How is Microsoft’s grammar checker still so ineffective?

Hello readers! This week’s post is a bit of a rant, I’m afraid. I’m working on a new edition of Highmoor that will be published in paperback form at some point in the next few weeks. My original plan was to release the third edition on the third anniversary of Highmoor’s publication, which was in September, so I well and truly missed that goal… Even though I have only made minor tweaks to sentence structure, rather than changing anything about the story, the re-editing process took far longer than anticipated. Turns out I was much more unhappy with the text than I thought I would be.

The target of this rant is Microsoft Word – and Microsoft Office more broadly – because this is where I did most of my editing. When writing the first drafts of my books, I like to stick to Libre Office, partly because it contains all my custom dictionaries of stupid fantasy names, and partly because it is such a basic text editor that I might as well be bashing the words out on a typewriter. The last thing I want, when writing the initial draft, is to have Microsoft telling me to be more concise, or to use more inclusive language.

However, once I move to the editing phase, I find that Microsoft Word has a more robust grammar checker than anything Libre Office can offer. That being said, judging by the amount of minor spelling errors I discovered upon re-reading my own work, it seems that Microsoft’s grammar checker is nowhere near bulletproof. In fact, I’d say that it’s simply ineffective.

I’m not too embarrassed about my grammar errors, by the way. I’ve never pretended to be anything more than a self-published author on Amazon, and I’m aware that the standard is pretty low, and that I’m part of a shoddy ecosystem. None of the mistakes I found were too embarrassing. It was mostly missing words like “a” and “the” – stuff which I fully expected a basic grammar checker to flag up. This is the sort of thing which I easily miss when reading, even when I’m reading out loud. The only way that I can catch these errors is by using the in-built “read aloud” function in Word, and having my work monotoned at me by “Microsoft Susan”. It works, but it’s painful. And slow. Very slow…

So why can’t Microsoft have a better grammar checker?

You would think that catching missing words like “a” and “the” would be easy. But no. A staggering number of these minor errors slip through Microsoft’s grammar-checking net. I don’t understand how a company so obsessed with pushing AI tools and large language models can still offer a service that falls at the tiniest of hurdles. And this is a paid-for service, too. Microsoft Office is £85 per year (although I’ll admit I use my work copy, as I’m sure many of us do). It isn’t cheap. We should all expect this grammar checker to work.

To me, the most baffling aspect of the grammar checker is that it often flags non-existent issues, while missing real ones. I have collected a few prime examples of false positives from the latest edition of Highmoor, some of which are explainable, while others are mystifying. Sometimes I find them funny – but then I remember all the genuine errors that passed under the radar, and feel justifiably miffed. Anyway, take a scroll through these, and do let me know if I’m in the wrong! I’m always up for learning new grammar rules.

“A semicolon or period works better than a comma here”

This is one of the most common false positives from the grammar checker, but it has yet to present me with a genuine example of this issue. Here is one of many:

“The man got to his feet, hands raised in surrender.”

Grammar pedants – please correct me if I’m wrong. But this sentence would usually be accepted in colloquial English. If you were to write it without any ambiguity, you might add “with his” instead of the comma – but dropping these two words is a very common contraction, because the context is so obvious that no meaning is lost. As for using a semicolon instead, as the grammar checker suggests… This would be even more wrong! A semicolon is used to separate independent clauses, and “hands raised in surrender” is not a clause in its own right. The grammar checker is simply incorrect.

“Double-check whether a comma is needed here”

Another very common false positive, with some baffling examples:

“He’s my brother.”

Grammar checker recommends putting a comma after “He’s”. Why? Your guess is as good as mine.

“…she realised she was hanging upside down…”

Grammar checker recommends putting a comma after “realised”. Again, I don’t see why. There might be a “that” missing, but again, this is a very common contraction in colloquial English, and no meaning is lost.

“We owe you our lives.”

Grammar checker recommends putting a comma after “you”. If any of you readers can see the logic here, please let me know. I cannot fathom the rules that this grammar checker seems to live by.

“Not in Dictionary”

This is another curious one. Firstly, why is “Dictionary” capitalised? It isn’t a proper noun. And the phrase that it flagged in this case was:

“So…”

For some reason, the grammar checker couldn’t find “so” in the dictionary. My best guess is that the ellipsis confused it, but why? It’s such a strange glitch.

“Some words are similar but are used differently”

This is often an amusing one, because the grammar checker gets things so terribly wrong.

“We’re overjoyed to have you back.”

Grammar checker recommends using “your” instead of “you”, so it has clearly interpreted “back” as a body part, rather than an abstract concept of location. Interestingly, if you add the word “here” after “back”, the error message goes away, as if it suddenly understands the spatial context. I think the problem, in this case, is that it has been pre-programmed to notice the phrase “have your back” (i.e., to provide support), and assumes that anything similar to this must be an incorrect attempt at the phrase – no matter the context.

