Researchers find that Simple English Wikipedia has "lost its focus"

In its September issue, the peer-reviewed journal First Monday published The readability of Wikipedia, reporting research which shows that the English Wikipedia is struggling to meet Flesch reading ease test criteria, while the Simple English Wikipedia has "lost its focus".

The statistical method developed by Flesch (1948) focuses on two core components of the concept of readability: word length and sentence length. The test is widely used in the US, with areas of application ranging from Pentagon files to life insurance policies. The concept has been adapted for other languages, including a German version by Toni Amstad (1978). The Flesch test uses the following formula to indicate the readability of a given text:^[1]

Higher scores indicate material that is easier to read, and lower scores that it is more difficult. While in theory the results can vary widely due to the artificial construction of very complex or simple sentences, in practice natural English typically results in a score between 0 and 100, which can be interpreted as shown in the table.

The authors assume that the English Wikipedia should score around 60–70 on average ("standard"), and Simple English, which explicitly aims at audiences with less advanced literacy skills, around 80 ("easy"). An older study, Besten and Dalle (2008), had found on the basis of the same test method that the overall readability of Simple had decreased from around 80 in 2003 to just above 70 in 2006.

The 2012 study examined two 2010 database dumps it sampled from English Wikipedia and Simple. For the study, the scientists filtered out lists, redirects, and disambiguation pages, and removed components such as tables, headings, and images. Thus, the study examined 88% of the English and 85% of Simple Wikipedia's articles in the database dump. In a second step, the methodology excluded short articles with fewer than six sentences (due to their likely wide fluctuation in readability).

The analysis found that English Wikipedia articles scored 51 on average ("fairly difficult") with more than 70% of all articles scoring less than the set goal of 60 ("standard"). Simple scored 62 on average ("standard") with 95% of all entries below the set 80 ("easy") goal. In addition, a set of around 9600 respective articles was comparable between both Wikipedia versions; Simple scored 61 on these, while the related English Wikipedia articles scored 49.

The paper argues that the creation of Simple as a solution for readability issues of the English Wikipedia with some audiences has run into difficulties. The average reading ease of Simple, while still above the English Wikipedia, declined compared to the findings of Besten and Dalle in 2008 (2003: 80, 2006: just above 70) to 62 on average. Based on the outlined methodology, the authors conclude that Simple has "lost its focus … this version now seems suitable for the average reader, instead of aiming at those with limited language abilities."

The English Wikipedia findings indicate that the results of another study in 2010, focusing on the readability of English Wikipedia entries on cancer (Signpost coverage), cannot be fully generalized. The paper in 2010 found that articles in the targeted topic area scored about 30 on average.

However, both studies show that the English Wikipedia potentially excludes major segments of the English-speaking world, including (for example) large parts of the US public. According to a major study on literacy in the US in 2002, 21–23% (extrapolated: more than 40 million people) "demonstrated skills in the lowest level of prose, document, and quantitative proficiencies".

The authors of the study on readability of Wikipedia have set up a demo site where users can calculate the readability of English and Simple English Wikipedia pages based on the automatic measure they deployed in the paper.

• WMF RfC: The ongoing RfC on whether to establish a legal fees assistance program for volunteer role-specific risks that go beyond the contributor defense policy already in place is in full swing.
• Audit committee call for volunteers: The WMF audit committee, overseeing financial and audit issues on behalf of the WMF board, has published a call for volunteers to serve on the 2012–13 committee.
• English Wikipedia report
  • Arbitration Committee: the committee took several actions this week, including a desysop and passing two motions regarding prior cases (Falun Gong 2, The Troubles). Four clarification and amendment requests remain open.
  • Main-page redesign competition: 23 proposals have been lodged, and discussion about the issue continues on the competition talk page. Editors are welcome to submit their own proposals until 30 September.
  • Pending Changes update: The Request for Comment on Pending Changes Level Two continues. Currently, there is an 11–2 vote on the following proposition: "... if there's no clear consensus for one particular point of view after a week (which seems likely at the moment), is there any objection to trying to come up with a third option in a 'committee' that anyone could join?" That committee page is at WP:PC2012/Committee.

← Previous "News and notes"

Next "News and notes" →

In this issue

10 September 2012 (all comments)

From the editor

Op-ed

In the media

Featured content

WikiProject report

Special report

News and notes

Technology report

Discussion report

Discuss this story

These comments are automatically transcluded from this article's talk page. To follow comments, add the page to your watchlist. If your comment has not appeared here, you can try purging the cache.

