Don’t like the way you write? An artificial intelligence app promises to polish your prose

This content has been archived. It may no longer be accurate or relevant.

From Quartz:

I am a professional writer, but I often hate my writing. I wish it was more concise and powerful. And it certainly doesn’t read as smoothly as the work of my literary heroes. Recently, I began to wonder: Could a software program make me better at my job?

The Hemingway App, an online writing editor created in 2013 by brothers Adam and Ben Long, promises to do just that. “Hemingway makes your writing bold and clear,” the site claims, so that “your reader will focus on your message, not your prose.” If you listen to the app’s advice, it will rid your writing of run-on sentences, needless adverbs, passive voice, and opaque words. There’s no guarantee you’ll crank out the next Farewell to Arms—but the goal is to get you closer to Ernest Hemingway’s clear, minimalist style.

The app uses a crude artificial intelligence that recognizes writing problems through natural language processing. When you copy and paste your text into the Hemingway Editor, it highlights sentences with possible issues in different colors and offers suggested changes. For example, if I write, “This Editor has been used since around 2013,” the words “been used” are highlighted green because I am using the passive voice.

Most of the recommendations offered by the Hemingway App are based on research into readability—that is, how easy it is to understand a given text.

. . . .

I also ignored some of the app’s suggestions. A few were just nonsensical. It suggested I replace the word “demonstrate” with simpler synonyms “prove” or “show,” but I was talking about people going to airports to protest. I rejected other suggestions for stylistic reasons. The app wanted me to remove “really” from the sentence “As it turns out, protest size really does matter.” But I wanted to keep the conversational tone.

In the end, I was able to bring the grade level of my story down from 13 to nine, and shed 34 words along the way. Then I gave the updated version to Kira Bindrim—a Quartz editor who’d edited the original story.

“I think my gut reaction is to prefer the original,” Bindrim wrote to me after reading the Hemingway version.

Link to the rest at Quartz

PG wonders if there’s an app that really works for professional authors.

PG also suggests that Hemingway might crash if he fed some typical legal documents into it.

18 thoughts on “Don’t like the way you write? An artificial intelligence app promises to polish your prose”

  1. im just wondering
    how one is a ‘pro’ writer
    who has an amateur tude

    I have a short list of words that I use too often. Just search for them global and all are highlighted in blue and I can choose to keep or delete. Easy.

    I dont think this idea of auto anything, will work with nonfiction in which voice and tone are everything. Maybe in tech, but have a hunch Guy Kawasaki would spurn it.

    • Word’s spellcheck picks up repeated words. Quite why it’s the spell and not the grammar check that does I don’t know but this is for the best as I ignore the green underlining of the grammar check when I’ve forgotten to turn it off.

  2. Nothing new here. There was a program that did the same thing for WordStar files on a KayPro running CP/M.

    The Air Force also had a program that did something similar for checking efficiency reports in the mid 80’s.

  3. Does it have a menu so one can choose what kinds of things to flag?

    I use the word, “that” too much. So I have Word search for “that,” and make sure it isn’t superfluous. I also continually confuse “form” and “from” when typing. So Word searches for them.

    Maybe someone does think they use too much passive. That might be a good thing to check.

    A tool that allows the writer to select for specific things can be useful.

  4. I’ve tried Hemingway. I purchased ProWritingAid and use it occasionally to keep myself on my toes. When I chose ProWritingAid, it had more functions that looked useful to me, but like all software, they compete and what one has, the other steals. There are also free online tools.

    I think this kind of tool is helpful, if you have the right attitude. You have to use their advice when it helps and ignore it when it doesn’t. They are not very sophisticated. They can spot over used words and some low-content sentences. When I find myself drifting into corporate-speak, they help me get my bearings.

    One of the most useful function is spotting over-used words and phrases, but ProWritingAid gets slow when the text gets long, which is exactly the most important use case for me. I wrote a little batch Perl script that is much faster and I prefer to use it.

