackage{acl} (no option) | Final copy, no line/page numbers |
| Non-anonymous preprint | \usepackage[preprint]{acl} | Page numbers shown, authors visible |
% Submission
\documentclass[11pt]{article}
\usepackage[review]{acl}
% Camera-ready (final)
\documentclass[11pt]{article}
\usepackage{acl}
% Preprint
\documentclass[11pt]{article}
\usepackage[preprint]{acl}
Standard Packages
Include these packages in every submission (review and final):
\usepackage{times} % Times Roman font
\usepackage{latexsym} % LaTeX symbols
\usepackage[T1]{fontenc} % T1 font encoding (required for proper hyphenation)
\usepackage[utf8]{inputenc} % UTF-8 source encoding
\usepackage{microtype} % Improved typography, saves space
\usepackage{inconsolata} % Monospace font for code
\usepackage{graphicx} % Figures and images
Title Box
If the title and author block overflows, increase \titlebox:
\setlength\titlebox{6cm} % or larger; must be >= 5cm
Title and Author
\title{Your Paper Title Here}
\author{First Author \\
Affiliation / Address line 1 \\
Affiliation / Address line 2 \\
\texttt{email@domain} \\\And
Second Author \\
Affiliation / Address line 1 \\
Affiliation / Address line 2 \\
\texttt{email@domain} \\}
For authors from different institutions, use \And between groups:
\author{Author 1 \\ Institution 1 \\\texttt{email1}
\And
Author 2 \\ Institution 2 \\\texttt{email2}}
To start a separate row of authors, use \AND:
\author{Row 1 Author 1 \AND Row 1 Author 2 \\
Row 2 Author 3 \And Row 2 Author 4}
In review mode, acl.sty automatically replaces the author block with "Anonymous ACL submission". Do not manually anonymize.
Document Structure
Standard NLP Paper Structure
\begin{document}
\maketitle
\begin{abstract}
Write your abstract here. Keep it concise (typically 150–250 words).
State the problem, approach, and main results.
\end{abstract}
\section{Introduction}
% Problem, motivation, contributions (explicit bullet list)
\section{Related Work}
% Categorize by approach or task, not just list
\section{Background / Preliminaries}
% Formal notation, task definition, background concepts
\section{Method / Model}
% Core contribution with sufficient detail
\section{Experimental Setup}
% Datasets, metrics, baselines, implementation
\section{Results and Analysis}
% Quantitative results, ablation, error analysis
\section{Conclusion}
% Summary, limitations, future work
\section*{Acknowledgments}
% Funding, collaborators
% then references, then appendix
Standard Section Order (ACL)
- Abstract — mandatory, state problem/method/results
- Introduction — motivation, problem, contributions (enumerate explicitly)
- Related Work — positioning vs. prior work, categorized
- Background / Preliminaries — formalism, definitions, notation
- Method / Model — core contribution
- Experimental Setup — datasets, metrics, baselines, implementation
- Results — experiments, results, analysis
- Conclusion — summary, limitations, future work
- Acknowledgments — funding, collaborators
- References
- Appendix / Supplementary
Citations
acl.sty loads natbib. Use these citation commands:
\citet{Gusfield:97} % "Gusfield (1997)" — author in text
\citep{Gusfield:97} % "(Gusfield, 1997)" — parenthetical
\citealp{Gusfield:97} % "Gusfield, 1997" — alternative (no parens)
\citeyearpar{Gusfield:97} % "(1997)" — year only
\citeposs{Gusfield:97} % "Gusfield's (1997)" — possessive (ACL-specific)
ACL Anthology Bibliography
Obtain the full ACL Anthology BibTeX: https://aclweb.org/anthology/anthology.bib.gz.
Include in your bibliography:
\bibliographystyle{acl_natbib}
\bibliography{anthology,your_file} % Include ACL Anthology + your file
References
\bibliographystyle{acl_natbib}
\bibliography{your_file} % single .bib file
\bibliography{anthology,your_file} % include ACL Anthology + your file
Obtain the full ACL Anthology BibTeX: https://aclweb.org/anthology/anthology.bib.gz.
Include doi or url fields in every bibliography entry for hyperlinks.
BibTeX Guidelines
- Use the accented-character commands from
acl.sty in .bib files (e.g., {\"a} for ä).
- Do not use Unicode characters directly in BibTeX entries.
- Alphabetize references in the output.
- List all authors (no "et al." in bibliography).
Figures
\usepackage{graphicx} % already in preamble
\begin{figure}[t]
\includegraphics[width=\linewidth]{your-figure}
\caption{Descriptive caption. Keep it informative and concise.}
\label{fig:your-label}
\end{figure}
% Wide figure spanning both columns
\begin{figure*}[t]
\includegraphics[width=0.48\linewidth]{fig-a}\hfill
\includegraphics[width=0.48\linewidth]{fig-b}
\caption{Side-by-side caption.}
\label{fig:side-by-side}
\end{figure*}
- Use PDF, PNG, or EPS for graphics.
- Match figure fonts to document fonts.
- Do not override default caption sizes.
- For subfigures, use the
subcaption package (compatible with acl.sty).
Figure Best Practices for NLP
\begin{figure}
\centering
\begin{tabular}{cc}
\includegraphics[width=0.4\linewidth]{example1} &
\includegraphics[width=0.4\linewidth]{example2} \\
(a) Case 1 & (b) Case 2
\end{tabular}
\caption{Examples of qualitative analysis showing...}
\label{fig:qual}
\end{figure}
Tables
Use booktabs-style rules. Avoid vertical rules.
\begin{table}[t]
\centering
\begin{tabular}{lcc}
\hline
\textbf{Model} & \textbf{Precision} & \textbf{Recall} \\
\hline
BERT & 85.2 & 83.1 \\
RoBERTa & 87.4 & 85.6 \\
\hline
\end{tabular}
\caption{Results on the dev set.}
\label{tab:results}
\end{table}
Professional Table Formatting
\begin{table}[t]
\caption{Comparison of models on standard benchmarks.