• Is there no simple way for editors to find out how complex their writing is? Lotus Word Pro uses to have the means for doing this, but I don't know of any other word processors that can do it. Maybe we might have a gizmo as part of the edit page. Apwoolrich (talk) 08:40, 11 September 2012 (UTC)[reply]
• I had not read to the bottom of the article when writing the above! Apwoolrich (talk) 08:43, 11 September 2012 (UTC)[reply]

• You definitely used to be able to do it using MS word too. I doubt they removed the functionality. Egg Centric 09:27, 11 September 2012 (UTC)[reply]

• MSWord in Office 2011/Mac has a "Show readability statistics" option in the spelling/grammar preferences (results are displayed as part of the grammar-checker tool). DMacks (talk) 13:08, 11 September 2012 (UTC)[reply]

• The demo site is nice, but I think you're on the right track to suggest something easier to access from WP directly would be helpful. I checked a few of "my" articles and was shocked at how low the scores were. I'll definitely make a conscious effort to check the readability of my contributions in the future. Matt Deres (talk) 10:58, 11 September 2012 (UTC)[reply]

• I find the Readability Calculator at online-utility.org (recommended at Simple Wikipedia) to be a big help and gives other readability indices in addition to Flesch. It even tells you which sentences particularly require improvement. Having said that, writing "simply" is fiendishly difficult. I've written a few articles at Simple Wikipedia and really struggle to get the Flesch above 65%. This one has a Flesch of 79 and it took a lot of work to get it there. Voceditenore (talk) 13:12, 11 September 2012 (UTC)[reply]
• A grade school primer ("Dick can run. Jane can run, too. See Dick and Jane run!") scores very high in "readability," but an encyclopedia written thus would sound idiotic. How do the scholarly encyclopedia Britannica and the children's encyclopedia Worldbook score? Edison (talk) 14:58, 11 September 2012 (UTC)[reply]
• I editted Simple English Wikipedia for a couple of years, and drop back occasionally. There are several ongoing problems:

There is a recommended word list. The words on the list were devised by someone who thought they knew what words people really needed. Not all the words on the list are "easy", and not all are "essential". In fact, many of the words that are absolutely essential for encyclopedic writing, are left out. I did make up a list, but don't have it to hand.

To write successful Simple English, one needs to have a very good grasp of the language. One needs to have the flexibility to reduce complex ideas to simple language. Basically, to do it well one needs to have experience in writing for children or for people who are learning English.

Some people who author those pages have limited language skills. Some of them with whom I have communicated have seen themselves as well suited for the purpose for the very reason that they had limited English language.

Having a background in teaching, and specifically in writing educational material for a museum, I know that this isn't the case.

Clarity, in simple words, is not easy to achieve. This is demonstrated by the fact that Simple Wiki editors sometimes string together a dozen one-syllable words in a non-grammatical way, just to avoid using one three-or four-syllable word correctly.

Among the requirements for writing good simple English articles are:

Deciding what vocabulary is essential for an article. This needs determining for every main subject area, and sometimes every individual topic. It may include some complex words that need to be explained within the context of the article itself.

Thinking around subjects for the easiest way to express them. The easy way doesn't always comply with the MOS of English Wikipedia.

I believe that Simple English Wikipedia could be very useful, if well done. But, unfortunately, the way to get it well done, is to get it done by experts.

And that defeats the purpose of Wikipedia. What we need is teamwork, because the people who have the knowledge, the people who are prepared to research to subjects, and the people who are good at writing simple, readable language are not always the same people.

Amandajm (talk) 15:40, 11 September 2012 (UTC)[reply]

An excellent analysis, Amandajm! Re Edison's query about how WP compares to children's encyclopedias. Britannica has the first paragraphs of entries in its Children's Encyclopedia (Ages 6-10) online. I compared the Flesch scores for "Navy" and "Frog" with the equivalent number of sentences from the first paragraphs in the Wikipedia and Simple English Wikipedia articles:

Navy

Children's Britannica = 79

Simple English Wikipedia = 76

Wikipedia = 18 (!!!)

Frog

Children's Britannica = 93

Simple English Wikipedia = 63

Wikipedia = 51

– Voceditenore (talk) 16:20, 11 September 2012 (UTC)[reply]

• I will have to pay attention to this as well (and, thank you for bringing this to my attention): I plugged in six lengthy passages of personal writing (some used on Wikipedia, some not, although my style is similar in both), and came back with Flesch Reading Ease scores between 7.09 [edit: 1.33 for a page out of my thesis - it doesn't get any worse than that!] and 27.28 - but I have always believed my writing was clear and understandable! My scores on the other metrics were equally abysmal, the worst being a 29.68 on the Gunning Fog index (that's, according to Wikipedia, "post-post-post-post-post-doctoral"); a 16.03 on the Coleman-Liau; a 24.48 on the Flesch-Kincaid grade level; a 28.45 on the Automated Readability index; and a 21.11 on the SMOG index (oddly enough for numerologists, another piece of my writing was a 17.77). I plugged in some of Gibbon's Decline and Fall and it scored marginally better than my average. I take it from the above comments that these are abysmally abysmal scores, and that my writings is essentially nonsensical to a large segment of the population. Any tips or suggestions on how to bring my writing in to, at the least, the Flesch Reading Ease 50-60 level? I tried explaining some difficult philosophical and theological concepts in simple terms and scored equally poorly, although I managed to get a 48.21 writing about the Fall of Constantinople. St John Chrysostom ^Δόξα_{τω ^Θεώ} 21:06, 11 September 2012 (UTC)[reply]

While I don't question the conclusion per se, the use of such a primitive metric is suspect. You can't measure how clear or accessible a piece is by sentence and syllable counts, but the difference to the reader is enormous. Simple routinely defines big words and then uses them repeatedly - this inflates the metric but doesn't pose any serious conceptual challenge to the reader. That said, Simple does need help with expansion and simplifying in a lot of areas if it is to keep growing while achieving its goal, rather than turning into an awkward subset of En. Dcoetzee 21:22, 11 September 2012 (UTC)[reply]

See Wikipedia:Village pump (proposals)/Archive 18#Suggestion: readability test(s) for Wikipedia articles (January 2008).

