Translation Test: Bashō — The Frog Haiku

Source: Matsuo Bashō (松尾芭蕉), 1686

Original: Japanese (classical; kanji + hiragana)

Status: Draft — first pass retained; second pass after EMD-002 added

Purpose

The frog haiku is the kireji / pure juxtaposition stress test. It is the most translated poem in East Asian literature and has no consensus English version after 340 years, for a specific linguistic reason: the kireji や (ya) is structurally untranslatable into any European language. Tonesu has two candidate tools — / and : — and this batch determines which one applies, with consequences for the interpretation of the entire poem.

Update note (May 2026): BSH-001 below preserves the original first-pass analysis written before — was admitted in EMD-002. BSH-002, appended at the end of this file, revisits the same haiku after prosodic suspension became available as a third and potentially better kireji candidate.

Primary tests: - Kireji as structural mark — does / (parallel partition) or : (topic frame) correctly encode the ya cut? The two marks encode different readings; choosing between them forces a decision that Japanese leaves open for 340 years - Agent-drop — the frog is grammatically present in line 2 (kawazu tobikomu — frog jumps-in); the perceiver of the sound in line 3 is absent; Tonesu has agent marking for the action but must choose how to handle the soundless final clause - Pure noun-phrase sentence — line 3 水の音 (mizu no oto = water's sound) is a bare noun phrase, no verb, no predicate; Tonesu requires a predicate form - Sound as event — so (acoustic signal) must serve as both a thing and an event; minimal predication needed - Frog vocabulary gap — no registered frog compound; zo-se-ma (aquatic-vertebrate class, kinds.md) used by contextual inference; GAP-BSH-001

Corpus sentences from this batch: S468–S470.

Source Text

Japanese (original):

古池や
蛙飛び込む
水の音

Romanization:

furuike ya
kawazu tobikomu
mizu no oto

Morpheme gloss:

furu-ike  ya      — old-pond    [kireji: cut / pause / juxtaposition marker]
kawazu    tobi-komu   — frog     leap-enter
mizu  no  oto         — water GEN sound

Literal (syntactically faithful but not poetic):

An old pond — / A frog jumps in / The sound of water

The contested problem: every English translator must decide what the dash, comma, line break, or silence is doing between "old pond" and "frog jumps in." Bashō left that structural relationship undefined by design. The poem's resonance lives in that gap.

Vocabulary Framework

Form	Reading	Construction	Notes
`re-fe-ma-ki`	old pond	`re` (recurrence/persistence) + `fe` (boundary) + `ma-ki` (water, MAT-002) = long-established bounded water = old pond	Compositional; `re` encodes long persistence through cycles; `fe-ma-ki` = bounded water = pond; head-final compound
`ma-ki`	water	`ma` (matter) + `ki` (motion) = flowing matter = water	Established MAT-002, S300–S306
`zo-se-ma`	aquatic vertebrate / frog (by context)	kinds.md node: matter-perceptual organism; the branch covering fish and aquatic vertebrates	No specific frog discriminator registered — GAP-BSH-001; context (pond + jumping) disambiguates to frog-class
`ma-ki-so`	sound of water / water-sound	`ma-ki` (water) + `so` (sound) = water-sound, head `so` modified by `ma-ki` = the sound that belongs to water	Compositional; `so` is the head (the acoustic signal); `ma-ki` modifies it as the source-domain
`pa`	exists / occurs	primitive R021; existential predicate — used here as minimal predicate for the pure-noun-phrase final line	Provides required predicate without adding semantic weight

`re-fe-ma-ki`: building "old pond"

The compound builds head-finally: - ma-ki = water (established) - fe-ma-ki = boundary-water = bounded body of water = pond (fe = boundary/limit; pond is water with a limiting boundary) - re-fe-ma-ki = recurring-bounded-water = a pond that has persisted through many cycles = old pond

re (recurrence/persistence/cycle) encodes oldness as temporal persistence — the pond that has been there through many seasons. This is not merely "a pond modified by time" but "a pond characterised by its having-been-there-across-cycles" — which is precisely the existential weight Bashō gives to furuike. An old pond in a haiku is not merely aged; it is a marker of established worldly continuity that the moment of the frog's entry briefly punctuates.

`ma-ki-so`: "sound of water"

Head-final direction: so = sound (primitive, the acoustic signal); ma-ki = water. ma-ki-so = water's sound, where the head is the acoustic event and water is its source modifier. Contrast so-ma-ki (which would have ma-ki as head = auditory-water, wrong direction). The form ma-ki-so is right: it is a sound, of water.

This is compositional; no W-number required. First attested S468.

The kireji `や` — `/` or `:` ?

