Grant responds to a voicemail from Doug in Louisville, Kentucky, who asks whether our phone’s autocomplete function will affect the way we talk and write. The answer is yes, partly because of Markov chains, or models describing a sequence of possible events. This is part of a complete episode.
Transcript of “Markov Chains and Your Phone’s Autosuggested Text”
Doug in Louisville left us an interesting message that I wanted to respond to.
He was asking about autocomplete on the phone, you know, when you’re typing and it suggests words.
And he says sometimes it’s not what he wants to say, but he says, sure, and goes with it instead.
And then he wants to know, is this going to have an effect on the way people talk or write?
And what I loved about that question is he’s kind of getting it to one of these core problems that people who study language have.
For example, if I’m looking only at edited texts to see how language is really used, and all of those texts use a style guide, then maybe all that I’m getting is the rules of the style guide and not natural, unbridled speech, unconscious, normal language, which is kind of what you want.
And it’s the same problem with autocomplete.
Before there was autocomplete, generally, you could just download everyone’s text messages with their permission and put them into a big database and examine that and say, this is what’s happening with the way people text and type on their phones, right?
But now with autocomplete, we can’t do that because it’s now polluted.
There’s poison in the well.
And so you have to find text that hasn’t gone through this autocomplete process, some program that doesn’t have that feature, for example, and only get from that particular bit of software.
The other thing that’s happening here that he kind of asked about, but I want to address indirectly, is something known as Markov change.
Markov was a Russian thinker and scientist who died in the 1920s, Andrei Markov.
And basically the way that his ideas are represented today is you have random events when you’re typing.
The computer doesn’t really know what letter is going to follow every other letter.
So what happens is every time you type a letter, it runs a series of probability programs to figure out what letter it thinks is going to be next.
It bases this on frequency tables.
For example, this word is more common than that word or this letter is never followed by these other letters.
So we’re going to just exclude those letters and things like that.
So you can do that fun game on your phone where you type in I, just the personal pronoun I,
And accept every word that it suggests in the middle of the suggestion list.
Just keep going and see if you get a comprehensible sentence.
And sometimes you do.
And the reason you do is because Markov chains,
Because of this math and these frequency tables
And other sophisticated sources of language data,
Which tell the computer and tell you,
Here’s what they might be thinking.
And so, yeah, he’s right.
Doug is absolutely right.
If a lot of people are just saying,
That wasn’t what I meant, but that works,
And go ahead and say yes, then they are polluting the data for people like me who go in and study that kind of information.
It’s a real problem.
Those kinds of autocomplete errors, so to speak, can change the way you talk with intimates.
We had a call years ago about a guy who tried to type the word cool, and it typed the word book instead.
And so he and all his friends used the word book instead of cool.
And this was a thing for a while.
A bunch of people were doing this.
I believe they’re called Cupertinos after some common Apple errors that were happening in some of their autocomplete.
But now this stuff is so sophisticated that sometimes it’s more right than I am.
I can’t believe how good it is.
It’s amazing, right?
And it does it in multiple languages, too.
If you turn on, say, Spanish on your phone as well as English, those Markov chains are right there helping you decide what comes next in a language that maybe you don’t speak as well as your other language.
Yeah, it can be a little uncanny.
Like, my phone knows what I’m thinking.
It’s typing the word before I say it.
That’s a little creepy, right?
So anyway, Doug, thanks for your call.
We really appreciate that.
The term you want to Google if you want more about this is Markov chains.
It’s M-A-R-K-O-V.
It’s super high-level stuff.
But I think you can find some easy explanations on Wikipedia.