—Wavelength (talk) 21:24, 11 September 2012 (UTC)[reply]

• To move away from the feature, I would advise users to go and have a look at the submissions for the new main page. I was pleasantly surprised, when I reviewed the submissions, due to a particular standout revision of the main page that caught my eye. --Izno (talk) 23:52, 11 September 2012 (UTC)[reply]

• This comment copied from SEWP. This is what I like to call a typical "So what?" research article. The data is freely available, and the tools to generate some numbers are easy to use. But, so what? What is the actual relationship between Flesch Reading Ease and readability? They fail to demonstrate that, so the numbers don't mean anything to actual readers. Quote: "the whole concept of readability cannot be covered". Then don't present this as having anything to do with serving readers.

Plus, their methodology is pretty weak. They remove all of the headings and incomplete sentences. Yet, any editor knows that good use of headings and image captions can make a text much easier to understand. But, if they don't do that, their handy, convenient measurement won't work. Pretty poor. Also, it would have been reasonable to exclude every article on SEWP that is tagged complex from the sample. The community has marked these specifically for readers as not simple. These articles are basically placeholder texts there to be simplified. How about vocabulary frequency (common or uncommon vocabulary) and vocabulary recycling (how aften a word occurs in the text)? I could go on, but I won't.

Finally, they seem to be unaware of newer, more sophisticated tools like Coh-Metrix that go some way towards identifying writing qualities and measuring things like coherence and cohesion that are critical to ease of understanding. Gotanda (talk) 05:40, 12 September 2012 (UTC)[reply]

I wouldn't disagree with you, especially being the author of a very similar paper which is released earlier, and yet covering the same topic more in details and trying to provide some deeper understanding of the phenomenon, instead of just reporting some numbers. 130.233.245.45 (talk) 11:42, 19 September 2012 (UTC)[reply]

• Simple English Wikipedia has a page 'Textual difficulty' which explains some of the basics. The Flesch formula has been extensively researched, far more than any other readability formula, and I can provide references to substantiate this. However, I want to make a more important point.

These formulae were all developed for, and tested on, readers of printed material. Our material is a hypertext (obviously), and this has consequences. In particular, links and wiktionary give ways for readers to check immediately the meaning of words they do not know. In practice, well-constructed Simple pages give the reader these facilities. Therefore, the readability scores are universally overestimating difficulty. On the other hand, printed text in the form of a book is still outstanding in legibility and searchability.

To do better, what we really need is research on people: how our readers actually use our system. In detail, how they find out the answers to questions or topics they wish to know about, how they deal with problems, what their reading ability is, and so on. Who needs to know that some of our articles are terrrible? Of course there are. But the system is self-improving, and I can think of many pages which are far better than they were. This is another of the fundamental issues which get ignored by outsiders looking in. Macdonald-ross (talk) 10:02, 12 September 2012 (UTC)[reply]

you definitely can't easily go from en:wp to Simple. There are many rules, like don't use "phrasal verbs". It would take me a long time to figure out what's ok there. And they're not so concerned with sourcing, in my experience. It's definitely a different culture. As far as how useful it is, I don't know. Are there measure of page views and such? MathewTownsend (talk) 00:49, 17 September 2012 (UTC)[reply]

Re Hypertext

The comment in the project page about hypertext making Simple English Wik easier than the scores indicate needs qualification. First, many of us in less-developed countries do not have fast computers. This means that linking to hyertexts is often eschewed or, if done, breaks the train of thought. Understanding a sentence entails comprehending its entirety. When one goes to a reference and then returns, the train of thought is disturbed. Try this in a language that you barely know to get the feel of such action. You will also see that the slow pace raises your affective barrier. Another problem with hypertext is that the explanations are often not much direct help: either they are so long that a low-level reader (or a person not conversant in the field under consideration) totally gets lost or confused. To see what I mean here, look at almost any article on a technical or even somewhat technical subject in Wik/Eng (e.g., arithmetic): more than likely, you will stumbling through one definition after another trying to get some sort or understanding. This is in part because Wik is a hypertext text with no hierarchy of definitions, unlike, say, a standard text (in, e.g., mathematics), which usually has monothetic definitions. another problem with Wik hypertextualization is that the blue text (along with other font variations) adds a factor of "difficultization" not considered in Flesch. Kdammers (talk) 04:41, 21 September 2012 (UTC)[reply]