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:
-
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. -
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
.qmdfile will count the code generating tables and figures, again inflating the word count. -
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
.qmdfile (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.
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.
You should definitely glance through the “How this all works”
section to understand… um… how it works.
quarto add andrewheiss/quarto-wordcountUsing {quarto-wordcount} requires Quarto version >= 1.3.0
This will install the extension under the _extensions subdirectory. If
you’re using version control, you will want to check in this directory.
You can specify one of three different output formats in your YAML
settings: wordcount-html, wordcount-pdf, and wordcount-docx:
title: Something
format:
wordcount-html: defaultThe 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: marginThe word count will appear in the terminal output when rendering the document. It shows multiple 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 also multiple shortcodes you can use to include different word counts directly in the document:
-
Use
{{< words-total >}}to include a count of all words -
Use
{{< words-body >}}to include a count of the words in the text body only, omitting the references, notes, and appendix -
Use
{{< words-ref >}}to include a count of the words in the reference section -
Use
{{< words-append >}}to include a count of the words in the appendix, which must be wrapped in a div with the#appendix-countid (see below for more details) -
Use
{{< words-note >}}to include a count of the words in the notes: -
Use
{{< words-sum ARG >}}whereARGis some concatenation of the four countable areas:body,ref,append, andnote.For example,
{{< words-sum body-note >}}includes a count of the words in the body and notes;{{< words-sum ref-append >}}includes a count of the words in the references and appendix
You can use shortcodes in your YAML metadata too:
title: Something
subtitle: "{{< words-total >}} words"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
You can see a minimal sample document at template.qmd.
The original wordcount.lua
filter came from Frederik Aust’s (@crsh)
{rmdfiltr} package.
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 --citeprocThe 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.luaQuarto 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:
- "/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:
- "/path/to/citeproc.lua"
- "/path/to/wordcount.lua"
- quartoThis 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.luaEventually 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.