John McCrae

Your ‘References’ Section Sucks!

… but don’t worry, so do most papers.

This is a guide about how to use BibTeX to manage your references, focusing on ACL-format and computer science publications. It is based on my experience as a reviewer and seeing many, many publications with horrible reference sections. The references section of your work is not a trivial part of your paper, it is a critical part of your work, that should be prepared with the same care as the rest of your paper.

Sin 1: Protect your titles

The most common issue I see is that titles of papers are not properly are not in the correct case. BibTeX will automatically adapt the case of titles so that they fit the style of the publication venue. For example, ACL wants titles to be in sentence case, while other venues want them to be in title case. You need to explicitly tell BibTeX to protect the case of certain words in titles, as follows

@inproceedings{smith2020awesome,
  title = {An Awesome Approach to {NLP} Tasks},
  author = {Smith, John and Doe, Jane},
  booktitle = {Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics},
  year = {2020},
}

If you don’t do this, BibTeX will convert “NLP” to “Nlp” or “nlp”, which is wrong. You should protect any acronym, proper noun, or any word that should not be changed.

Sin 2: Poor source of .bib files

Many people use Google Scholar to find publications and it has a feature to get BibTeX entries. DO NOT USE IT! The BibTeX entries from Google Scholar are often incomplete or incorrect. Instead, use the following sources:

Sin 3: Incomplete entries

Another common issue is that entries in the .bib file are incomplete. At a minimum, you should provide the following fields for each entry type:

Make sure to fill in all relevant fields. Missing information can make it difficult for readers to locate the referenced work. This can normally be fixed by using good sources as above.

Sin 4: Overuse of arXiv

A reference to arXiv should only be used for papers that have not passed peer review. A paper that has not passed peer review should probably not be cited in the first place, so find the peer-reviewed version of the paper and cite this.

Sin 5: Citation as a noun

In most citation styles it is not correct to use the citation as a noun, e.g., “In (Smith et al., 2020), they show that…”. Instead, you should use the correct LaTeX command to cite the paper as part of the sentence, for example:

\citet{smith2020awesome} show that...

Here is a table of common citation commands and their outputs:

Command Output
\citep (Smith et al., 2020)
\citet Smith et al. (2020)
\citealp Smith et al., 2020
\citeyearpar (2020)
\citeposs (ACL only) Smith et al.’s (2020)

Some citations styles use different commands such as \shortcite or \newcite, so check the documentation of the style you are using.

For IEEE style citations, it is allowed to use the citation as a noun, e.g., “In [1], they show that…”. I don’t like this but it is allowed.

Sin 6: Not Grouping References

If you wish to reference multiple papers at once, do not write each reference separately like this: (Smith et al., 2020), (Doe et al., 2019), (Johnson, 2018). Instead, group them together in a single citation command like this:

\citep{smith2020awesome, doe2019great, johnson2018fantastic}

This will produce a single citation with all the references grouped together, like this: (Smith et al., 2020; Doe et al., 2019; Johnson, 2018).