- From: Charles McCathieNevile <charles@w3.org>
- Date: Fri, 4 May 2001 03:08:26 -0400 (EDT)
- To: Matt May <mcmay@bestkungfu.com>
- cc: <w3c-wai-gl@w3.org>
I agree that flesch-Kincaid and gunning fog are at best very rough guesses. As you say, they reward short sentences made of short words. It is true that this is only a rough measure. And I agree that to give good emasure, some grammar checking helps. And spelling chicking is not quite as high a priority - if the spellchecker can recognise the word most people will too, whatever the spelling. But checking whether words are in a common dictionary (more or les the same process but using a different dictionary) would be a useful test. If we could specify three or four simple tests at the same time, and provide tools to do these tests, we would be well on the way to helping people get a rough idea. Another test that Jonathan Chetwynd has often suggested is word count. If there are more than 30 words on the page, he says, people that he works with find it very hard. This would also be useful. (Note that all this is in theory required by WCAG 1.0...) cheers Charles McCN On Thu, 3 May 2001, Matt May wrote: I've done some research on reading level algorithms, and I think what I've found is enough to give us pause with respect to putting any real faith in the numbers that are produced, or basing any guidelines on them. The first one I researched was the Flesch-Kincaid grade level. The algorithm is as follows: (.39 * w) + (11.8 * s) - 15.59 where w is the average number of words per sentence, and s is the average number of syllables per word. Negative FK results are reported as zero. Numbers over 12 are reported as 12. FK is based on the Flesch Reading Ease score, which is itself just a measure of syllable count and sentence length. The modified algorithm was published in 1975 by a researcher trying to create readable documents for enlisted personnel in the US Navy, and doesn't appear to have any real attachment to education, much less cognitive disability. The subjects in Kincaid's research were adults, presumably skewed 18-30 and male (and by definition skewed American), and the resulting algorithm really wasn't intended to be utilized as widely as it is. The other major grade-level index is the Gunning Fog index. Its algorithm is: (w + h) * .4 where w is the average number of words per sentence, and h is the percentage of words with three or more syllables ("hard words") Gunning Fog is capped at 17, where "17-plus" is suggested as postgraduate-level writing. Both of these algorithms reward short, monosyllabic sentences, irrespective of how many of these sentences are necessary to communicate the point. These indices are also not meant to rate an entire document, but rather extrapolate scores from small passages (100 words). I picked a couple of random examples to illustrate results. The first sentence of Lincoln's Gettysburg address rates a 12 (the raw number is closer to 18) on FK, while Gwendolyn Brooks' poem "We Real Cool"[1] rates a 0. I was a third-grader when I studied Lincoln, not a post-doctorate, and I think I still got the gist. :) Now, here's the part that bugs me. First off, by interjecting sentences like this previous one (short, no polysyllabics), I can lower the overall score of this message. And if the goal of a site is to work its way down to a prescribed reading level, then that is likely to be what they try to do: they'll boil down the content by working it until it comes up with a low enough score, and say that's that. No real usability or accessibility gain can be found by fostering this type of practice, where people are writing to the index, rather than to the reader. Secondly, syllable count is a really weak measure of complexity. Is "Germany" a more difficult message to communicate than "France"? Do 7-year-olds know what a brad is? The idea is based on an assumption that longer words are harder than short ones. There's a correlation there, but it's not reliable. Just as relevant to comprehension are educational environment, cultural influence, and above all, context. My last problem is that none of these actually check spelling or grammar. It seems those might be somewhat relevant, as well. The information that I read seemed to suggest (where it didn't say outright) that reading-level indices have been used largely as pseudoscience: an overly simplistic "scientific" numeric answer to the readability of a system. I found a document[2] which lines out ten principles of clear statement. I think this is a lot closer to my ideal of providing content providers with solid guidance for good writing, and in fact, we may want to consider asking to incorporate these principles. This document comes from the creators of the Gunning Fog index, circa 1973, and at the end, it emphasizes that systems like the fog index, while of some utility, are not a panacea: "It is important not to over-use the fog index. Use it only occasionally to spot-check your writing. Don't write to make a good fog index score. That will make you write short, choppy sentences. Like these. Instead, learn and practice using the 'Ten Principles of Clear Statement.' If you observe these guides to good writing, your writing will naturally grow easier to understand." [1] http://www.poets.org/poems/poems.cfm?prmID=1233 [2] http://muextension.missouri.edu/xplor/comm/cm0201.htm - m -- Charles McCathieNevile http://www.w3.org/People/Charles phone: +61 409 134 136 W3C Web Accessibility Initiative http://www.w3.org/WAI fax: +1 617 258 5999 Location: 21 Mitchell street FOOTSCRAY Vic 3011, Australia (or W3C INRIA, Route des Lucioles, BP 93, 06902 Sophia Antipolis Cedex, France)
Received on Friday, 4 May 2001 03:08:40 UTC