The Future

Why Google docs is gaslighting everyone about spelling: an investigation

Turns out, it’s the internet’s fault.

The Future

Why Google docs is gaslighting everyone about spelling: an investigation

Turns out, it’s the internet’s fault.

Unless you’re a nerd, or the victim of some embarrassing autocorrect, you hopefully don’t think about spell checkers all too often. They’re one of those magical internet tools that are designed to work seamlessly in the shadows. So long as they don’t screw up, they shouldn’t occupy your mind much at all.

But of course, like all things, spell checkers screw up pretty regularly, and thanks to their general ubiquity nowadays, these errors are usually quickly and widely noticed. When it comes to issues like the iPhone’s strange late-2017 aversion to the letter ‘i’, it’s pretty easy to point the blame at a freak glitch or bug, but not all problems are as simple. Take, for example, the curious case of Google Docs:

Oddly enough, the issue isn’t that Google Docs’ spell checker is wrong, but that it’s too right. Allow me to explain: Spell checkers rely on a number of tests and programs in order to properly own you for using “their” instead of “there,” but the two big ones are called isolated-word error correction and context-dependent word correction. As boring and stodgy as they may sound, the concepts behind them are actually pretty simple. Isolated-word error correction essentially means that the spell checker is just trying to figure out whether or not the combination of letters you typed is a real word, and if not, it’ll try and figure out what you meant to type. It generally does this by comparing your spelling error against a whole corpus of other errors to figure out what real word the average person means to type when they make that particular mistake. This is pretty accurate on a word-to-word level, but not so accurate on the sentence level, as it can lead to grammatical monstrosities like “Eye halve a spelling chequer” to technically be labeled as correct.

That’s where context-dependent word correction comes in. At the most basic level, this works by looking at the words in a sentence or paragraph together and assessing the likelihood of these pairings (e.g. “eye halve” is way less likely than “i have” so it’s probably wrong). In order to do these types of predictions, the spell checker has to have a huge body of correct sentences and paragraphs to check your writing against.

For a company like Google, that’s no problem, as its bots have basically crawled and cataloged the entire web. So, in 2012, Google Docs introduced a content-dependent word correction system that would provide suggestions that were informed by the internet at large. As a whole, this new system has worked splendidly, as Google’s massive trove of sample data is able to grow and adapt with the web, instead of merely being sporadically updated like a traditional dictionary. However, this change also likely accounts for the bizarre errors experienced by users like those mentioned above, as the one downside to using the internet as your dictionary is that, well, you’re using the internet as your dictionary. Though you may end up with “covfefe” marked as correct within mere hours of its creation, you also run the risk of marking less common spellings and phrasings as incorrect based on their lower overall use.

That being said, it’s worth noting that Google is rather secretive when it comes to, well, everything. So this answer hasn't technically been confirmed by the company itself. We’ve reached out to Google for comment and will update this story if we hear back.