The repository contains a template for writing NLP/ML papers (using latex). By no mean one could write great articles by filling a template, but we hope this project could help less experienced researchers to grasp key aspects of an academic paper.
In the template, we try to give a detailed todo list for each paper section
(see texts in docs/antnlp-tawp.pdf
),
a workflow for a new writing project,
and a flavor of using latex.
We suggest using github repos to manage tex files.
To start your own projects from this template, click Use this template
(see creating a repo from a template).
Suppose your project's name is "antnlp-tawp".
antnlp-tawp/docs
: latex source files for the main paper. We use\input
commands in the main tex fileantnlp-tawp.tex
to facilitate collaborative writing.antnlp-tawp/images
: images (prefer pdf format) used in the paper. You may have nested directories if some images are generated by other programs (e.g., source codes of matplotlib).antnlp-tawp/supplementary
: latex source files for the supplementary.
Instead of "fig1.pdf, fig2.pdf, paper.tex, my-paper.tex", please files with clear meaning (e.g., "pr-curve.pdf", "inorder-dep-parsing.tex").
When you start a new paper, we suggest going with the following steps
- list motivations (e.g., problem importance/challenges) in the introduction section. It helps you to build a complete picture of the paper.
- list experiments.
- draw the key figures of your model. It helps you to organize the approach section.
- the approach section.
- the introduction section and the abstract.
- the experiments section.
- the related work section.
- the conclusion.
The writing should start at least 1 month ahead of deadlines.
- latex software: texlive. To be friendly to unicodes, we recommend compiling with xelatex.
- latex source files
- add a line break for each sentence (clauses, or long phrases). Don't worry about the formatting, unless you adding an empty line, latex automatically wraps them into one paragraph in the final pdf file. Short sentences are prefered both for experssing ideas and reviewing.
- latex packages
- equations:
IEEEeqnarray
package - tables: avoid vertical lines. Usually,
\toprule
,\midrule
and\bottomrule
are enough.
- equations:
- figures
- make sure to include vector graphics (pdf(recommended), eps, svg) in your final paper, rather than jpg, png, bmp.
- to crop blank margin of a pdf, try
pdfcrop
in texlive. - figures with latex symbols and equations
- Mac: keynote + LaTeXiT
- Windows: ppt + kLatexFormula
- Linux: wps+
images/latex2emf.py
(a linux version of kLatexFormula is available. However, the current version of linux wps can not directly import emf file from clipboard. We still need to do the importing job manually.)
- or you can try matplotlib (be sure that ticklabels, annotations, legend, etc are in proper font size (the default setting is usually too small))
- even tikz (use additional .tkz file to place your tikz figure source file, rather than in the main tex)