Punctuation as a giveaway
Recently, I've seen a lot of comments on the web that the use of em dashes is a dead giveaway that an article has been written by AI. This immediately made me think of my own use of dashes and semicolons. I don't use AI for text generation, but I wondered if my writing might be mistaken for AI because of my use of punctuation. I decided to take a deeper look at the whole area.
Punctuation symbols
Let's start by looking at the symbols themselves.
| Symbol | Name | Commentary |
|---|---|---|
| — | Em dash | Not easily available from my keyboard. Named because it's the width of a capital M. HTML: — Markdown: --- |
|
–
|
En dash | Not easily available from my keyboard. Named because it's the width of a capital N. HTML: – Markdown: -- |
| - | Hyphen | Available easily on my keyboard (minus sign) |
| ; | Semicolon | Easily available from my keyboard. |
| , | Comma | Easily available from my keyboard. |
Grammatical use
I'm not going to go into grammar too much here, I'm the wrong person to do that (my favorite grammar book is "Rules for writers" by Hacker and Sommers, check it out if you want a good grammar reference). Commas and semicolons have different grammatical purposes and their use goes back a long time. Hyphens are a more modern invention and seem to have some of the same usage of both commas and semicolons; a sort of generic punctuation mark.
As far as I can tell, em dashes, en dashes, and hyphens are used for more or less the same grammatical purpose; they're interchangeable. Some websites suggest that there is a difference in grammatical use between — and –. These are reputable websites.
Usage in the real world by people
Recent writing seems to favor the use of - rather than ;, especially in short form communications like text messages or even emails. I've noticed some modern authors are using hyphens instead of semicolons, in fact, I've met a professional writer who always used hyphens and never semicolons. Overall, semicolon usage seems to be in decline.
If I'm typing in text, normally I only use characters easily available from my keyboard, unless I'm using a special character like a currency symbol (e.g. €). In other words, it's unlikely I'll use em dashes or en dashes. Given that it's hard to tell the different dashes apart, it's hard to understand why anyone (any human) other than a professional typesetter would use a dash other than a - (hyphen). In the sentence below, have I used an Em dash or an En dash or even a hyphen?
"David lived in Paris 2005–2010."
Is it a reliable AI detector?
Recent English usage seems to favor - over ;, so you can see why an AI might learn to use - rather than ;. As I said earlier, there are some websites that distinguish different uses between —, –, and -, so it's possible an AI will apply these rules too. You can sometimes tell non-native English speakers because their English is too good, they don't make the mistakes native speakers do, and something similar may be happening here. An AI may be applying a "dashes" rule that a native writer wouldn't.
Is it a smoking gun proof? Probably not. I'm sure there are writers who love different dashes, and of course, the software they're using may convert hyphens into different types of dashes for them. But it is a strong indicator.
I find distinguishing between dashes hard, but peeking at the underlying HTML or Markdown gives way the use of em-dashes and en-dashes immediately. So if you have access to the text, you can check.
By contrast, the use of a ; may indicate a human writer, until of course, AIs learn how to use it (im)properly.