Count words

What counts, and what does the counting

The word count is not ours. Words come from the browser's own Unicode segmenter, asked for its default locale rather than a named one, and the page keeps the pieces it marks as word-like. Nothing here adds a rule for hyphens, apostrophes or digits on top of that — what you get is the segmenter's judgement about where words begin, which is why the answer is right in Chinese and Japanese and why two browsers on two machines can differ very slightly on the same text. On a browser old enough to lack the segmenter, the page falls back to splitting on whitespace and tells you the figure is approximate, which is the number simpler counters give you all of the time without saying so.

Paragraphs and lines are counted by our own rules and they are not the same rule. A paragraph boundary is a blank line — two line breaks with nothing but spaces or tabs between them — so a Windows text file dropped in with its carriage returns intact counts as a single paragraph; text pasted into the box never has that problem, because a text box hands over line feeds. Lines are counted the way an editor's gutter counts them: an empty box is nought lines, one word is one line, and a file that ends in a newline has an empty last line which counts. That is deliberately not what the command line's wc reports, which is the number of line terminators.

Two smaller decisions settle most arguments about the numbers. Characters without spaces removes anything the browser calls whitespace, which includes the non-breaking space that arrives with text copied off a web page — but not a zero-width space, which is not whitespace and stays in the count. And the longest word and the average word length are measured in code points rather than in the grapheme clusters the headline character figure uses, so a word carrying an emoji or a combining accent measures longer here than it looks. The frequency table hides about sixty common English function words unless you ask for them, and folds case with the plain English rule regardless of what language you are writing in.

When you want something else

When a specific program is going to do the judging, only that program's number counts. A journal that says eight thousand words including references means the figure Word produces, and Word is counting fields, footnotes, captions and text boxes according to its own settings. This page will land close on ordinary prose and it will not match to the word, so submit against the tool that is marking you rather than against this one.

For anything more than one document, the counting is a shell line. wc -w takes a whole folder at once and streams files larger than your memory; for Markdown, HTML or LaTeX, piping the file through pandoc to plain text first removes the markup so that code fences, link targets and tags are not counted as prose — which is the gap between a repository word count and a real one. Neither is something a paste box can do for four hundred files.