What generative predictive text can teach us about writing style

What is predictive text and how does it work?

By analyzing writing behavior, machine learning curates a personalized glossary of words and phrases people generally use for various circumstances. To do this, predictive text uses language modeling, the use of statistical techniques to determine probability and guess when a writer plans to use them next. Over time, predictive text improves to the point where a user starts a sentence and can finish it accurately using predictive text, saving them the time of typing it all out manually.

Write with Confidence using Editor

Elevate your writing with real-time, intelligent assistance

Learn more

Predictive text was first developed for people with motor disabilities or poor typing skills, but it provides many benefits for users with a wide variety of typing abilities as well. Predictive text can:

• Improve typing speed.
• Improve typing speed.
• Serve as an organic text editor.

Predictive text reveals truths about human behavior and writing styles

Researchers are fascinated with how predictive text influences human behavior and have conducted numerous studies on the subject. Through their findings, we writers can learn a lot about our own writing—both with and without predictive text.

Be concise and edit your work

A Harvard study found when people use predictive text, they write shorter sentences than when typing out a sentence manually without predictive text. Predictive text suggestions encourage a writer to skip over words that may not be essential to the overall message. For example, without predictive text, a user might write, “Do you think I should go ahead and buy the red shirt?” With predictive text, a user might write, “Do you think I should buy the shirt?” Both sentences carry the same meaning, but the second is more concise. In the study, most people preferred the shorter, more concise version.

What does this tell us about our writing styles? When left to our own devices, we tend to provide additional details not necessarily essential to our overall message. Your message strengthens when you write it concisely. There are times when the more concise message reigns superior: In an email, research paper, or text message; but there remain instances when you benefit from the more embellished message, like when writing a creative writing piece. Edit your work, resist the urge to overembellish, and cut out the nonessential parts to strengthen your message.

You are the authority of your writing

Another study found that predictive text heavily influences a writer’s text. Many writers in the study went with the suggested text even if it contradicted what they were originally trying to say. They assumed the predictive text was more accurate and authoritative than themselves and doubted their own writing style and opinions. Other writers struggled against the predictive text and noted biases in the suggestions. For example, when writing a negative movie review, the predictive text kept offering positive suggestions even though it was not in line with the writer’s message. While predictive text may seem like an authority on writing, it’s not. It’s simply a suggestion. It’s up to the writer to choose whether to take the suggestion or write something else in line with what they’re really trying to say.

What can we as writers learn from this? Be confident in your own abilities and opinions. You have a well-developed vocabulary, excellent grammar skills, and strong opinions that the world needs to see, and no language model can replace you.

Context is everything

Perhaps the most comforting takeaway for writers entering the age of AI, is that context is everything, and only humans can provide that. Predictive text uses an algorithm model called the “Markov chain.” In doing so, it offers suggestions based on the most recent text. However, the Markov chain algorithm has a short memory and does not consider the previously written text. When using predictive text creatively, a sentence will often loop in circles—not understanding that it said the same thing twenty words ago.

What can we learn from this? Our words are only as good as the mind behind them, and you need an actual human to provide valuable context, direction, and editing.