The conscious, the subconscious, and the neural network of your smartphone’s keyboard

Before showing up for an examination, the man telephoned Sigmund Freud to fix an appointment. He also made it a point to find out the consultation fee: ten dollars, for the first session.

Sigmund Freud is one of the most well-known figures in psychology even today, mainly for his theories on the subconscious mind. Modern psychology has moved on since then, and Freud is often criticised because his methods were anecdotal and not scientific. Most people, however, still associate “psychology” with Freud’s methods: listening to people speak, observing their behaviour, and digging up hidden childhood memories to explain why those people acted the way they did.

That’s probably what Freud did in this session. The patient then asked the fee again, adding that “I don’t like to owe money to anyone, especially to doctors; I prefer to pay right away.” At least, that’s what he meant to say.

What he instead said was: “I prefer to play right away.”

The patient’s mistake is an example of what is now known as a “Freudian slip”: when a mistake in talking reveals your inner thoughts and desires. Some of these feelings may be unknown to even the person to has them. This was part of Freud’s more general ideas on the unconscious part of your mind, and how it influences memory and behaviour.

That’s why Freud immediately got suspicious. Did the “word” play mean the patient was “playing” with him—that is, pretending to be honest but actually planning to escape without paying? Freud couldn’t do anything, of course, even when the patient reached into his pocket and realised, with apparent surprise, that he had only four dollars in his pocket. He promised to pay the rest by cheque.

A few weeks later, Freud sent a bill to the patient’s address, only to have it returned by the post-office with the message: “Not found”.

Nowadays, you might think there’s less chance to catch a Freudian slip, because people don’t talk. They text.

Well, the point is moot, because it turns out most slips aren’t “Freudian” in nature. Freud-style explanations ran very deep, such as the one related by his friend Dr. Ferenczi, of a middle-aged woman who’d heard of a certain psychiatrist but forgotten his name. It turned out the woman had lost her husband and was too old to remarry, which meant she didn’t want to be reminded of youth or old-age. That’s why her mind also suppressed the name of the psychiatrist, Jung (pronounced ‘yoong’, which sounds quite similar to ‘young’).

In most cases, the reason for slip-ups doesn’t run so deep. It’s more to do with the way language is generated in the brain — which is more technical, but basically means your brain’s working fast to pick out words, and it sometimes picks the wrong ones by mistake.

Some studies still support part of Freud’s theory. People who thought they might receive an electric-shock were more likely to make speech mistakes related to shocks, for instance. And the concept has been around long before Freud: as he himself acknowledged, these revealing slip-ups have appeared in many literary works, including those of William Shakespeare.

However, the fact is, most speech errors are not because of repressed memories — and if they are, they’re usually not as repressed as Freud seemed to suggest. They’re simply language mistakes, like skipping a word, or repeating sounds in the wrong place, or blending two words that are fighting to be used. Or just picking a wrong but similar-sounding word, a mistake even your Autocorrect-enabled keyboard can make.

But wait. Is picking a wrong-but-similar word the only error a keyboard commits? Or do its mistakes point to something deeper?

It was the early ’90s, and the team working on Microsoft Word was busy improving the product.

New employee Dean Hachamovitch, was especially interested in the “glossary” feature, which let you automatically insert stuff. For example, you could type “insert logo”, press the F3 key, and voilà— your company’s JPEG image would appear in place. What Hachamovitch realised was, you could also use it to correct common typing mistakes, like the ubiquitious “teh”.

His idea: make an entry so that, by pressing F3, you can replace “teh” with “the” in the same way you replace “insert logo” with a picture. Then, he had a brainwave. Instead of using F3, he used a much more convenient command-key: the space bar.

And, just like that, the world’s first “AutoCorrect” feature was born.

The early autocorrect used hard-coded lists of common typos and their correct versions. If there was a typo not on the list, you had to add it yourself. Which was still convenient…but it could be made even better. Hachamovitch once gave a presentation to his daughter’s third-grade class on the new feature. Soon after that, he received many emails from parents along the lines of “Thank you for coming to speak to my daughter’s class, but whenever I try to type her name it always gets changed into ‘The pretty princess’”.

The newer AutoCorrects were smarter. They used in-built dictionaries, intelligently picking out the best match for whatever was typed. But dictionaries get outdated. Languages evolve. And it’s a lot of work to keep up with the changing usage of words. That’s why modern autocorrect systems like Google’s don’t have a dictionary at all.

Instead, they’re programmed to learn the language on their own.

You have an incomplete sentence. How do you predict what’s going to come next?