Best results in \textbf{bold}. $\dagger$ indicates $p < 0.01$ vs. BERT.}
\label{tab:comparison}
\centering
\begin{tabular}{lcccc}
\toprule
\multirow{2}{*}{Model} & \multicolumn{2}{c}{Task 1} &
\multicolumn{2}{c}{Task 2} \\
\cmidrule(lr){2-3}\cmidrule(lr){4-5}
& Acc & F1 & Acc & F1 \\
\midrule
BERT & 85.2 & 84.1 & 78.3 & 77.5 \\
RoBERTa & 87.4 & 86.2 & 80.1 & 79.2 \\
\textbf{Ours} & \textbf{89.1} & \textbf{88.3} & \textbf{82.4} & \textbf{81.8} \\
\bottomrule
\end{tabular}
\end{table}
Equations
\begin{equation}
\label{eq:loss}
\mathcal{L} = -\sum_{i=1}^{N} \log p(y_i \mid x_i; \theta)
\end{equation}
Label all equations; reference with \eqref{eq:label}.
Mathematical Notation for NLP
\section{Preliminaries}
Let $\mathcal{D} = \{(\mathbf{x}_i, y_i)\}_{i=1}^N$ denote a dataset of
input-output pairs, where $\mathbf{x}_i \in \mathcal{X}$ is a sequence of
tokens and $y_i \in \mathcal{Y}$ is the corresponding label.
We denote the model's parameters by $\theta$. The conditional probability
of output $y$ given input $\mathbf{x}$ is:
\begin{equation}
p_\theta(y \mid \mathbf{x}) = \text{softmax}(\mathbf{W} \mathbf{h}_\theta(\mathbf{x}))_y
\end{equation}
where $\mathbf{h}_\theta(\mathbf{x})$ is the encoding of $\mathbf{x}$.
Appendix
Place appendix content after \appendix:
\appendix
\section{Additional Experimental Results}
\label{sec:app-results}
\section{Proofs}
\label{sec:proofs}
The \appendix command switches section numbering from numbers to letters (A, B, C...).
Page Limit
ACL 2026: 8 pages for the main paper, excluding references and appendix.
Check the official CFP for the exact page limit and whether it includes tables and figures. Excess pages may be desk-rejected.
Anonymity and Double-Blind Review
- In review mode, author names are automatically replaced with "Anonymous ACL submission". Do not add acknowledgments or self-citations that reveal identity.
- Remove or anonymize references to prior work by the authors (or use generic placeholders that you restore for the camera-ready).
- Do not upload anonymized drafts to public repositories.
- Check the CFP for specific anonymity policy updates.
Self-Citation Anonymization
% In submission (anonymized):
...as shown in prior work~\citep{anonymous:2024:ourprior}...
% In camera-ready:
...as shown in our prior work~\citep{smith2024prior}...
Camera-Ready Checklist
Writing Conventions (NLP)
Problem Definition
\section{Task Definition}
We focus on the task of $X$, formally defined as follows.
\textbf{Input:} A sequence of tokens $\mathbf{x} = (x_1, \dots, x_n)$.
\textbf{Output:} A structured object $y$ (e.g., a parse tree, label sequence).
\textbf{Evaluation:} We measure performance using standard metrics
including F1 score for token-level tasks and accuracy for sentence-level tasks.
Dataset Description
\subsection{Datasets}
We evaluate on four standard benchmarks:
\begin{itemize}
\item \textbf{Dataset A}: 10K training, 1K dev, 1K test examples.
Source: URL. Domain: newswire. Language: English.
\item \textbf{Dataset B}: 50K training, 5K dev, 5K test examples.
Source: URL. Domain: social media. Language: multilingual (10 languages).
\end{itemize}
Preprocessing: tokenized using spaCy, lowercased, rare words replaced with <UNK>.
Error Analysis
\subsection{Error Analysis}
We manually analyze 100 randomly sampled errors.
\textbf{Error categories:}
\begin{itemize}
\item \textbf{Long sentences} (34\%): Performance degrades for inputs > 50 tokens.
\item \textbf{Rare entities} (28\%): Named entities not seen during training.
\item \textbf{Ambiguity} (22\%): Genuinely ambiguous inputs where even humans disagree.
\item \textbf{Other} (16\%): Annotation errors, preprocessing issues.
\end{itemize}
Qualitative Examples
\subsection{Qualitative Examples}
\autoref{tab:examples} shows representative examples from our model.
\begin{table}[h]
\caption{Qualitative examples comparing model outputs.
``GT'' denotes ground truth, ``Err'' denotes error type.}
\label{tab:examples}
\centering
\begin{tabular}{p{3cm}p{3cm}p{2cm}}
\toprule
Input & Output & Type \\
\midrule
The dog chased the cat. & nsubj(dog, chased) & Correct \\
John gave Mary a book. & ROOT(gave) & Correct \\
\bottomrule
\end{tabular}
\end{table}
Supplementary Materials
Check the ACL 2026 CFP for the supplementary materials policy. Common rules:
- Supplementary files are not part of the review process.
- Keep supplementary material minimal and clearly labeled.
- Do not exceed file size limits specified by the submission system.
Typical Supplementary Contents
% In main paper:
Results on additional datasets are provided in Appendix A.
% In supplementary:
\section{Additional Datasets}
\section{Extended Error Analysis}
\section{Hyperparameter Sensitivity}
\section{Model Architecture Details}
Checklist Before Submission