The kireji ya has been translated into English as: - A dash: "old pond — a frog jumps in" (most common) - Nothing: just a line break (weakest) - "and": "old pond and a frog jumps" (adds causality that isn't there) - "where": "old pond, where a frog jumps" (makes it a relative clause — wrong) - Ellipsis: "old pond..." (trailing off — wrong direction; the poem moves forward)

No English solution is correct because English has no structural mark that means: I am presenting these two images as formally paired without specifying their relationship; hold both in mind simultaneously and let the resonance arise.

Tonesu has two candidates:

Mark	Function	What it does with the haiku
`/`	parallel partition	Signals co-equal structural pairing between two clauses; does not specify the relationship type; content supplies it
`:`	topic frame	Establishes the left-side noun phrase as the thematic anchor of the following clause

These are not the same reading of the haiku.

The / reading says: two co-equal images held in juxtaposition — the pond is not more than the event; both are equal partners in the resonance. This is the Bashō-as-Zen reading: the poem presents two equally weighted moments of consciousness.

The : reading says: the pond is the thematic anchor; the frog and sound are what happen against it — foreground/background. This is the Bashō-as-painter reading: the pond is the canvas; the frog is the mark on it.

The finding: Tonesu forces this choice, exposing the interpretive fork that Japanese ya keeps open. Both S468 (using /) and S469 (using :) are valid Tonesu renderings — they encode different critical readings of the same poem. The language is more precise than the source on this axis. Whether that precision is a gain or a loss depends on whether you think a poem should commit to its interpretation or remain open.

Agent-drop and the final clause

In Japanese line 3, there is no verb and no subject. 水の音 is three particles: water + genitive + sound. Tonesu requires a predicate. Options:

ma-ki-so pa = water-sound exists/occurs — minimal; pa is the existential predicate; says only that the sound is present
se lo-ma-ki-so = perceiving water-sound — names an act of perception, but invents a perceiver (even an implicit one)
ma-ki-so alone — ungrammatical in Tonesu (no predicate); but if haiku as a genre were established as a no-predicate form, the genre itself supplies the implied pa

S468 and S469 use ma-ki-so pa. The existential pa is the lightest predicate available — it says "this sound exists" without attributing it to a perceiver or a cause. This preserves the impersonal quality of the source.

Gap BSH-002: Tonesu cannot match Japanese's zero-predicate noun-sentence for this line. The existential pa is the minimum; it slightly increases the semantic content of "water's sound" by asserting existence rather than simply naming. This is a real translation loss.

Verse-by-Verse Analysis

S468 — Full haiku — kireji as `/` parallel partition (BSH-001-A)

re-fe-ma-ki  /  la-zo-se-ma  ki  lo-re-fe-ma-ki,  ma-ki-so  pa

Written: refemaki / lazosema ki lorefemaki, makiso pa

Parse: - re-fe-ma-ki = old pond — the left image - / = kireji: parallel partition; co-equal structural pairing - la-zo-se-ma = the aquatic-vertebrate / frog (agent) - ki = moves / enters (physical motion, ki = displacement) - lo-re-fe-ma-ki = patient: the old pond (into which the frog moves) - , = prosodic pause separating the action from the result - ma-ki-so pa = water-sound exists

Reading: The pond and the event are co-equal partners in the poem. The structural pairing (/) presents them without subordinating either. The relation between them — stillness punctuated by motion, age punctuated by instantaneous presence — is left for the reader to supply. The poem is a juxtaposition, not a sentence with a subject and a predicate.

Note on pond repetition: re-fe-ma-ki appears both in line 1 (as left image of the / pairing) and as lo-re-fe-ma-ki in line 2 (patient of the frog's motion). Japanese avoids this: furuike in line 1, then tobikomu in line 2 (the pond is implied — the frog jumps in; no explicit second mention). Tonesu has no discourse-anaphora for established referents — the full compound must repeat. This is GAP-BSH-003: no pro-form for recently-established discourse referents. The repeated compound is slightly more redundant than the source.

S469 — Full haiku — kireji as `:` topic frame (BSH-001-B)

re-fe-ma-ki  :  la-zo-se-ma  ki  lo-re-fe-ma-ki,  ma-ki-so  pa

Written: refemaki : lazosema ki lorefemaki, makiso pa

Identical except / is replaced by :. The topic frame makes the old pond the thematic anchor: "as for the old pond — the frog enters, water-sound occurs." The pond is no longer a co-equal image; it is the stage setting. The event is a comment about the pond: this is what happens here, in this enduring place.

This reading is more temporal: the old pond persists; the frog's leap is a momentary event against that persistence. The deep/present contrast is foregrounded. Many Zen readings of the haiku favour this structure — the sound is not just a sound; it is what the pond receives in this moment.

Structural note: : requires sentence-initial position. re-fe-ma-ki : is sentence-initial, so this is valid. No parse conflict.

