Why Can You Read Words with Scrambled Letters? How Typoglycemia Works

What is typoglycemia?

The word describes how text stays readable even when letters inside words are swapped. In English, a passage beginning "Aoccdrnig to a rscheearch at Cmabrigde Uinervtisy" is the famous example, and text with the same idea also circulated widely in other languages, including Japanese, as a chain email.

But "typoglycemia" itself is an internet coinage, combining "typo" with "hypoglycemia" (low blood sugar), and it isn't an academic term. The phenomenon has been studied, but keep in mind that this word isn't its formal name.

The conditions: not every swap stays readable

The sample text reads well because the swaps are carefully chosen. Break the conditions, and it quickly stops being readable.

A scrambled "hlelo" reads easily because you can predict a greeting. The same kind of swap makes an unfamiliar proper noun, or strings where the order itself carries meaning, such as numbers and model numbers, completely unreadable. In other words, this happens because you're matching against words in your memory rather than reading the sequence of letters.

• In alphabetic words, the pattern of letters that rise above or drop below the line (b, d, k, p and so on) forms the word's outline, so whether this is preserved affects readability.
• Japanese text in hiragana alone maps one character to one sound, so swaps have a strong effect, but the visual length of a word is preserved, which helps with restoring it.
• In Japanese mixing kanji and kana, the kanji act as landmarks for chunks, so the meaning comes through even if the kana are somewhat jumbled.

The truth about the "Cambridge University study"

The English sample begins with "according to a study at Cambridge University", and this has carried over into versions in other languages. But no corresponding Cambridge study has been found. It's thought to be a line added to make it look sourced as it spread as a chain email in the early 2000s.

A Cambridge researcher has pointed this out and published an explanatory page, which also shows that "you can read it as long as the first and last letters are right" is an oversimplification. In practice, how hard the text is to read varies greatly with where the swaps fall.

Letter position and word recognition have been studied for some time. A British doctoral thesis from the 1970s dealt with how letter position affects reading words, and it's often cited to explain this phenomenon. Studies measuring eye movements have also reported that text with swapped inner letters can be read, but about 10% more slowly, and even more slowly when the first letters of words are swapped.

You miss typos because of the same mechanism

Failing to find typos in your own writing isn't only a matter of attention. The cause is the same mechanism as typoglycemia: rather than reading letter by letter, you match against words in your memory.

• As the writer, you already know what word comes next. Strong prediction means you skip ahead without looking at the letters actually written.
• The closer a typo is in shape to the right word, as with "form" and "from", the less you notice it. Errors that look very different catch your eye more easily.
• Autocorrect and input slips are real words in themselves, so they don't show up as misspellings. Only context can reveal this kind of error.
• The more you reread the same text, the stronger your prediction gets. This is why rereading right away does so little.

So proofreading calls for deliberately creating a state where prediction doesn't work. Here are the main methods.

How to read with prediction switched off

1. 1

   Reread after some time has passed

   Right after writing, the content is still in your memory, so prediction is strong. Leave it for even half a day, and as your memory fades, the letters become visible.

2. 2

   Have it read aloud

   Text-to-speech turns the written letters straight into sound, so your predictions can't get in. Repeated or missing words stand out to the ear as something unnatural. If it pauses in an odd place, the sentence structure itself may be hard to read.

3. 3

   Read backwards from the end

   Reading from the last sentence to the first stops prediction from the flow of the story, so you focus on how each sentence is written. To check letter by letter, look at a reversed version of the text, which can't be read for meaning and makes you follow the letters themselves.

4. 4

   Change how it looks

   Changing the font, text size, line spacing or background color makes the same text look different, so you're less likely to skip over a shape you've memorized. Printing it out works for the same reason.

5. 5

   Check one type of thing at a time

   Look at only one type at once: typos, numbers, proper nouns, link destinations. Prediction doesn't work on numbers and proper nouns, but errors there have a big impact, so check them separately from the body text.

Numbers and proper nouns are where this phenomenon works least. They don't get the "it just reads" effect, so errors are easier to notice, yet wrong digits look similar and get missed. If your text has a length limit, make the last step of proofreading a character or word count to confirm it fits. How to proofread by reading aloud is covered in tips for proofreading by reading aloud, and how to count in how to count words and characters.

FAQ

What is typoglycemia?

It's a word for how text stays readable even when letters inside words are swapped. It's known from examples such as "Aoccdrnig to a rscheearch at Cmabrigde Uinervtisy". But the word is an internet coinage combining "typo" and "hypoglycemia", not an academic term. The phenomenon happens because you match against words in your memory rather than reading letter by letter.

Is the Cambridge University study real?

No study corresponding to that line has been found. It's thought to have been added to make the text look sourced as it spread as a chain email in the early 2000s. A Cambridge researcher has pointed this out on an explanatory page. Letter position and word recognition have long been studied, but the explanation attached to the sample text is oversimplified.

Can any text stay readable when scrambled?

No. It only works when the first and last letters are fixed, the swaps are between neighbors, the words are short, the context lets you predict the word, and the word is familiar to begin with. Unfamiliar proper nouns, technical terms, and strings where order itself carries meaning, such as numbers and model numbers, can't be read at all.

Why can't I find the typos in my own writing?

As the writer, you know what word comes next, so you read ahead by prediction without looking at the actual letters. It's the same mechanism as typoglycemia. What helps is deliberately creating a state where prediction doesn't work: read after some time has passed, have it read aloud, read backwards from the last sentence, or change the font or background color.