Bashō's frog has been carried into English more than a hundred times. Every version says something Bashō didn't — not from carelessness, but because English has no grammar for what the poem leaves open.
古 furu old |
池 ike pond |
や ya cutting word |
蛙 kawazu frog |
飛 tobi fly / leap |
込 -komu go into |
水 mizu water |
の no 's / of |
音 oto sound |
That is nearly the whole of it. Seventeen morae — not syllables; we will come to the difference — in three measures of five, seven, five. An old pond. A cutting word. A frog, a leap into the water, and the sound that makes. There is no scene more ordinary, and few that have been fought over harder.
The fight is not about the words. 古 is old; 池 is pond; nobody disputes the dictionary. The fight is about everything Japanese declines to specify and English cannot withhold. To put this poem into English is to be forced to decide a series of things Bashō left open — and every decision adds information that was not in the original. The translator does not lose meaning. He gains it, against his will.
Japanese, here, withholds four things English insists on. Each is a slot the grammar of the target language will not let you leave empty:
None of these is a failure of any one translator. They are structural. The two languages share no slot for these things, the way the side and the diagonal of a square share no common measure — and as with that older incommensurability, the gap is not bridged by being more careful. It is bridged by choosing, and the choice is the loss.
Below is the poem as a set of dials. Each one is a thing Japanese left open and English will not. Leave them all where they start — unset — and you have the original's silence, and no English at all. To produce a single line of English you must turn every dial; the meter at the top counts how many decisions you have been forced to make that Bashō never made.
The point of the instrument is the point you cannot reach: the one with every dial unset. English has no word that means a-frog-or-frogs, no tense between jumps and jumped, no mark that is the cutting-word and not a dash. The original lives at that point. Every translation is a different way of leaving it.
Lined up, the canonical English versions are not better and worse attempts at one fixed target. They are different settlements of the same four open questions — plus a fifth the dials can't hold, which is how literal to be about 水の音, "the sound of water." Some keep the plain genitive; some name the splash that Bashō only implies.
| Translator | Number | Tense | The cut (や) | The sound |
|---|
Read down the columns and the incommensurability is plain. There is no agreement, because there is nothing to agree on: the original does not answer these questions, so every column is a record of guesses, not of fidelity. The cell that says the most is the last one. Where Bashō wrote 水の音 — flatly, water's sound — Watts wrote Plop! and Ginsberg Kerplunk!. They are not wrong. They are translating the sound itself instead of the words for it — abandoning the lexicon to keep the event. That is a defensible trade, and it is a trade; the dials cannot show it, because it leaves the grid entirely.
The original does not answer these questions. So every English version is a record of guesses, not of fidelity — and lined up, the guesses are the poem's true commentary on itself.
Almost everyone is taught that a haiku is five-seven-five syllables. It is not. The count is in morae — in Japanese, 音 (on) — and a mora is not a syllable. A syllable is one beat of the mouth; a mora is one beat of timing, and a single syllable can hold two of them. The poem scans:
fu · ru · i · ke · ya (5)
ka · wa · zu · to · bi · ko · mu (7)
mi · zu · no · o · to (5)
The split matters because the mora carries weights a syllable-count throws away: a long vowel is two morae, and a final ん (-n) is a mora all its own. A word English would call one syllable can fill two of haiku's seventeen beats. So an English "5-7-5" — which counts syllables — is already in a different metric from the original it claims to honor. The form, too, is incommensurable: you cannot keep the count, because you cannot keep the unit. Most of the translators above quietly gave up 5-7-5 and were right to. It was never the same seventeen.
The dictionary survives translation. The grammar does not. What this poem is made of — number left open, time left uncut, a particle that stops the line without saying anything — are exactly the features that live in a language's grammar rather than its words, and grammar is what does not cross. You can carry old and pond and frog into any language on earth. You cannot carry the not-deciding, because the moment you speak English you have decided.
And there is a loss deeper than grammar, one no dial can show. 蛙 is read kawazu — not the everyday kaeru but the poetic-register word, and in the classical waka tradition the kawazu was the frog that sings: for centuries the convention was the naku kawazu, the croaking frog, prized for its voice. Bashō's quiet violence — the whole turn of the poem, what Haruo Shirane calls its haikai twist on the inherited tradition — is that he gives the sound not to the frog's voice but to the water it disturbs. He takes the most codified poetic sound in the canon and replaces it with a plain splash. An English reader cannot hear this, because English has no kawazu carrying nine hundred years of singing behind it. The word is translatable; the allusion the word is arguing with is not.
This is the same shape as the first stratum laid in this ground. Incommensurable showed that the diagonal of a square has no common measure with its side: no fraction, however fine, lands on √2. Here the side is one grammar and the diagonal is another, and the thing with no common measure is a frog that is neither one nor many, leaping in a tense that is neither now nor then. The proof there was that the gap can be named exactly — odd against even, forever. The proof here is softer but the same in spirit: you can name exactly what is lost, dial by dial, and naming it is the closest thing to keeping it.
So the hundred frogs are not a hundred failures. They are a hundred soundings of a single point no English can occupy — and read together, lined up against the open original, they are the most honest commentary the poem has. The translations cannot reach Bashō. But the shape of how they all miss is exactly the shape of the pond.
In the spirit of this ground, the places this piece strains or guesses, set down plainly rather than smoothed over:
On the readings. 蛙 is read kawazu here — the classical, poetic-register word for frog (it carries the waka singing-frog convention discussed above), as against the everyday modern kaeru. The gloss treats 飛び込む as one verb, tobikomu ("to jump in"), which it is; I split it into 飛 (tobi, fly/leap) and 込 (-komu, to go into) only to show the compound, not to claim two words. That や is a kireji, that Japanese marks neither number nor article here, and that the verb is non-past — these are uncontroversial facts of the grammar (set out in Henderson's Introduction to Haiku), not interpretive claims. The poem was composed in the spring of 1686 and published that year in the Kawazu awase ("Frog Contest") anthology of Bashō's circle.
On the translations. Every quoted version is given verbatim, with its translator named. Bashō's frog has been mistranslated less often than it has been mis-attributed: plain, article-free renderings circulate online under a dozen different names, and the bare "Old pond / a frog jumps in / the sound of water" is best treated as a literal crib with no single owner — I have labeled it so rather than hang it on a translator it may not belong to. Where I was not confident of an attribution, I left the version out. The fullest scholarly gathering is Hiroaki Sato's One Hundred Frogs (1983), which collects well over a hundred English versions; this piece samples it, and is not a substitute for it.
On the dials. The instrument is a schematic, not the translations themselves. Snapping a translator's "choices" onto five dropdowns flattens a living line into a grid — the quoted text beside each is the truth, the dial positions are my reading of it. The "decisions forced: 5" is a count of independent open questions, not a measurement; I resisted dressing it up as Shannon bits, because the options are not equally likely and that number would be a costume, not a fact.
On the type. The Japanese above is set in whatever mincho serif your system provides: this site self-hosts only Latin fonts and makes no third-party requests, so it will not fetch a Japanese webface to match. If the characters look plainer than the English around them, that is the cost of the promise, kept.