Humans can use their knowledge of language, and the context of what’s been said so far, to make an intelligent guess. Modern machines, on the other hand, have to rely on statistics.

Markov analysis is a simple way of predicting what comes next. Read through a lot of text — say, The Secret Garden by Frances Hodgson Burnett— and make a note of what word follows what. For example, “fresh” is followed by “air” 63.8% of the time, by “scent” 6.3% of the time, and so on. It doesn’t have to be one word either: deeper levels of analysis use longer prefixes, like “What usually follows ‘in the secret ___’?”. With these numbers, you can make a good guess at what the person is trying to type.

If they write “good” they probably mean “night”, but if the type “White” then it’s more likely “Knight”.

And you can go further. Instead of only fixing typos, you can guess what they want to type before they even start typing it. That, of course, is the basis of your modern Android’s “autosuggest” feature.

Today’s keyboards go beyond Markov, of course. They use “neural networks”, a way of calculation modelled on the way the brain learns (or one of the way the brain learns). Apart from studying the whole Internet of texts, they get instant feedback from their millions and millions of users. They learn which typos are most common, which phrases people like to be left alone. The result is a constantly updating algorithm, that picks up words and phrases as they’re being made.

And that’s not all. Keyboards don’t just learn to write. As you use them for your notes and texts, they learn to type like you.

The key to training your keyboard is to always correct its mistakes. Don’t let it get away with “freaky” when you meant “freshly”.

Over time, your keyboard will pick up your style. It’ll learn when you capitalise words, and what private phrases you use. My phone knows to follow “sit” with “rep”, a short form of “situation report” that comes up often while planning Snipette.

The corollary is that if you let autocorrect get away with its mistakes, it’ll soon learn all the wrong things, and you’ll find yourself battling your keyboard more and more often.

Typing is no longer a one-way thing. Often, your phone comes up with a very apt suggestion, that fits better than what you’d have written yourself. Other times, it just writes for you when you’re feeling lazy: my aunt, demonstrating her new smartphone to my brother, began to type “Sent with you sitting nearby” — but autosuggest nudged her to make it “…with you sitting in the region” instead. That’s a slightly extreme example — but as any app designer will tell you, the more available an option is the more likely you are to use it.

Along with the conscious and subconscious mind, today’s writing is now seeing a third factor at play: the automated algorithms of the smartphone keyboard.

What you write can reveal a lot about you. That’s kind of the point of writing. But now, technology is taking the idea further.

Crystal is a service that analyses emails for you, detects the personality of the sender, and gives you tips on how to format your next emails to them. Meanwhile, text-prediction has gone far beyond autosuggest. The Augmented Eternity project plans to “resurrect” people by making chatbot editions of them. Trained on the writings of their originals, these algorithms will chat with you in the same style as the person they’re meant to represent.

Right now, the plan is to load famous historical figures (including, incidentally, Sigmund Freud). But it’s equally possible to make a chatbot of ordinary people like you and me — provided we’ve written enough material to train the bots on, of course.

That idea is actually pretty creepy. Today, people “memorialise” the Facebook pages of departed loved ones, and write posts to remember them. The next step could be to actually chat with them — or rather, with their Augmented Eterntity clone. People can ask them things, and find out things about them, even when they’re long gone.

But perhaps we don’t need to look so far, to find clues of the people around us. I recently got a message from a friend, talking about two of her friends — let’s call them ‘Atul’ and ‘Vijay’. Then she followed up with a message “I meant ‘Rakesh’, not ‘Vijay’!”

It turned out she usually spoke about ‘Atul’ and ‘Vijay’ together, so her keyboard had inserted the latter in this case too.

Until now, keyboard errors have been quite generic. The hundreds of examples on sites like Damn You, Autocorrect may hint about the idiosyncrasies of language, or about people in general, or even about the difficulty of training bots to speak.

But as keyboards get smarter, as they become more attuned to your specific styles, their mistakes will become more specific too. Phrases that you’d only use with some people may accidentally leak over to others.

Keyboard errors can’t reveal your secret hidden thoughts. But if you talk to your friends about your “dumb boss”, autocorrect might accidentally mangle your workplace note on the “dummy box” — and then where will you be?

What autocorrect errors can do is, reveal what you keep secret from some people and not from others. Like the Freudian slip, they make it possible to let out what you’d like to have kept hidden.

Are we witnessing the rise of the Freudian typo?

Have something to say? At Snipette, we encourage questions, comments, corrections and clarifications — even if they are something that can be easily Googled! You can also sign up for our weekly email updates.

Want to know more? Sources and references for this article can be found here.