Turkish Readability Test

What the tool does

The tool counts three values in plain Turkish text: syllables (H), words (W) and sentences (S). From these it derives the average word length in syllables (OKU = H / W) and the average sentence length in words (OCU = W / S), then applies two formulas developed for Turkish and states which band each score falls into.

Each sentence is listed on its own line with its word and syllable count. Sentences above your threshold are highlighted and labelled “long”, so colour is never the only signal. Numbers, links, e-mail addresses, emoji and code fragments that were left out of the measurement are reported separately.

How to use

1. Paste the text. Results refresh shortly after you stop typing; up to 100,000 characters are processed.
2. Choose the measurement. The default analyses the whole text; the optional 100-word sample mode measures only the beginning.
3. Set the long-sentence threshold between 5 and 60 words. The default is 20.
4. Read the warnings: short samples, missing punctuation, unknown abbreviations and excluded items can all move the score.
5. Copy the summary, download the details as JSON, or add the text to a share link if you want to.

The page interface is in English, but the analysis always uses Turkish rules. These formulas are not meant for English text.

The Ateşman formula

Ateşman = 198.825 − 40.175 × OKU − 2.610 × OCU

Ersin Ateşman adapted the Flesch Reading Ease approach to Turkish and published this formula in 1997 in issue 58 of Dil Dergisi. Higher scores mean easier text. The tool uses these bands: 90–100 very easy, 70–90 easy, 50–70 medium, 30–50 difficult, 0–30 very difficult. Classification uses the unrounded score. Very short words and sentences can push the score above 100 and very long ones below 0; the raw score is not clamped and is flagged as out of range.

The Çetinkaya–Uzun formula

Çetinkaya–Uzun = 118.823 − 25.987 × OKU − 0.971 × OCU

Proposed in Gökhan Çetinkaya’s 2010 doctoral research together with Uzun, this formula interprets the score as reading levels: below 35 is frustrating, 35–51 suits instructional reading with support, and 51 or above suits independent reading. The coefficients and sampling approach are discussed in the Turkish Ministry of Education readability publication. The bands do not tell you which age or school grade a text is right for, and the tool never presents a score as a diagnosis of a reader.

Counting Turkish syllables, words and sentences

• Text is normalised to NFC and lower-cased with Turkish rules, so İ and I map to the right letters.
• Syllables equal vowels: a e ı i o ö u ü plus â î û. ğ never forms a syllable; tren has 1, saat 2 and kooperatif 5.
• In-word apostrophes are kept: Ankara’da is one word with 4 syllables. Hyphenated compounds count as separate words.
• Numbers, links, e-mail addresses, emoji and code are excluded; the tool does not invent how a number would be read aloud.
• Abbreviations such as Dr., Prof. and TBMM use a versioned exception table with their spoken syllables. Vowel-less capitals that are not in the table are assumed to be spelled out and are flagged.
• Sentences come from the browser’s segmenter; false breaks after abbreviations and initials (A. Yılmaz) are merged back.

Short samples

Both formulas rely on averages, so in a text of a few sentences one long word can swing the score. Texts under 100 words therefore show a clear short-sample warning next to the result. In sample mode the first 100 words are measured; the sentence containing the 100th word counts as the share of its words that fall inside the sample, and that share is displayed.

Example and interpretation

The example text averages 2.54 syllables per word and 11.3 words per sentence, which Ateşman rates as medium and Çetinkaya–Uzun as instructional. To make such a text easier, split the highlighted long sentences first, then replace heavily suffixed words with plainer ones. Short sentences alone do not make good writing, so meaning and flow still need a human read.

Limits and privacy

• Scores are adapted automatic estimates. They do not measure meaning, factual accuracy, how well a particular reader understands the text, or search rankings.
• The abbreviation table is limited and versioned by date; unknown tokens are listed on screen.
• Text is processed in your browser, never sent to a server and never written to storage. It only becomes part of a link if you choose to include it when sharing.