    A decade or so ago I was a contributor to a style and grammar extension for LibreOffice. It was built on a engine that uses Yacc-like capabilities to parse sentences with fair accuracy and a massive set of contributed rules for flagging issues and making suggestions. My experience was that your results depended on the quality of the rule set you happened to have. It was an international effort and rules sometimes crept in that were written by well-meaning non-native speakers. I believe it’s still being improved on if anyone is curious. (LanguageTool). I quit messing with it when using LibreOffice got to be too much of a pain because too many people I work with insist on Word.

    LanguageTool was definitely not machine learning. (It may have some elements today, I don’t know.) I suspect that Hemingway and ProWritingAid are the same kind of technology.

    I’ve sketched ideas for a neural net based tool, that would combine with a Yacc type parser, but I prefer writing English to computer languages these days, so I haven’t put much effort into it, especially because Msft appears to be paying kids in Redmond to build more machine learning into Word.

    • Any time someone (or something) says that adverbs and passive voice are always bad and should never be used, I know they don’t have any idea what they’re talking about and I no longer need to listen to whatever else they say.

      Also, “hard to read” is not exactly useful, actionable feedback.

      I guess maybe these sorts of apps are useful for kids writing papers or adults who don’t know anything about writing who are doing reports or something. But for fiction writers? I don’t see it.

  5. These pieces of software may be of limited use for people who write nonfiction and aren’t professional writers, but they are death to fiction which is all about voice. Either they make the sentences flat and uninteresting, or they constantly flag fictional routine usages as errors.

    • I would suggest that nobody should pay the slightest attention to them in dialogue. I have met perhaps two people that speak the same way as they write – among many, many who are far more formally educated than I.

  6. This sounds dreadful. Word processing software has come with grammar and style features for a long time. They are the first thing I turn off, should the default setting have them on. Flagging “really” and whatever it thinks are passive constructions is a dead giveaway that this is more of the same. Flagging passives is a sure sign of general uselessness. Good writers routinely use passive constructions where appropriate, a century of bad writing advice notwithstanding. Indeed, even the people giving this advice use passive constructions routinely. E.B. White, of Strunk & White, certainly did. And why not? Passive constructions are a valuable tool in English. It would be absurd to shackle yourself by not using them. Of course they can be used poorly, and a good editor will suggest revisions. But a mindless ‘passive-Bad!’ editor? Not so much.

    I am also bemused by the premise that reducing the reading level of a text is the same thing as improving it.

    • I use the word counter (essential, if quirky – depending on how you select text, it gives you slightly different counts). I use the spell checking, as typos happen. (It’s also quite useful, with a custom dictionary, to avoid calling a character “McAlister” in one place and “MacAllister” in another. Also, whether I put two “l”s in “McAlister” – the name is most commonly with one “l,” not two – fun fact…)

      I use the word counter at the end of every revision (for production tracking). I only use the spell checker at a “draft” point. On the “finished” product, I do run the grammar check – I used to have the “passivity” habit to excess, but rarely change things nowadays. If it were to highlight something like “demonstrate,” I would ignore its suggestions – but at least consider changing the word to “protest,” a more commonly used word (or “riot,” if that is what I am really describing).

  7. ‘but I often hate my writing’

    Bad start. So many articles begin with a premise – upon which the whole article leans – slipped in without much of a comment.

    I find more and more of these premises are wrong.

    ‘Everyone knows ebook sales have dropped’

    ‘The emergency response team failed to…’

    ‘Amazon is bad because…’

    People then read the article looking for a solution to a problem – that isn’t.

    What writing has made me is far more critical (in the sense of critiquing, not so much as of fault finding though there’s that, too) of what I read.

    Occupational hazzard, I understand.

    But these people get published, and sometimes paid, based on tripe.

    It annoys.

    The idea that an AI can write anything other than sports scores or weather reports (which even professionals struggle to make new, given that there’s almost nothing new in the content except dates and names) is still ludricrous.

    And the idea that they would give a professional writer usable suggestions is more ludricrous still.

    Grow up, pro. Learn to write. And use machines as God intended us to: to count the adverbs and the number of places you’ve written ‘its’ – so the human can do its job.

Comments are closed.