S470 — Compressed form — agent suppressed (BSH-001-C)

re-fe-ma-ki  :  ki  lo-ma-ki,  ma-ki-so  pa

Written: refemaki : ki lomaki, makiso pa

The most minimal rendering. Agent (la-zo-se-ma) is dropped — the frog disappears as a named entity; only the motion event remains. lo-ma-ki substitutes for lo-re-fe-ma-ki — the patient is simply "the water" (not the full old-pond compound), dropping the re (age) and fe (boundary) modifiers. The final line is unchanged.

What this achieves: The haiku's spirit without the frog's identity. The motion event ki lo-ma-ki = motion into water is pure event, not animal-event. This is closest to Bashō's most radical reading — some commentators say the frog is almost irrelevant; what matters is the sound that follows. The agent-dropped form makes the frog's absence structural.

What it loses: The frog is actually in the poem — kawazu is not grammatically absent in Japanese; the agent who performs tobikomu is explicitly named. S470 removes that. This is a compression beyond what the source licenses. S468 or S469 are the primary translations; S470 is an exploratory variant.

Grammatical note: ki lo-ma-ki without la-{agent} — is this licensed? GAP-BSH-004 — Resolved. Grammar §Ellipsis Pattern 3 (context drop) licenses omission of any argument when fully recoverable from discourse context. Within the BSH-001 batch, zo-se-ma (the frog) is explicitly established as agent in S468 and S469; agent drop in S470 is discourse-licensed. In a standalone reading, the agent is scene-recoverable (pond + motion + sound evoking the frog). S470 remains an exploratory compression beyond what the source licenses (Japanese names the frog explicitly); the grammar is not violated, but the translator's choice to omit it goes further than the original.

BSH-001 Translation Comparison

S	Kireji	Agent	Old pond	Sound
S468	`/` (co-equal)	`la-zo-se-ma` explicit	`re-fe-ma-ki` × 2	`ma-ki-so pa`
S469	`:` (topic anchor)	`la-zo-se-ma` explicit	`re-fe-ma-ki` × 2	`ma-ki-so pa`
S470	`:` (topic anchor)	dropped	`re-fe-ma-ki` + `lo-ma-ki`	`ma-ki-so pa`

BSH-001 Batch Summary

Entry	Form	Test
S468 (BSH-001-A)	`refemaki / lazosema ki lorefemaki, makiso pa`	kireji as `/`; co-equal juxtaposition; Bashō-as-Zen reading
S469 (BSH-001-B)	`refemaki : lazosema ki lorefemaki, makiso pa`	kireji as `:`; topic anchor; Bashō-as-painter reading
S470 (BSH-001-C)	`refemaki : ki lomaki, makiso pa`	agent-dropped maximum compression; event-without-actor

Key finding: Tonesu exposes the 340-year critical debate about ya as a structural choice between two available marks. The / reading (co-equal juxtaposition) and the : reading (topic anchor / foreground-background) encode genuinely different interpretive stances on the poem. Japanese ya refuses to make this choice; Tonesu cannot avoid it. This is the clearest case yet where Tonesu's greater structural precision is simultaneously a gain (it clarifies what the interpretive options actually are) and a loss (it commits the translator where the poet chose not to).

The language comparison: Japanese's richest poetic tool is the suppression of grammatical structure — kireji, zero-particles, unpredicated noun-phrases — creating resonance through what is not said. Tonesu's design philosophy is the explicit articulation of structure. The haiku tests whether an analytically-explicit language can express a poem built entirely on structural withholding. The answer: partly. S468 and S469 are good translations; neither is a haiku.

New vocabulary introduced: - re-fe-ma-ki (old pond; compound; first attested S468) - ma-ki-so (water-sound; compound; first attested S468)

Open questions logged: - GAP-BSH-001: no registered frog compound; zo-se-ma (aquatic-vertebrate) used by contextual disambiguation; a proper discriminator sub-node should be established in kinds.md when frog-reference recurs - GAP-BSH-002: Tonesu cannot produce a zero-predicate noun-sentence; pa (existential) is the minimum addition; this is a structurally unavoidable translation loss - GAP-BSH-003: no discourse-anaphora pronoun for recently-established referents; re-fe-ma-ki must be spelled out twice; this inflates compound length - GAP-BSH-004: ~~agent-drop for pure motion-event clauses — is ki lo-X without la-{agent} grammatical? S470 tests the limit; not yet resolved~~ Resolved (March 2026): Grammar §Ellipsis Pattern 3 licenses agent drop when fully context-recoverable. In the batch zo-se-ma is established in S468/S469; S470's drop is discourse-licensed. The compression beyond source is a translator's choice, not a grammar violation.

