blatherskite: (Default)
blatherskite ([personal profile] blatherskite) wrote2016-08-05 11:03 am

Why readability indices (formulas) are nonsense

A colleague recently wondered about the usefulness of readability formulas such as those that most word processors offer. My take? They're nonsense. Not only useless, but possibly actively misleading.

What evidence do I offer as proof?

First, take any elegantly written sentence of about 10 words in length, then rearrange the words to reverse the meaning. Randomize the words to make the sentence incomprehensible without heroic effort. Rearrange the words with malice aforethought to make the sentence as difficult as humanly possible to comprehend. You'll calculate precisely the same readability index value.

Second, take any sentence and insert random (or deliberately malicious) punctuation after every word. Surprise! You get the same readability index.

Third, take any easily comprehended sentence and replace all the words with nonsense words of the same length (use random letters or change all the letters to "z"). You'll generally get exactly the same index value.

Any soi-disant "readability" index that cannot provide realistic differences in readability values for the different phrases in each of these three tests is useless. Q.E.D. To obtain a valid measure of readability, it's necessary to parse the sentence for structure and meaning. Currently, it takes human eyes to make this judgment.

But if you want proof from a real expert, try Brad Connatser's article "Last Rites for Readability Formulas".