There are many good resources on the structure and philosophy behind academic writing. I mention some of these in the first section below. But this note is different: it is simply a (growing) list of things that I keep on telling students when reviewing paper or thesis drafts. These are mainly about nitty-gritty things (instead of the bigger story, which is arguably more important). But it is still important to get these things right.

Two disclaimers. First, parts of this note convey my personal writing preferences (and it might sound quite opinionated, because it is). I try to state when something is a stylistic choice rather than a rule which you should really follow. Second, I wrote parts of the note in anger, normally after reviewing many paper and thesis drafts in a short space of time. So sorry if I sometimes sound a bit blunt!

## Writing structure and philosophy

There are many good resources on the structure and bigger picture behind academic writing. Here are some of my favourites on specific topics:

## We don’t clap hands for basics

We all make mistakes and try to learn from them. But some mistakes make me really sad. Here are some basics that I expect when a student gives me a draft:

• Spelling mistakes. With today’s word processors, editors and tools, there is no excuse for spelling mistakes.
• If you have a figure or table, you need to refer to it and describe it somewhere in the text.
• If I gave feedback on a previous draft, I expect that you work the recommendations into the entire document. I often fix a mistake only where it occurs the first time. I then assume you will fix all occurrences of this mistake throughout the entire draft.

• With whatever you do (spelling, figures, table layout, capitalisation, section headings), be consistent!
• Choose either British or American spelling, and stick to it. This means that you won’t have “initialize” in Section 2 and “initialise” in Section 4.
• Avoid subjective and emotive language. For instance, when describing results, don’t say that the results are “disappointing” or “remarkable” or “impressive”. Simply state the facts: “model X performs 10% worse than model Y”. Sometimes you can say that something is unexpected or surprising, but then you need to unpack why you say so, and help the reader see what you are seeing.

## Mathematics

• Consistent mathematical notation. If $x$ is used to represent something in Section 1, then $x$ shouldn’t be used for something else in Section 3. Sometimes it is difficult to be entirely consistent (this is often the case in bigger documents like a thesis). But then you need to tell the reader clearly that you are being sloppy with your notation (and why).
• We sometimes use “proper” words or abbreviations to denote an operation within a mathematical expression. For instance, we use “max” to denote maximisation or “log” to indicate taking the logarithm. In these cases, the operation should be denoted with an upright font, i.e. $\textrm{max}(x)$, not $max(x)$. The latter implies a multiplication of variables: $m \cdot a \cdot x \cdot (x)$. You should also use an upright font if you use a word as a sub- or superscript to denote some properties of a variable: for instance, you should write $y_\textrm{pair}$ and not $y_{pair}$.
• There are different ways to refer to an equation in the text. Choose how you will refer to equations and do so consistently. For instance, either write “the loss is shown in (3.3)” or “the loss is shown in Equation 3.3”. Do not mix the two!

## Figures and tables

• Figure and table captions are always full sentences (ending with a full stop). The caption should be self-contained, i.e. it should be possible for the reader to get a fairly good understanding of the figure or table by just reading the caption. This means that “Model diagram” or “Illustration of the AwsmeCat approach” are not good captions.
• If you did not make a figure yourself, it requires a citation in the figure caption otherwise it is plagiarism. You can also ask the author directly whether it is okay to use their figure; then you should not only cite them, but also thank them in acknowledgements. The best approach is probably to just remake the figure, so that it is consistent with the other figures in your document (but even then, it is good to add a citation to the figure caption).
• Try to place figures or tables close to where they are first mentioned.
• Try to put tables and figures at the top or bottom of a page. This is not a hard rule (although it might be for some journals or conferences). But a hard rule I do use is that I do not split paragraphs with a figure. This means that, if the figure is not at the top or bottom of the page, it will be in-between the end of a section and the heading of the next section.
• If you use LaTeX, use proper names when labelling your figures or tables. For instance, don’t use \label{fig7}, but use something like \label{fig:tsne_acoustic_embeddings}. This is not so important for a short paper or report, but for a long document, proper labels can be very helpful (especially when using a LaTeX editor with autocomplete that gives you a list of defined labels after typing \ref).