BSH-002 — Second Pass After Prosodic Suspension [S1217–S1219]

Purpose: Revisit the frog haiku after EMD-002 admitted — as a prosodic suspension mark. BSH-001 showed that forcing a choice between / and : exposed the poem's interpretive fork, but it also over-committed the translation. This second pass asks whether — can preserve more of Bashō's withheld relation by suspending the pond-image without deciding whether what follows is co-equal juxtaposition, topic-comment, or something looser than either.

Primary tests: - whether — is a better kireji candidate than / or : for the primary rendering - whether the poem benefits from suspension rather than explicit structural commitment - whether second mention of the pond can be reduced without losing the event too badly - whether the agent-dropped compression becomes more haiku-like once the opening cut is handled by —

Secondary tests: - whether BSH-001 should now be read as interpretive expansions rather than primary renderings - whether fe-ma-ki is a useful partial-repeat compromise between full repetition and over-compression

Corpus sentences from this batch: S1217–S1219.

BSH-002 Vocabulary Framework

Form	Reading	Notes
`re-fe-ma-ki`	old pond	established from BSH-001
`fe-ma-ki`	pond	compositional first attested here as reduced second mention of the pond
`zo-se-ma`	aquatic vertebrate / frog (by context)	same contextual frog solution as BSH-001
`ma-ki-so`	water-sound	established from BSH-001
`—`	prosodic suspension	EMD-002 / G028 reuse

BSH-002 Sentence Analyses

S1217 — Full haiku — kireji as `—` prosodic suspension (BSH-002-A)

re-fe-ma-ki  —  la-zo-se-ma  ki  lo-re-fe-ma-ki,  ma-ki-so  pa

Written: refemaki — lazosema ki lorefemaki, makiso pa

Reading: Old pond — a frog enters the old pond, water-sound exists.

Notes: This is now the strongest primary rendering. — suspends the old pond without deciding whether the next line is a co-equal partner (/) or a comment anchored to it (:). That gets closer to Bashō's deliberate under-specification. The cost is that the full pond compound still repeats, so the line remains heavier than the Japanese.

S1218 — Full haiku — suspension with reduced second pond mention (BSH-002-B)

re-fe-ma-ki  —  la-zo-se-ma  ki  lo-fe-ma-ki,  ma-ki-so  pa

Written: refemaki — lazosema ki lofemaki, makiso pa

Reading: Old pond — a frog enters the pond, water-sound exists.

Notes: This is the best economy compromise. The poem names the old pond once, then reduces the second mention to fe-ma-ki (pond) rather than repeating the full aged-boundary image. It is still more explicit than Japanese, but it avoids making the source sound as though it repeats itself more than it does.

S1219 — Suspension with agent suppressed (BSH-002-C)

re-fe-ma-ki  —  ki  lo-fe-ma-ki,  ma-ki-so  pa

Written: refemaki — ki lofemaki, makiso pa

Reading: Old pond — motion into the pond, water-sound exists.

Notes: This remains exploratory rather than primary. Once — carries the opening cut, the agent-dropped version sounds more poem-like than S470 did, but it still compresses beyond the source by removing the frog altogether.

BSH-002 Comparison

S	Kireji	Agent	Pond repetition	Why keep it
S1217	`—`	explicit	full `re-fe-ma-ki` repeated	strongest structural fidelity to Bashō's withheld relation
S1218	`—`	explicit	reduced to `fe-ma-ki` on second mention	best balance of suspension and economy
S1219	`—`	dropped	reduced `fe-ma-ki`	most haiku-like compression, but furthest from source grammar

BSH-002 Batch Summary

Entry	Form	Test
S1217 (BSH-002-A)	`refemaki — lazosema ki lorefemaki, makiso pa`	`—` as primary kireji candidate with full explicit frog event
S1218 (BSH-002-B)	`refemaki — lazosema ki lofemaki, makiso pa`	reduced second pond mention under suspension
S1219 (BSH-002-C)	`refemaki — ki lofemaki, makiso pa`	maximum compression under the new suspension mark

Updated finding: After EMD-002, — is the best primary Tonesu rendering of ya. / and : still matter, but they now look less like candidate translations and more like deliberate interpretive glosses of what Bashō leaves structurally open. — does not solve the zero-predicate or repetition problems, but it does remove the largest avoidable over-commitment in BSH-001.

What improved: Tonesu no longer has to decide the haiku's deepest relation at the moment of the cut. That is a real gain. The poem remains heavier than Japanese because ma-ki-so pa still needs predication and the pond still needs some repeated mention, but the opening cut is now much closer to the source's suspended resonance.

Still unresolved: - GAP-BSH-001 remains: no dedicated frog discriminator - GAP-BSH-002 remains: no zero-predicate noun-line equivalent - GAP-BSH-003 remains partially relieved but not solved: fe-ma-ki reduces repetition, but there is still no true discourse pro-form for "that already-established pond"