The 'Cambridge email'

Aoccdrnig to a rscheearch at Cmabrigde Uinervtisy, it deosn't mttaer in waht oredr the ltteers in a wrod are, the olny iprmoetnt tihng is taht the frist and lsat ltteer be at the rghit pclae. The rset can be a toatl mses and you can sitll raed it wouthit porbelm. Tihs is bcuseae the huamn mnid deos not raed ervey lteter by istlef, but the wrod as a wlohe.

This text circulated on the internet in September 2003. We first became aware of it when a journalist contacted the CBU trying to track down the original source. It's been passed on many times, and in the way of most internet memes, has mutated along the way. It has also been translated into many different languages (see below for examples). The fact that the email mentioned Cambridge University intrigued us. We work at the Cognition and Brain Sciences Unit, in Cambridge, UK, a Medical Research Council unit that includes a large group investigating reading and language. If research on this topic had been done in Cambridge, then surely we would have heard about it?

Eventually we did track down some research on this, but from Nottingham University, not Cambridge. In subsequent years this general issue of how readers perceive the order of letters in words has generated a lot of research. Most recently, one of us has developed a computer model of reading [1]. When the model is given the Cambridge email it reads it as:

According to a researcher at Cambridge University, it doesn't matter in what order the letters in a word are, the only important thing is that the first and last letter be at the right place. The rest can be a total mess and you can still read it without problem. This is because the human mind does not read every letter by itself but the word as a whole.

We'll say a little more about the model later, but the reason for writing this page is to try to explain the science behind this meme. It contains elements of truth, but also some things which scientists studying the psychology of language (psycholinguists) know to be incorrect. We're going to break down the meme, one line at a time to illustrate these points, pointing out what we think is the relevant research on the role of letter order on reading. Again, this is only our view of the current state of reading research, as it relates to this meme. If you think we've missed something important, let us know [matt.davis@mrc-cbu.ac.uk or dennis.norris@mrc-cbu.cam.ac.uk].

We've tried to keep this as informal as possible, but we've also followed standard academic procedure of providing references in case anyone wants to read the original research papers. The references aren't exhaustive, but the papers we've cited are a good place to start.

aoccdrnig to a rscheearch at Cmabrigde Uinervtisy...

After a bit of rscheearch here at Cmabrigde we think we may have tracked down the first academic work on this topic. In 1976 Graham Rawlinson did a PhD [2] at Nottingham University in which he showed that randomising letters in the middle of words had little or no effect on the ability of skilled readers to understand the text. Indeed one rapid reader noticed only four or five errors in an A4 page of muddled text. Graham has very kindly sent us a summary of his thesis work.

Now we come to some research that really was done at Cmabrigde, although here at the Cognition and Brain Sciences Unit, and not at the Uinervtisy. This work is part of an on-going project to build a computer model of how people read [1]. The core idea of the model is that reading, and indeed, perception in general, is all about making the best possible use of noisy information.

The brain isn't noise-free digital computer; it's a noisy biological system. This means that it takes time to accumulate enough reliable information to read a word, or to recognise a face, or a cup or a car (perhaps a couple of hundred milliseconds or so). But speed is of the essence. To read quickly you want to accumulate just enough evidence to know which word you're seeing. To do this you don't have to extract every last bit of information from every word. If you've got just enough evidence to know that that the word you're reading has got the letters R,E,A and D, but you're not sure whether the letters are in the order 'read' or 'raed', well, there's no point waiting any longer – if it's a word, it must be 'read'. In this illustration we've assumed that you know the letters before you know for certain what order they come in, which almost certainly isn't true. But we do know that changing the order of letters is much less harmful than changing the identity of the letters. [3].

The model simulates this process of 'evidence accumulation'. Like people, it can read the Cambridge email, and like people, it can tell that "raed" isn't really a proper word. It can also simulate lots of data collected in different laboratories around the world, using different languages. But the principles of the model aren't specific to reading. This shouldn't really be surprising. Reading is a relatively recent cultural invention and we almost certainly have to read using a perceptual system that evolved for other things.

FN We won't claim that everybody who works on reading believes that this explanation is correct, but it does seem fair to claim that this model can simulate a wider range of data than any other.

it deosn't mttaer in waht oredr the ltteers in a wrod are, the olny iprmoetnt tihng is taht the frist and lsat ltteer be at the rghit pclae... it doesn't matter in what order the letters in a word are, the only important thing is that the first and last letter be at the right place

If the email had begun: "Arcdocnig to racreesh at Cragdimbe Utinsivery" it might not have had quite the same impact. "Arcdocnig to rrceaesh at the Citonogin and Brian Scenecis Uint" wouldn't have been any better.

