Readability Score

An indicator that expresses how easy a text is to read as a number. Flesch Reading Ease and Flesch-Kincaid Grade Level are representative examples.

Readability score is the general term for indicators that quantitatively evaluate how easy a text is to read. A representative example is Flesch Reading Ease, published by Rudolf Flesch in 1948. The higher the score, the easier the text is to read: plain English falls in the range of 60 to 70, writing for university students in the range of 30 to 50, and writing aimed at readers at a university-graduate level in the range of 10 to 30. It is sometimes described as "an indicator from 0 to 100," but the formula itself has neither an upper nor a lower limit. The theoretical maximum, when every sentence consists of exactly one word of one syllable, is 121.22, and if a sentence is extremely long the value becomes negative (a long sentence in Moby-Dick is calculated at -146.77). A score below 0 is not an abnormal value but an indication that the sentences are too long.

The formula for Flesch Reading Ease is "206.835 - 1.015 × (total words / total sentences) - 84.6 × (total syllables / total words)." It is determined by only two quantities: the number of words per sentence and the number of syllables per word. The derived indicator Flesch-Kincaid Grade Level shows the US school grade level needed to understand a text, and it is calculated as "0.39 × (total words / total sentences) + 11.8 × (total syllables / total words) - 15.59." Grade Level 8 means the level of an 8th-grade student in the US, that is, a level that can be understood with 8 years of schooling, and a Reading Ease of 60 to 70 roughly corresponds to this range.

Because there are only two variables, you can also calculate in advance how much a rewrite will help. Splitting sentences that average 30 words into sentences of 15 words each raises the score by 1.015 × 15 ≒ 15 points. Rephrasing technical terms to lower the number of syllables per word from 1.7 to 1.5 raises it by 84.6 × 0.2 ≒ 17 points. This means there are situations where making the words simpler is more efficient than cutting long sentences. Conversely, simply adding more periods while keeping polysyllabic words will level off at some point. Which lever to pull is decided by looking at which of these two averages is further from the target range.

Besides the Flesch indicators, there are several other readability indicators, such as the Gunning Fog Index (based on long words and sentence length), the Coleman-Liau Index (based on character counts and requiring no syllable counting), and the SMOG Index (based on the proportion of polysyllabic words). Conventions differ by field: Coleman-Liau is used in environments where syllables cannot be counted, and SMOG is used for medical and public health documents. Many SEO tools display Flesch Reading Ease, but no search engine has announced that it uses readability scores themselves for ranking. Treat the score in a tool as a guide for editing, on the premise that it is not a number that promises rankings.

The first stumbling block in practice is that the score is strongly influenced by how the text is divided. Because the rules for estimating the number of syllables (whether to look words up in a dictionary or to estimate from sequences of vowels) differ from one implementation to another, the score for the same text can shift by a few points depending on the tool. When headings or bullet points without periods are joined into a single sentence, the average sentence length jumps, and the score sinks regardless of the quality of the body text. URLs and code fragments are also counted as words. Therefore, rather than comparing absolute values from different tools, the reliable way to use the score is to compare before and after a rewrite with the same tool. If you write a threshold into an internal standard, decide in advance which tool to measure with and whether to include headings and bullet points.

The English Flesch indicators cannot be applied as they are to evaluating the readability of Japanese. This is because the way syllables are counted in Japanese differs from English, and word boundaries are also not clear in the written form. Instead, average sentence length (characters per sentence), the proportion of kanji, the frequency of conjunctions, the rate of passive voice, and similar measures are used as indicators. The guideline of 40 to 60 characters per sentence and a kanji proportion of 20 to 30% has been handed down in editorial practice, but because these are not values derived statistically like the Flesch coefficients, use them as rules of thumb on the premise that they are adjusted to the target readers. If you match the places where someone close to your readership got stuck while reading against the average sentence length, you will find a threshold that suits your own content.

A common misconception is to think that the higher the score, the better the writing. In reality, there is an appropriate range depending on the target readers and the purpose. A medical paper does not need to be turned into writing for elementary school students, and conversely, content for children should not make heavy use of technical terms. The score is only a reference indicator, and it does not guarantee the accuracy of the content. In relation to character counting, if you know the character count, word count, and sentence count, you can get the average sentence length by division, so you can obtain one highly effective indicator without using a dedicated tool. While confirming whether you have reached the target character count, you can also check whether the total number of characters divided by the number of sentences exceeds the upper limit of the guideline. Even this routine alone catches a good number of hard-to-read drafts.

Share this article