A bit more on table layout. If you use LaTeX, make sure to learn how to use the tabularx and booktabs packages. The tabularx package allows you to stretch your table to (a fraction of) the text or column width. The booktabs packages helps with layout and making tables pretty, but more importantly, the first two sections of the booktabs manual give an opinionated rant about how to format tables. I recommend that you read those two sections and follow the guidelines. I specifically subscribe to the two rules given there:

1. Never use vertical rules.
2. Never use double rules.

This means that tables would typically look something like this:

This is somewhat of a personal choice, but I do think that removing vertical rules makes tables look a lot better and improves readability.

## Bibliography

Make sure that the references in your bibliography are consistent:

• I generally make sure that the referencing format for all conferences papers are exactly the same and that the format for all journal articles are the same.
• Decide whether you are going to abbreviate conferences or not, and then do so consistently. E.g., in the same bibliography you should not have “in Proceedings of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)” and “in Proc. ICASSP”. Choose one format and stick to it. I prefer to use the shortened abbreviation format everywhere (since this saves space and is easier to keep consistent), but the important thing is to be consistent.
• If you include the page numbers and the location for a conference, you need to do this for all the conference papers you cite. I prefer not to include these details for conference papers (again, since this then makes it easier to maintain a consistent bibliography).
• Similarly, if you add page numbers, the volume and the number for a journal article, you should do so for all the journals. I tend to include this information for journals.
• If there is a conference or a journal version of an arXiv paper, cite the published version.
• Decide whether you are going to add the DOI and online link to papers, and then do so consistently. Here I would (almost definitely) advise to leave this information out, since it can be hard to obtain for all publications.
• Do not trust BibTeX entries drawn automatically from Google Scholar or applications like Mendeley. You need to go through these to make sure they are correct and that they are consistent with your own formatting.

As an example, consider the following bibliography:

1. L. Besacier, E. Barnard, A. Karpov, and T. Schultz, “Automatic speech recognition for under-resourced languages: A survey,” Speech Commun., vol. 56, no. 1, pp. 85–100, 2014.
2. Y. Zhang and J. R. Glass, “Unsupervised spoken keyword spotting via segmental DTW on Gaussian posteriorgrams,” in Proc. ASRU, 2009.
3. K. L. Levin, A. Jansen, and B. Van Durme, “Segmental acoustic indexing for zero resource keyword searh,” 2015 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2015.
4. A. Jansen, E. Dupoux, S. J. Goldwater, “A summary of the 2012 jhu clsp workshop on zero resource speech technologies and models of early language acquisition,” in IEEE International Conference on Acoustics, Speech and Signal Processing, May 2013, pp. 8111–8115.
5. M. Heck, S. Sakti, and S. Nakamura, “Learning supervised feature transformations on zero resources for improved acoustic unit discovery,” IEICE T. Inf. Syst., vol. 101, no. 1, pp. 205–214, 2018.

Both the journals [1] and [5] are abbreviated and contain the volume, number and page numbers, so these are consistent. The conference papers [2–4] are not consistent: [2] is abbreviated, while [3] and [4] are not. The year occurs twice in [3], which could be okay but then this should also be done in [2] and [4]. In reference [4], page numbers are given as well as the month of the conference; again this could be okay, but then this should be added for [2] and [3] as well. There is an additional issue in [4]: “jhu clsp” are abbreviations which should be capitalised, i.e. “JHU CLSP”.