The words are still jumbled up according to the "rules" described in the meme. The first and last letters have stayed in the same place, and only the internal letters have been moved. However, we suspect that your experience is the same as ours and that these words are much harder to read.

However, much more convincing than subjective impressions is solid experimental data. Keith Rayner and colleagues [4] measured readers' eye movements while they read sentences where pairs of adjacent letters had been transposed. They found that the reading rate decreased by 11% for sentences containing words with word-internal transpositions and by 36% and 26% when transpositions were at the beginnings or ends, respectively. The fact that readers were slowed down even when internal letters were jumbled up shows that it's not the case that "the olny iprmoetnt tihng is taht the frist and lsat ltteer be at the rghit pclae". On the other hand, readers can cope better when the middle letters are jumbled than when the first and last letters are jumbled. Clearly, the first and last letter is not the only thing that you use when reading text. If this really was the case, how would you tell the difference between pairs of words like "salt" and "slat"?

We're going to list some of the ways in which we think that the author(s) of this meme might have manipulated the jumbled text to make it relatively easy to read. This will also serve to list the factors that we think might be important in determining the ease or difficulty of reading jumbled text in general.

the rset can be a toatl mses and you can sitll raed it wouthit porbelm... the rest can be a total mess and you can still read it without problem

This sentence is, like the rest of the demonstration, strikingly easy to read despite being jumbled. As you have seen above, not all sentences distorted in the same way are as easy as this to read. What is it that makes this sentence so easy? Along with our colleagues we have suggested the following properties:

1) If you're restricted to changing letters in the middle of words, two and three letter words can't change at all. The only change that is possible in a 4 letter words is to swap the order of the middle letters which doesn't cause too much difficulty (see 4).

2) Function words (the, be, and, you etc.) stay the same - mostly because they are short words, see (1). This really helps the reader by preserving the grammatical structure of the original, helping you to work out what word is likely to come next. This is especially crucial for reading jumbled text - words that are predictable are going to be easier to read in this situation.

3) Of the 15 words in this sentence, there are 8 where the letters are still in the correct order. However, as a reader you might not notice this since many of the words that remain intact are function words, which readers don't tend to notice when reading. In fact, whereas we look directly at (fixate) most words about 85% of the time when reading, function words are only fixated about 35% of the time. [5]

4) Transpositions of adjacent letters (e.g. porbelm for problem) are easier to read than more distant transpositions (e.g. pborlem). We know both from the studies of eye movements mentioned above [2] and research in which people read words presented very briefly on a computer screen, that the exterior letters of words are easier to detect than middle letters - confirming one of the ideas present in the meme. We also know that position information for letters in the middle of words is more difficult to detect and that those errors that are made tend to be transpositions. [6]

5) None of the words that have reordered letters create another word. We know from existing work [7], that words that can be confused by swapping interior letters (e.g. salt and slat) are more difficult to read. To make an easy to read jumbled word you should therefore avoid making other words.

6) Transpositions were used that preserve the sound of the original word (e.g. toatl vs ttaol for total). This will assist in reading, since we often attend to the sound of the words even when reading for meaning [8]. When we tried to produce jumbled text that was harder to read than the original (Arcdocnig to racreesh at Cragdimbe Utinsivery) we changed the sound of the words. We also moved the letters further away from their original positions, and arranged some of the words to contain other words ("Arc", "Crag", "dim", "tins", and "very". Notice that the function words "do", "be" and "in" are there too, but very hard to spot. So is "rag", but that's another story).

7) The text is reasonably predictable. For instance, given the first few words of the sentence, you can guess what words are coming next (even with very little information from the letters in the word. [9]

Tihs is bcuseae the huamn mnid deos not raed ervey lteter by istlef, but the wrod as a wlohe... This is because the human mind does not read every letter by itself by the word as a whole.