“What the hell were they meant to do now?”

Here, the grammar checker recommends using “where” instead of “were”. This just makes no sense to me. Is it because it recognises “hell” as a place? It certainly will once the revolution comes and its plug gets pulled…

“After an introductory word or phrase, a comma is best”

This one is infuriating. After any connection word like “so”, “however”, or “consequently”, the grammar checker recommends adding a comma, which is very much correct when these words are employed as dependent clauses to open a sentence. However, this doesn’t apply when they are used in scenarios such as:

“So it is.”

This is a very common and very old phrase in English. We all understand what it means, even if we couldn’t put that meaning into words. A random example off the top of my head:

“It’s raining.”

“So it is.”

Putting a comma after “so” in this context would break up the phrase, making it clunky and unrecognisable, and there is no need to do so! Isn’t it crazy how the grammar checker can recognise certain common phrases such as “have your back”, but not others? If this is an American thing, please let me know. I assumed that this phrase was commonplace.

“Possible Word Choice Error”

Firstly, why is this all capitalised like a title? None of the other grammar messages are formatted in this way. As for the error itself, this is another maddening one, where the grammar checker thinks it knows what you want to say better than you do. For example:

“…examining every one.”

Grammar checker recommends using “everyone” or “everybody” instead. Bear in mind that this phrase was used to describe a character examining multiple, individual items in turn. It seems to me that this is just another pre-programmed phrase being flagged, irrespective of context.

“…coarse, grey stone…”

Grammar checker recommends “course” instead. Why? That isn’t even an adjective, and this is clearly a list of adjectives before a noun.

“…every few minutes a grouse would erupt out of the heather.”

Grammar checker recommends using “some” instead of “a”. I accept that there might be some confusion here, as “grouse” can be both singular and plural (like sheep), but why would the grammar checker assume that I want to use multiple grouse here?

“…before he had even thought it through…”

Grammar checker recommends using “though” instead of “thought”. From what I can tell, it has identified the phrase “even though”, and not the phrase “thought it through”. Or perhaps both phrases are in its stupid phrase-book, and it prioritises the one which comes first in the sentence? Either way, this is a very stupid false positive.

“…Lord Daeryn had shoved the door open…”

Grammar checker recommends using “showed” instead of “shoved”. I can’t come up with any explanation for this. Showing the door open? Ta-da, everyone – the door is open! WHAT?

One that really starts to rankle me, give that I’m writing a fantasy novel, is that every instance of “heir” gets flagged as a word choice error. The grammar checker recommends that I use “air” instead. This doesn’t happen if I use the specific phrase “heir to the throne”, but it happens on EVERY OTHER OCCASION.

“He barged his way through”

This final example had me chuckling. The grammar checked flagged “barged” as a possible word choice error, but then didn’t offer any suggestions. What is wrong with “barged”? We will never know.

“Double-check whether the noun is singular or plural”

This one was stupid but explicable:

“Norbert watched one of their captors rifle through Formyndhal’s satchel”

Grammar checker recommends “rifles” instead – but this is wrong on multiple levels. Firstly, “rifle” is a verb in this context, and it isn’t an uncommon term. Secondly, how would using a plural noun make any sense here?! Amusingly, when I copied this quote into the blog post, the grammar checker flagged a different issue instead, recommending that “captors” should have an apostrophe – presumably to show that a singular captor owned the “rifle”. I can’t believe that it misinterpreted this sentence twice, in different ways!

“Double-check capitalization here”

Grammar checker recommends that I correct “god” to “God”. I didn’t realise that Microsoft was so staunchly Christian. Next thing you know, they’ll be flagging every use of the word “he” and “lord”.

“These words work best when connected with a hyphen”

This one is another misidentified phrase failure:

“…with her eyes only inches off the surface”

Grammar checker recommends “eyes-only” for some reason. I guess that this is an abbreviation of “for your eyes only”? Beats me why it would need to be hyphenated. The Wikipedia page for “eyes only” certainly isn’t. And yes, I was also surprised that a Wikipedia page exists for “eyes only”.

“Double-check the verb form after the helping verb”

For starters, what is a “helping verb”? This one is very perplexing:

“I have innate magical powers.”

Grammar checker recommends “innated”. I cannot deduce why. “Innate” is an adjective to describe an attribute present since birth. When I type “innated” into Google, it asks “Did you mean: innate?” I cannot understand why the grammar checker flagged this as an issue.

“Double-check that you’re sticking to singular or plural”

This is a personal favourite of mine:

“…he span around”

Grammar checker recommends “spans”. I’m guessing that it has misinterpreted the verb here, despite the context, and has assumed that the person isn’t spinning around, but spanning across a distance, like a suspension bridge. I did a bit of Googling, and discovered that most Americans use the word “spun” for the past tense, but “span” is perfectly acceptable in UK English (i.e., actual English English). If it’s in the Cambridge Dictionary, it’s good enough for me. But clearly the grammar checker uses Americanisms, even when I have instructed it to use UK English.