## LaTeX-specific recommendations

• I recommend using the microtype package. It basically moves and spaces letters so that a document looks more beautiful. More importantly, it often reduces the length of a paper since it reduces the number of dangling words (when the last word of a paragraph ends up on a line of its own).
• Put each sentence on a new line in your LaTeX code. Alternatively, partial sentences can be put on new lines. But don’t have a whole paragraph on a single line. There are several reasons for this:
• If you use version control (git) and a whole paragraph is on a single line and you change a single word, then that whole paragraph will be flagged as a change. This can be especially irritating when collaborating with others on a LaTeX document through git.
• Similarly, if you do a diff on a document later to see changes, then the whole paragraph will be marked as a change.
• Some LaTeX editors (including Overleaf) allows you to do backward search, where you can click on a PDF preview of the document and then jump to the corresponding line of code. (Forward search is the opposite, where you jump from a line in the code to the corresponding point in the PDF.) If you place each paragraph on a single line in your code and you perform backward search, then you would jump to the paragraph and will need to find the particular sentence or word that you are looking for. If sentences or partial sentences are placed on new lines in your LaTeX source, then it is much easier to find your place when doing backward search.
• Splitting your LaTeX source into multiple files:
• Break up your document into separate .tex files. Then in the main file, use \input to pull in the separate .tex files. This means that I would typically have a introduction.tex, related_work.tex, models.tex, etc. file for a single paper.
• This again helps when collaborating on a single document with multiple authors through git, making it easier to isolate where changes have happened.
• It also really helps with larger documents like theses.
• In Overleaf I have seen that it might be better to use a single file, since separate files sometimes break the forward and backward search functionality. (This is maybe one more reason why I am not a huge Overleaf fan, but I think it’s just me.)
• More on version control with LaTeX:
• Make sure that .pdf, .aux, .bbl and other file types produced as part of the document build process are not included or tracked in your repository (basically anything that is not source or figures). I use .gitignore files to do this.
• Make sure your editor doesn’t overwrite changes when pulling in changes for an open file. I once had a horrible experience with a collaborator using a LaTeX editor which really broke our document. We were collaborating from a common git repository. The collaborator was pulling from git using the command line. But his editor wasn’t reloading the files after a pull. This meant that he would pull (without getting the updates), continue to edit, and then my changes were lost the moment he saved. I guess this is an edge case with a particular way of collaborating and editing, but it caused a lot of sadness.
• List of figures / list of tables: Use shortened versions of the full figure/table caption, no more than one or two lines. A good choice for a short version would typically be the first sentence of the caption. You can do this do this in LaTeX using the command \caption[short version]{long version}.

## Conventions

These are some personal conventions I follow (and might also recommend/force onto students).

• Abbreviations. In a paper, I write out abbreviations—such as hidden Markov model (HMM)—in the abstract, at first use in the body, and also in the conclusion. I do this since I think these sections should be self-contained. I also (normally) write out abbreviations in section headings.
• Tenses. It is often tricky to know what tense (past, present, future) to use for different sections. My own convention is always to write in the present tense, except for the Conclusion, which I write in the past tense. Many students tend to write the Abstract in the future tense, e.g. “we will propose a model”, but I prefer to also write the Abstract in the present tense, e.g. “we propose a model”. There are some cases where I would make an exception in the body and write in the past tense instead of the present tense; this is normally when I refer back to something that has happened in a previous section or chapter, e.g. “In Section 3.1 we found that …”.
• Bulleted lists. I try to follow this guideline on how to punctuate bullet point lists.
• “3” vs. “three”: I write out the words for numbers up to twelve, and from then on write out the number. E.g. I would write “13 layers”, but would write “five captions”.
• In a thesis, book, or longer report, I often start a chapter with a paragraph explaining what the reader should expect in that chapter. I also often finish with a final section like “Chapter summary”, explaining what the reader should take away from that chapter and how it links with what happens next and in the rest of the document.

## General writing strategies

• Struggling to write. Like everyone, I sometimes get writer’s block. One thing I sometimes do which helps, is to write using pen and paper, away from my computer. There is something about this analogue process that helps with the struggle of starting to write. I realise this might not work for everyone, but I’ve written large parts of some papers in my notebook (sometimes on a long flight—I don’t like taking out my laptop).
• After explaining something once in a paper, I sometimes feel like explaining it in a slightly different way in a follow-up sentence or paragraph. This is normally a great indication that I didn’t explain the concept well in the first place: I need to go back and rewrite it.

## Stellenbosch University templates

A LaTeX template for reports following the guidelines of the E&E department at Stellenbosch University can be downloaded here.

By (un)commenting the appropriate lines in the root LaTeX file, the front page can follow the format for:

• Skripsie (i.e. final year project) reports
• MEng theses

## Acknowledgements

I changed the way I format tables after I received an email from Iain Murray; I had given a talk with some nasty looking tables on some of the slides, and he sent a kind email with some great recommendations.

Leanne Nortje edited and made suggestions for parts of the LaTeX template above. I also received good recommendations from Ludwig Schwardt.