high angle photo of robot

Can a Grammar Checker Replace an Editor?

Oh, c’mon folks. If you’re reading this blog, you shouldn’t even have to ask if a grammar checker can replace an editor.

A high-angle image of Pepper, a white, humanoid robot with large black eyes and a small mouth.
Not today, science. (Alex Knight / Pexels.com)

Of course not.

And I’m not just saying that because I am an editor and I’m scared of the robots coming for my job. Trust me. We haven’t taught the robots how to do that job well enough to leave it up to them. Maybe someday in the future we can, but for now, an editor is always going to beat out an AI.

What Can an Editor Do That an AI Can’t?

The biggest problem with AI right now is context. Seriously, check out some of the Twitter threads about training neural networks. You have to feed AI massive amounts of data for it to be able to accurately pick things up.

My favourite is one about training an AI to recognize pictures of sheep in unexpected places. The neural network had been fed many, many photos of sheep in expected places. It was pretty good at picking them up when they were in meadows or barns—you know, places you normally expect sheep.

Two sheep have laid down in the middle of a country roadway in the mountains. A third sheep crosses from the right of the frame.
This picture of sheep would probably confuse it. (Amelie Lachapelle / Pexels.com)

But when a sheep suddenly showed up on a beach or on top of a mountain, the AI couldn’t parse what it was “looking” at. So it spit out all kinds of nonsensical descriptions of the image, none of them including sheep.

What does that have to do with language? Well, language is highly malleable. We have twenty-six letters, just under 175,000 words in the OED, and probably more.

And we can put all these words into novel arrangements to create new meanings—sentences people have never uttered before. And those arrangements make sense, even though we’ve never heard them before.

So, what this all adds up to is that language is highly context dependent. If we’re talking about our friend Martha, and I say “she,” you know that means Martha. Metaphor works because of the reader’s ability to put it into context, to make meaning of it.

You can probably guess where this is going. Since there are so many novel arrangements of words, context is instrumental in figuring out what anyone means. AI doesn’t work well with the novel. Like I said, it needs massive amounts of “same” data to learn what to do. You can teach it to deal with the novel, but it takes time. Human language is just too flexible at this point.

Since editors are human, they’re much better at picking up contextual clues.

Okay, But Grammar Checkers Can Still Check My Grammar!

You’d think, right? Grammar checkers have come a long way, but a lot of grammar is also contextual. Take, for example, “it’s” and “its.”

Which one is correct depends on its use in the sentence—the context. Both words are technically correct, so a spellcheck alone will rarely pick up incorrect usage. Same with something like “Mondays” versus “Monday’s”—one is plural, the other is possessive, and the grammar checker isn’t going to flag your usage.

Grammar checkers are much better today at flagging erroneous usage, but that doesn’t mean they get it right 100 per cent of the time. Subject-verb agreement? Sometimes Word gets it right. And then sometimes it makes completely nonsensical suggestions, like “the pig are.”

What? How does that even seem remotely correct?

Further, grammar checkers are only as good as their programming. I’ve ranted about their incorrect flagging of passive voice. The AI is only flagging what it’s been programmed to flag. People with a faulty understanding of what the passive voice is and how it works programmed that in there.

So the grammar checker will tell you that you have a ton of passive voice, even if you don’t.

Beyond that, there are some nuances that AI is just not advanced enough to check at the moment. Do you have dangling modifiers or disembodied body parts moving around? The AI isn’t going to catch this, even though these errors can make your writing harder to understand (or more hilarious to read).

AI Doesn’t Impose Consistency

The other issue here is that the grammar checker and spellchecker aren’t going to impose consistency on what amount to matters of style. If you live in Canada, both “judgement” and “judgment” are acceptable spellings. If you flip-flop between them in your manuscript, neither grammar check nor spellcheck will save you.

And AI certainly isn’t reading for correct comma usage or consistency in how you handle your em dashes or ellipses.

This comes back to context quite often. AI, at the moment, can’t read the same way a human being can. It can only parse language against an internal database of programming. It understands syntax and, to some degree, grammatical context. But it’s not going to understand if you’re talking about one female character, then start talking about another as “she.” It will not see a problem—but your human reader will.

AI Has Biases (Thanks Programming)

Editors also have biases. I, for one, am a proponent of the Oxford comma. I prefer Latinx to “Latin@” and Indigenous Peoples to … basically anything else. “He or she” and “s/he” can take a bloody hike.

These biases extend to grammar rules as well. I’ll split infinitives and end sentences in prepositions, because I think those rules are silly. I also understand passive voice differently, and I’d really love it if grammar checkers like Hemingway stopped going after adverbs. Yes, yes, you can usually use a “stronger” verb, but people also … use a lot of adverbs. So if you’re trying to evoke a conversational tone of voice, ditching all your adverbs is … odd.

So, editors have biases. But so do AI grammar checkers. Their biases come from the people who programmed them. That’s how the AI determines something is “right” or “wrong” (or at least needs attention).

Unfortunately, that means AI is rigid when it comes to applying anything. If the AI’s programming says “passive voice is bad,” then the AI is going to flag whatever it thinks is passive voice. And it will flag it every time the writing trips the parameters of the programming.

That means you get a lot of “false flags.”

The same is true if your AI has programming that says adverbs are bad or that you shouldn’t split infinitives.

You Can Still Use a Grammar Checker

Does this mean AI-powered grammar checkers are totally useless? Not by a long shot!

What I’m saying here is, as good as grammar checkers might be today, they still have their limitations. That means you can’t rely on them to pick up everything. There are some things AI just cannot handled at this point in time. They’re just not sophisticated enough at this stage to handle the nuance of human language.

Will they eventually get there? Maybe. There’s allegedly some AIs out there that can write fake news so well, it’s hard to tell it apart from human-generated content. (I don’t necessarily think that means it’s good writing, of course.)

Human language and language processing is one of the harder things to get right because it involves so much flexibility. I mean, there are a lot of aspiring writers who have trouble crafting sentences that flow well. Wooden, stilted dialogue is incredibly common; it takes a real ear to listen and hear how people actually talk in order to get to a point where dialogue sounds naturalistic.

So, if even humans have trouble writing it, how is an AI supposed to be able to handle that level of nuance?

That doesn’t mean grammar checkers don’t have a place in your toolkit. They can be an excellent addition to your team. Whether you’re self-editing, getting help from a friend, or hiring a professional, an AI grammar checker is often a great idea.

Leave a comment