Can AI help us with our writing?

I spent my last three posts discussing the failure of artificial intelligence, even state of the art AI, to understand what we say to it and what it says to us. But understanding, surely, isn’t everything. Real-world communications aside, how does AI fare re one of Catherine’s and my other core interests: linguistic tools for clear, coherent sentences?

Two elements of clear coherent sentences are word choice and punctuation, and some of the more common word choice and punctuation errors have long been handled by basic word processors. In the last decade, however, more sophisticated tools have emerged: tools like ProWritingAidGinger,  WhiteSmoke, and Grammarly. The latter, with over 10 million users, is hard to miss unless you completely avoid YouTube. If you survey Grammarly’s Internet reviews, you get the sense that it is the most sophisticated—and most expensive—of the new tools. It’s also the easiest one to get information about without buying a subscription. So Grammarly is my focus today.

To what extent does Grammarly’s writing feedback go beyond that offered, say, by Microsoft Office? To what extent does Grammarly help people revise entire sentences? Answering these questions directly involves accessing Grammarly’s deluxe version at $29.95 per month. So my answers—incomplete though they may be—come instead from reviews written by existing subscribers, and from the promotional videos offered by Grammarly itself.

Many of the Internet’s examples of Grammarly’s feedback are about obvious or common punctuation and word-choice errors: misplaced commas (as in “I’d really appreciate it, if”); “your” vs. “you’re”, “its” vs. “it’s”, “their” vs. “they’re”; “lie” vs. “lay’; “affect” vs. “effect”; “than” vs. “then”; “would of” for “would have.” But Grammarly also detects some less common errors like “thorough” for “through.”  And, going beyond outright errors to matters of style, it suggests synonyms for unnecessarily fancy words (“utilize”), for over-used or bland words like “good”, “small,” “strong,” and “interesting,” and for words that you, the writer, have used repeatedly.

Grammarly also flags words for outright deletion: potentially gratuitous intensifiers like “really”, “very,” “actually,” and “definitely”, along with “needless to say”, and “personally” in “I personally.” Also flagged are wordy phrases: 

Original phraseGrammarly’s suggestion
It is necessary for us toWe should
It’s critical for us toWe must
I think we shouldLet’s
in terms ofregarding/concerning
make an effort totry
make a decisiondecide


Some of Grammarly’s suggestions address both wordiness and blandness:

Original phraseGrammarly’s suggestion
inspect carefullyscrutinize
really hungryravenous
really bigtremendous
really helpfulbeneficial

At the level of grammar, Grammarly detects the missing ending in “many option”, the wrong pronoun form in “She helped myself”, the tense error in “if you will leave now”, and the missing preposition in “you’ll be amazed what you can do.” It also detects sentence fragments and comma splices. It suggests shortening noun phrases of the form “the X of the Y” or “the X for the Y” (“the scope of/for this project”) to “The Y’s X” (“this project’s scope”). For sentences of the form “If you X, or if you Y” it suggests eliminating the second “if you” (“If you X or Y”). It also flags passive voice, which it is not a fan of, as well as clauses that end with a preposition, like the one you just read. Any tendency to blithely split infinitives is also reliably commented upon.

But Grammarly’s syntax skills are limited. It correctly flags “It not only expands your mind but also your soul”, noting that “the structure after ‘not only’ does not match the structure after ‘but also’”. But rather than hazard a guess at revision, it simply says “consider revising”. It correctly identifies the dangling modifier in  “Following from that, another option is to introduce a menu”, noting that “‘following from that’ does not appear to be modifying the subject ‘another option’”. But then it simply adds “Rewrite the sentence to avoid a dangling modifier” (followed by a canned mini lesson on misplaced modifiers). It readily flags long sentences and suggests that you split them up; it doesn’t tell you how. Sometimes it simply says “rewrite the sentence.”

Its disdain for passive voice is so reflexive and undiscriminating that it flags sentences like “when the active subject is unspecified, passive voice is often less wordy” and “I’m tempted to do all the editing myself”.

And sometimes it flags well-placed modifiers as misplaced, as in this example from English Stack Exchange: “Seeing the potential in data, the industry has already started to employ cognitive computing.”  As far as Grammarly is concerned, industries can’t see things.

Other examples of Grammarly finding errors where there are none include these, from Reddit:

Original phraseGrammarly’s suggestion
He has workHe has worked
They’re creating them on purposeThey’re cheating them on purpose
to potentially republish your work at a later dateto republish your work at a later date potentially

At the level of tone, Grammarly has been trained, via machine learning focused on key words and phrases and on punctuation, to rate a text’s friendliness, confidence, and level of formality (with phrases like “don’t you think” flagged as showing low confidence). Calculating syllables per word, words per sentence, and the number of unique or rare words, it also rates texts for readability. While such ratings serve as useful approximations, more goes into both tone and readability than what is captured by metrics that, as we’ve seen even in state of the art AI, operate mostly at the level of individual words and phrases.

Which means that there’s a whole set of sentence-level revision tools that Grammarly doesn’t even touch—and which (since the run-on sentence fragment which I’m going to—because I’m tempted to—blithely end this post with can’t really handle them) will have to wait for my next post.

Links

4 thoughts on “Can AI help us with our writing?

  1. Surely Grammarly didn’t recommend “ravenous” as a rewrite for “really helpful.” Were you hungry when you posted this?

    Like

    1. I guess so! I became aware of the error a few days ago but couldn’t be bothered to fix it–until now. I was wondering if anyone would notice.

      Happy New Year, Beth!

      Like

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s