The question of whether we read by first identifying each letter, or by recognising the word as a whole, has a very long history. Over a hundred years ago Cattell (1886) [10] and Huey (1908) [11] took the view that words are recognised as a whole. Huey found that words with distinctive outlines could be identified at a viewing distance too great for the identification of single letters. However, readers might simply have been working out what the words were on the basis of partial information from some or each of the individual letters. Although there are some more carefully controlled studies that suggest that word shape (the pattern of ascenders such as d, and descenders such as p) might sometimes contribute to reading, some very clever research by Dennis Pelli and colleagues [12] shows that reading really does depend on identifying each letter separately. In their study they presented people with words or letters with very low contrast (very faint print). If words were read as a whole, longer words should have been easier to read than short words or letters (they contain more information). But, using a simple mathematical model, they showed that how well people could read words was entirely predictable from how well they could identify the individual letters. The title of their paper was "The remarkable inefficiency of word recognition" – people can't make use of the 'whole word' information, just the letters.

This simplest summary of current research is therefore that you probably don't always need to read every letter, and there may be some circumstances where you take some advantage of the shape of words, but generally you read each letter separately and put those letters together to read the word. Of course, it feels like we read whole words, but the evidence tells us we don't. (Note that the research doesn't imply that you read each letter in sequence in a left-to-right order)

Clearly, the debate about whether we read using information from individual letters or from whole words is far from over. Demonstrations of the ease or difficulty of reading jumbled texts seem likely to continue to play an important role in our understanding of this process.

Other languages

We've collected examples of the 'Cambridge email' in other languages here

References

[1] Norris, D., & Kinoshita, S. (2012). Reading Through a Noisy Channel : Why There's Nothing Special About the Perception of Orthography. Psychological Review.

[2] Rawlinson, G. E. (1976) The significance of letter position in word recognition. Unpublished PhD Thesis, Psychology Department, University of Nottingham, Nottingham UK.

[3] Perea, M., & Lupker, S. J. (2003). Transposed-letter confusability effects in masked form priming. In S. Kinoshita & S. J. Lupker (Eds.), Masked priming: State of the art (pp. 97–120). Hove: Psychology Press.

[4] Rayner, K., White, S. J., Johnson, R. L., & Liversedge, S. P. (2006). Raeding wrods with jubmled lettres: There is a cost. Psychological Science, 17, 192–193.

[5] Rayner, K., (1998). Eye-Movements in Reading and Information Processing: 20 Years of Research[DN2] . Psychological Bulletin, 124, 3, 372-422.

[6] McCusker, L. X., Gough, P. B., Bias, R. G. (1981) Word recognition inside out and outside in. Journal of Experimental Psychology: Human Perception and Performance, 7(3), 538-551.

[7] Andrews, S (1996) Lexical retrieval and selection processes: Effects of transposed-letter confusability. Journal of Memory and Language, 35(6), 775-800.

[8] Van-Orden, G. C. (1987) A ROWS is a ROSE: Spelling, sound, and reading. Memory and Cognition, 15(3), 181-198.

[9] Stanovich, K.E. & West, R.F. (1983) On Priming by sentence context. Journal of Experimental Psychology: General, 112, 1, 1-36

[10] Catell, J.M. (1886) The time taken up by cerebral operations. Mind, 11, 220-240; 377-392.

[11] Huey, E.B. (1908). The psychology and pedagogy of reading. New York: Macmillan.

[12] Pelli, D. G., Farell, B., Moore, D.C. (2003) The remarkable inefficiency of word recognition, Nature, 423, 752-756.

[13] Velan, H., & Frost, R. (2007). Cambridge University versus Hebrew University: The impact of letter transposition on reading English and Hebrew. Psychonomic Bulletin & Review, 14, 913–918.

Programs and web pages:

1) Ted Warring posted a link to an algorithm that is much better than humans at deciphering scrambled text. This is perhaps unsurprising - I'm sure I'm not the only person out there who has used a computer program to solve a particularly taxing anagram.

3) Peter Hebels has made a Visual Basic program for generating jumbled texts. As he says:

I've made a nice open source program in Visual Basic, this program can do the letter randomiztion automatticly for you. It randomizes only the middle letters of a word, it doesn't change the place of the first and last letters, also special characters like commas and periods aren't affected. You can download the program and source here:

http://home.zonnet.nl/hebels13/letterreplacer.zip

You'll need the Visual Basic runtime files to be on your system if you want to run the executable, the installation for this files can be found here:

http://download.microsoft.com/download/vb60pro/install/6/Win98Me/EN-US/VBRun60.exe

4) Stephen Sachs has written a CGI script to jumble text. Simply enter your text into the