Quarto word count

• Why counting words is hard
• Using the word count script
  • Using as an extension
  • Using without an extension
• How this all works
• Appendices
• Example
• Credits

Why counting words is hard

In academic writing and publishing, word counts are important, since many journals specify word limits for submitted articles. Counting how many words you have in a Quarto Markdown file is tricky, though, for a bunch of reasons:

1. Compatibility with Word: Academic publishing portals tend to care about Microsoft Word-like counts, but lots of R and Python functions for counting words in a document treat word boundaries differently.

   For instance, Word considers hyphenated words to be one word (e.g., “A super-neat kick-in-the-pants example” is 4 words in Word), while stringi::stri_count_words() counts them as multiple words (e.g. “A super-neat kick-in-the-pants example” is 8 words with {stringi}). Making matters worse, {stringi} counts “/” as a word boundary, so URLs can severely inflate your actual word count.

2. Extra text elements: Academic writing typically doesn’t count the title, abstract, table text, table and figure captions, or equations as words in the manuscript.

   In computational documents like Quarto Markdown, these often don’t appear until the document is rendered, so simply running a word-counting function on a .qmd file will count the code generating tables and figures, again inflating the word count.

3. Citations and bibliography: Academic writing typically counts references as part of the word count (even though IT SHOULDN’T). However, in Quarto Markdown (and all other flavors of pandoc-based markdown), citations don’t get counted until the bibliography is generated, which only happens when the document is rendered.

   Simply running a word-counting function on a .qmd file (or something like the super neat {wordcountaddin}) will see citekeys in the document like @Lovelace1842, but it will only count them as individual words (e.g. not “(Lovelace 1842)” in in-text styles or ‘Ada Augusta Lovelace, “Sketch of the Analytical Engine…,” Taylor’s Scientific Memoirs 3 (1842): 666–731.’ in footnote styles), and more importantly, it will not count any of the automatically generated references in the final bibliography list.

Using the word count script

This extension fixes all three of these issues by relying on a Lua filter to count the words after the document has been rendered and before it has been converted to its final output format. Frederik Aust (@crsh) uses the same Lua filter for counting words in R Markdown documents with the {rmdfiltr} package (I actually just copied and slightly expanded that package’s inst/wordcount.lua). The filter works really well and is generally comparable to Word’s word count.

The word count will appear in the terminal output when rendering the document. It shows three different values: (1) the total count, (2) the count for the document sans references, and (3) the count for the reference list alone.

133 total words
-----------------------------
76 words in text body
57 words in reference section

There are two ways to use the filter: (1) as a formal Quarto format extension and (2) as a set of pandoc filters. You should definitely glance through the “How this all works” section to understand… um… how it works.

Using as an extension

Install the extension in an already-existing project by running this in your terminal:

quarto add andrewheiss/quarto-wordcount

Or create a brand new project by running this:

quarto use template andrewheiss/quarto-wordcount

You can then specify one of three different output formats in your YAML settings: wordcount-html, wordcount-pdf, and wordcount-docx:

title: Something
format:
  wordcount-html: default

The wordcount-FORMAT format type is really just a wrapper for each base format (HTML, PDF, and Word), so all other HTML-, PDF-, and Word-specific options work like normal:

title: Something
format:
  wordcount-html:
    toc: true
    fig-align: center
    cap-location: margin

Using without an extension

Alternatively, if you don’t want to install the extension, download the two Lua scripts wordcount.lua and citeproc.lua, put them somewhere in your project, and reference them in the YAML front matter of your document. Make sure you also disable citeproc so that it doesn’t run twice.

title: Something
format:
  html:
    citeproc: false
    filters: 
      - at: pre-render
        path: "citeproc.lua"
      - at: pre-render
        path: "wordcount.lua"

Or as another option, install the extension like normal, but don’t use the wordcount-html format and instead refer to the Lua filters that live in the extension folder:

title: Something
format: 
  html:
    citeproc: false
    filters: 
      - at: pre-render
        path: "_extensions/andrewheiss/wordcount/citeproc.lua"
      - at: pre-render
        path: "_extensions/andrewheiss/wordcount/wordcount.lua"

How this all works

Behind the scenes, pandoc typically converts a Markdown document to an abstract syntax tree (AST), or an output-agnostic representation of all the document elements. In AST form, it’s easy to use the Lua language to extract or exclude specific elements of the document (i.e. exclude captions or only look at the references).

