Readable word jumbles and artificial intelligence

A few yreas ago I was snet an eaiml cliamnig taht Crmgidbae Uienvitsry rrhesaerces had devoecsrid taht txet rieamend rlaebade if the frist and lsat lertets of ecah wrod had teihr poinsitos fexid, whlist the reinimnag letrtes wree all mxied up. Celarly the sepnllig of all one, two and there-lteter wdros wolud raemin crrocet.

Or, to put it another way…

A few years ago I was sent an email claiming that Cambridge University researchers had discovered that text remained readable if the first and last letters of each word had their positions fixed, whilst the remaining letters were all mixed up. Clearly the spelling of all one, two and three-letter words would remain correct.

To put this to the test, I wrote a small program that a would mix text up in this way. It would also, as a comparison, mix up text without anchoring any of the letters. I did this simply for the sake of interest and as a bit of fun.


However, this can be used when discussing artificial intelligence with pupils. The story goes something like this.

How are we able to read? Surely by knowing the spelling of each word. Individual words can then be added together to form sentences and voilà, we can read.

Jumbling words shows that this simple and intuitive answer is clearly wrong, since it is possible to read pieces of text where most of the words are incorrectly spelt. This simple experiment can be extended to show how knowing the correct context of the text that you are attempting to read, normally makes reading easier.

Therefore reading is not just about the spelling of words and it appears that knowing the context before you start reading aids readability. This points towards one of the problems of trying to program artificial intelligence. We are often trying to mimic or reproduce human processes that we still do not fully understand.

The word-jumbling program Azimnag is a freeware download. Amazing! There is a more comprehensive discussion of this phenomenon in the book, Living in a digital world: Demystifying technology in the section of Chapter 5 titled, “Why is AI so difficult?”

This entry was posted in Secondary computing and tagged . Bookmark the permalink.