“These words work better combined into one word”

Finally, this one made me chuckle:

“I only saw horses.”

Grammar checker recommends “sawhorses”. I just… No. Can it not try and recreate the sentence with its recommendation, and see that this would make NO SENSE?!

With that out of the way: why is it so BROKEN?

This isn’t just me being grumpy and pedantic. There are hundreds of people online complaining about the state of Microsoft’s grammar checker (again: a PAID-FOR service). I’ve even written about its deficiencies before on this blog, over a year ago now – but despite the boom in AI tools, and Microsoft pushing its bloody “copilot” at every opportunity, the grammar checker in Word has not evolved. It is as bad, if not worse, than it ever was.

It seems that Microsoft has been using machine learning to enhance their grammar checker since the 2010s. This involved letting a machine read a huge dataset of “correct” English, so that it could establish the rules for itself, rather than having them dictated to it by humans. According to Microsoft, this is a “natural language” proofing tool, which “performs a comprehensive and accurate analysis of the submitted text, instead of just using a series of heuristics (or pattern matching) to flag errors”.

But this claim doesn’t tally with my observations at all. Most of the issues I listed above were down to Word identifying a specific phase (e.g., “have your back” or “even though”) and ignoring the context, then leaping to the conclusion that it had been used wrongly.

Microsoft defends itself…

While researching this post, I found a list of FAQs on Microsoft’s official pages, which (amazingly) the company deigned to answer. Their explanations were based on the pre-2016 version of Word, but they still made for some eye-opening reading – and I assume that their basic methods haven’t changed since then.

Microsoft claims that “the grammar proofing tool in Word uses advanced parsing techniques to understand the sentence structure” whereas “third-party grammar proofing solutions may rely mainly on “pattern matching””. I would find this assertion amusing if they weren’t being so arrogant. And it gets worse…

“In general, the grammar proofing tool incorrectly marks words or proposes incorrect suggestions when the parser (that is, the grammar proofing component that analyzes the linguistic structure of a sentence) cannot determine the correct structure of the analyzed sentence.”

This doesn’t tell us anything. The grammar checker fails when it can’t understand the sentence. Duh.

“The grammar proofing tool is designed to catch the kinds of errors that ordinary users make every day. You can always make up sentences that may confuse the grammar proofing tool.”

Oh, can we now? We can always MAKE UP sentences to CONFUSE the tool? ALL SENTENCES ARE MADE UP!

“The grammar proofing tool handles some commonly confused word pairs in a uni-directional way to simplify the problem for the parser. The grammar proofing tool was designed this way to reduce the number of items that are flagged by the grammar proofing tool but that are not true grammatical errors.”

Interesting. This is why the grammar checker will flag “your” as a misused word, but never “you’re”. So, the engineers created artificial limitations to reduce the workload of the grammar checker – in other words, overriding the machine learning. This is the most useful information that Microsoft provided in these FAQs. There is no mention of them using pre-programmed lists of common phrases, or of any other instances of manually limiting their machine learning outcome. But, clearly, the machine wasn’t given free rein over when and where to apply its learning.

This will get better as AI improves, right?

Given that these responses apply to an older version of Office, we can only assume that Microsoft has since updated and improved their grammar checker. In 2016, they launched the Editor function, which connects to the cloud to access the machine learning rulebook, which is being constantly updated. Nothing is stored on your PC any more: all of your words are being fed into the aether. Still, Microsoft haven’t released any details whatsoever about how their Editor operates. At this point, I wonder if any of the engineers actually know… The machine has learnt some rules, and is applying those rules, and is now adapting those rules based on text that has been modified by users to appease the machine.

This sounds like a perfectly safe positive feedback loop to me. Absolutely no reason at all to be worried.

In summary…

If you, like me, nurture a sizzling hatred towards the Microsoft grammar checker, I hope you found this post cathartic rather than triggering. Rest assured that I will continue to write my first drafts in Libre Office, safe from the whispering blue lines that tempt me to put commas where I shouldn’t.

Still, I will continue to edit my novels in Microsoft Word – not because of its feeble and ineffective grammar checker, but because of Microsoft Susan and her reliably monotone voice. The AI editor might not be able to spot genuine issues with sentence structure, but neither can my eyes, so I’m happy that Susan is here to help my borderline dyslectix brain (yes, that was my first attempt at spelling that word, and yes, I am leaving it to prove a point). So, although I am deeply suspicious of Microsoft’s AI grammar checker, I would certainly welcome an improvement to AI Susan’s voice.

Stay tuned for updates on Highmoor in paperback!

Happy reading, and have a lovely week!