Quarto was designed to be language-agnostic, so {rmdfiltr}’s approach of using R to dynamically set the path to its Lua filters in YAML front matter does not work with Quarto files. (See this comment from the Quarto team stating that you cannot use R output in the Quarto YAML header.)

But it’s still possible to use the fancy {rmdfiltr} Lua filter with Quarto with a little trickery!

In order to include citations in the word count, we have to feed the word count filter a version of the document that has been processed with the --citeproc option enabled. However, in both R Markdown/knitr and in Quarto, the --citeproc flag is designed to be the last possible option, resulting in pandoc commands that look something like this:

pandoc whatever.md --output whatever.html --lua-filter wordcount.lua --citeproc

The order of these arguments matter, so having --lua-filter wordcount.lua come before --citeproc makes it so the words will be counted before the bibliography is generated, which isn’t great.

{rmdfiltr} gets around this ordering issue by editing the YAML front matter to (1) disable citeproc in general and (2) specify the --citeproc flag before running the filter:

output:
  html_document:
    citeproc: false
    pandoc_args:
      - '--citeproc'
      - '--lua-filter'
      - '/path/to/rmdfiltr/wordcount.lua'

That generates a pandoc command like this, with --citeproc first, so the generated references get counted:

pandoc whatever.md --output whatever.html --citeproc --lua-filter wordcount.lua

Quarto doesn’t have a pandoc_args option though. Instead, it has a filters YAML key that lets you specify a list of Lua filters to apply to the document at specific steps in the rendering process:

format:
  html:
    citeproc: false
    filters: 
      - at: pre-render
        path: "/path/to/wordcount.lua"

However, there’s no obvious way to reposition the --citeproc argument and it will automatically appear at the end, making it so generated references aren’t counted.

Fortunately, this GitHub comment shows that it’s possible to make a Lua filter that basically behaves like --citeproc by feeding the whole document to pandoc.utils.citeproc(). That means we can create a little Lua script like citeproc.lua:

-- Lua filter that behaves like `--citeproc`
function Pandoc (doc)
  return pandoc.utils.citeproc(doc)
end

…and then include that as a filter:

format:
  html:
    citeproc: false
    filters: 
      - at: pre-render
        path: "/path/to/citeproc.lua"
      - at: pre-render
        path: "/path/to/wordcount.lua"

This creates a pandoc command that looks something like this, feeding the document to the citeproc “filter” first, then feeding that to the word count script:

pandoc whatever.md --output whatever.html  --lua-filter citeproc.lua --lua-filter wordcount.lua

Eventually the Quarto team is planning on allowing filter options to get injected at different stages in the rendering process, so someday we can skip the citeproc wrapper filter and just do something like this:

format:
  html:
    filters:
      post:
        - '/path/to/wordcount.lua'

But that doesn’t work yet.

Appendices

In academic writing, it’s often helpful to have a separate word count for content in the appendices, since things there don’t typically count against journal word limits. Quarto has a neat feature for automatically creating an appendix section and moving content there automatically as needed. It does this (I think) with a fancy Lua filter.

However, Quarto’s appendix-generating process comes after any custom Lua filters, so even though the final rendered document creates a div with the id “appendix”, that div isn’t accessible when counting words (since it doesn’t exist yet), so there’s no easy way to extract the appendix words from the rest of the text.

So, as a temporary workaround until I can figure out how to make this Lua filter run after the creation of the appendix div, you can get a separate word count for the appendix by creating your own div with the id appendix-count:

# Introduction

Regular text goes here.

::: {#appendix-count}

# Appendix {.appendix}

More words here

:::

That will create this word count:

5 in the main text + references, with 4 in the appendix
-------------------------------------------------------
5 words in text body
0 words in reference section
4 words in appendix section

Example

You can see a minimal sample document at template.qmd.

Credits

The wordcount.lua filter comes from Frederik Aust’s (@crsh) {rmdfiltr} package.