Haiku / Tanka Syllable (mora) Counter

Why an unpublished haiku or tanka deserves careful handling

A mora-count check is almost always done mid-composition: you are revising a draft haiku for a poetry contest, polishing a tanka before sending it to a literary magazine, or fine-tuning a verse before a haiku club meeting. All of those moments involve work that is not yet public. Unpublished poetry carries both copyright significance and a creative vulnerability — it is an expression that the author has not yet chosen to release.

Editors and writing coaches face the same issue when checking the syllable counts of manuscripts entrusted to them by authors. Passing a client’s poem through a server-side tool is, arguably, a breach of the implicit confidentiality that surrounds unpublished work. A mora-count check seems trivial, but the vehicle matters.

The logging risk of online poetry analysis services

Server-side mora-count services receive your poem as an API request, process it remotely, and return the result. That flow produces a server log: a timestamp and the text of the verse. Services that offer morphological analysis ‘for free’ often monetise through advertising or by accumulating text data, sometimes reserving the right in their terms to use submissions for ‘service improvement’.

The commercial market for AI trained on Japanese poetry is growing. Uploading draft verses to an unvetted service is a potential — even if unintended — contribution to a corpus being used for commercial training. The risk is structural, not a matter of any one operator’s intent.

kuromoji and in-browser mora decomposition

This tool runs kuromoji.js (Apache 2.0) with the IPADIC dictionary inside your browser. When you paste a verse, the browser’s JavaScript engine tokenises the text, converts each token to its katakana reading, and then splits the reading into mora. The rules — yōon syllables (キョ, シャ, etc.) merge with the preceding consonant and count as one; sokuon (ッ), hatsuon (ン), and chōonpu (ー) each count as one — are implemented in local JavaScript.

Aside from the one-time dictionary download (~12 MB, cached after first use), all processing runs offline. Open DevTools → Network and edit your verse repeatedly: no outbound request carries your text. The revision cycle — write, count, revise, recount — stays entirely on your device.

Fitting a local mora check into a composition workflow

A practical approach: use the sample button to verify the tool is working correctly on the well-known Bashō verse, then switch to your own draft. This keeps the test and the live composition separate, and means you only enter unpublished work after confirming the tool behaves as expected.

For poetry contest submissions in particular — where the work must be original and unpublished — running all pre-submission checks in a browser-local tool is a clean habit. It removes any ambiguity about whether the poem was ‘submitted’ somewhere before the official submission.

Mora vs. syllable and the rules specific to haiku counting

Japanese counts sound units in mora (拍), not syllables. A mora is a unit of equal time length: in hiragana, single characters like あ・い・う・え・お or か・き・く・け・こ are one mora each, while yōon clusters like きゃ・しゅ・ちょ are two characters but one mora. Sokuon ッ, hatsuon ン, and chōonpu ー each count as one mora on their own. 東京 (トーキョー) is four mora (ト・ー・キョ・ー), and 札幌 (サッポロ) is four mora (サ・ッ・ポ・ロ) — the lengthening and the small tsu pay for their own beat. If you want to see the full kana reading of a kanji-heavy draft before counting, kanji-to-hiragana runs the same kuromoji pipeline.

In haiku and tanka the mora rule is applied strictly. Bashō’s 古池や蛙飛び込む水の音 parses as フルイケヤ・カワズトビコム・ミズノオト, exactly 5-7-5 = 17 mora. Modern and colloquial haiku sometimes deliberately bend the form, using 字余り (extra mora) or 字足らず (short mora) as technique. This tool reports the over/under against the strict pattern, but for free-style writers — Santōka Taneda, Hōsai Ozaki, or contemporary colloquial tanka by Machi Tawara and Hiroshi Homura — that diff is best read as “distance from convention” rather than as an error.

School and modernity differences in haiku rules

Traditional haiku enforces the 5-7-5 mora pattern, a seasonal word (季語), and a cutting word (切れ字: や / かな / けり). Different haiku associations take different views: the Nihon Dentō Haiku Kyōkai (traditional) holds firmly to the pattern, while the Gendai Haiku Kyōkai (modern) tolerates extra mora and even drops the seasonal word. Tanka has the same split, with the Araragi school favouring realism and stricter form, and contemporary poets using colloquial Japanese that runs over or under 5-7-5-5-7-7 freely. This tool reports the count; the literary judgment of which deviation is intentional craft is up to you.

Translated haiku in English typically cannot match the mora count because syllables and mora are different units — English syllables can be much shorter, so 5-7-5 in English syllables produces a poem with more information density than the Japanese original, defeating the form’s signature brevity. Translators like Robert Hass, R. H. Blyth, and the editors of The Haiku Anthology (Cor van den Heuvel) generally drop the strict syllable count in favour of conveying meaning. Conversely, running the Japanese translation of a non-native poet’s haiku through this tool reveals where cultural differences in rhythm bend the form. The practical use of the counter is as a distance-from-convention gauge, useful precisely because Japanese mora rules are sufficiently rigid to be measured. For per-token POS and reading analysis of the same verse, wakati-tokenize pairs naturally with this tool.