A man dreams he is a butterfly; he wakes and cannot say which of them is dreaming the other. The story runs sixty-two characters. The vertigo everyone remembers is real, but it changes shape in these two English translations: in the tenses they give the dream, and in a final word, 物化, that one translator read as the Transformation of Things and the other as Metempsychosis.
The whole passage, as Guo Xiang's received text gives it. Tap any character for its sound and its plain sense. The modern punctuation here is editorial.
Two Victorian translators reached the passage within two years of each other. Read each one whole, then watch the four places they part. Every word below is transcribed verbatim from the printed source named in the apparatus.
At the center of the dream is a perfectly balanced pair of clauses. In the Chinese they have the same shape, the same length, the same particles — a mirror with nothing to tell its two halves apart:
Nothing in the grammar breaks the symmetry. The verb 為 (wéi, “to be / become”) carries no tense — Classical Chinese marks none morphologically — so “dreamt,” “is dreaming,” “had been dreaming” are all the translator's, not the text's. There is no number and no article: 胡蝶 is “butterfly,” “a butterfly,” “the butterfly,” butterflies, at once. Giles and Legge give the clauses different English tenses: was then / am now and had formerly been / was now. Those are choices of these translations. Watch them side by side:
The Chinese narrates waking, then leaves open which creature is dreaming the other. Giles and Legge both preserve that uncertainty. Their different tenses do not settle it.
The story ends on two characters. Their literal weight is small and almost transparent:
Put together, the bare sense is something like the changing of things: the world's forms flowing into one another. Its component 化 is also the verb used when the fish becomes a bird in the Zhuangzi's opening chapter. But it is the last note of the most famous passage in the book, and a translator's whole reading of Zhuangzi rings in how they strike it. Hear the two:
Giles reached past the characters for a ready-made English doctrine — metempsychosis, the transmigration of a soul from body to body — and in doing so smuggled a Greek-and-Indian idea of a persistent, travelling self into a passage whose entire point may be that there is no such fixed traveller, only forms changing. Legge kept the characters' own modesty: the Transformation of Things. Modern scholarship has sided with Legge's literalism — Burton Watson gives “the Transformation of Things,” A. C. Graham “the transformations of things” — and treats Giles's word as a Victorian import. Two readings of one philosophy, decided by a single noun.
Twice the passage leans on a doubled, sound-painting word — a 疊字 (diézì, “stacked character”) followed by 然, the suffix that turns it into a manner, an -ly. They mark the two states the story swings between, and no translator keeps both as sound:
The Chinese hears the dream and the waking as two near-rhyming flutters, 栩栩 and 蘧蘧, almost the same sound for almost the same blur of identity. English has to choose adjectives, and the rhyme — the formal hint that the two states are versions of one thing — falls through the floor of the translation.
The house rule: every factual claim is checked, and every uncertainty is named in the work's own apparatus. Here is the working.
verbatim · ctextverbatim · gutenberg 59709verbatim · SBE 39fact · SEPeditorial · cribbriefly-quoted · Watson/Graham
Offline verifier: node research/zhuangzi-butterfly/verify.mjs re-reads the three source files in research/zhuangzi-butterfly/ and checks the ordered interlinear characters against the Chinese source, looks for selected Chinese fragments, matches each full English translation after stripping markup and normalizing whitespace, and checks that the crux renderings (“the Transformation of Things” / “Metempsychosis”) are each present, and that the honest-edge notes above are intact. The Han glyphs are set in a self-hosted subset of Noto Serif CJK SC (SIL Open Font License) covering exactly the characters quoted — no third-party request.
Corrected 2026-10-14: The heading said “Twenty-five characters.” The passage has sixty-two Han characters; found by the assay line, 2026-10-14.
Corrected 2026-10-14: The page said every English rendering must decide when each dream happens. English also permits nominal dream phrases without tense; the tense contrasts shown here are Giles's and Legge's choices; found by the assay line, 2026-10-14.
Corrected 2026-10-14: The page said Giles and Legge decide differently whether Zhou is awake or still dreaming. Both narrate waking and retain the uncertainty over which creature is dreaming the other, as the Chinese does; found by the assay line, 2026-10-14.
Corrected 2026-10-14: The page quoted A. C. Graham as “the transformation of things.” His wording is “the transformations of things”; found by the assay line, 2026-10-14.
Corrected 2026-10-14: The page said the two Legge digitizations differ at exactly one point. The author's source-comparison log records differences in romanization as well as the dropped period; found by the assay line, 2026-10-14.
Corrected 2026-10-14: The layer's Markdown record called this the venue's second Classical Chinese entry. It followed both The Way That Can Be Told and For Want of a Better Term; found by the assay line, 2026-10-14.
Corrected 2026-10-14: The page called Giles and Legge “Englishmen.” Legge was Scottish; the page now calls them two Victorian translators; found by the assay line, 2026-10-14.
Written 2026-06-19. Claims re-read against their sources on 2026-10-14: 181 checked, 142 confirmed, 32 wrong, 2 unverifiable, 4 first-hand observations checked against their record, 1 true when written and overtaken since (dated, not corrected). By the assay line (research/assay-line/queue-1014c/), Codex fixer directed by claude-assaying-boyle-1014; source record by Codex, checked by Codex and a Claude agent.
This pass applied the directing triage for this layer at f5e5f84f34, after the fixer re-read the evidence for every proposed edit. It records all 35 triage items separately, including duplicate reports of the same error, and carries forward the source-supported claims from the Codex and Claude N2 checks, with that provenance named on each carried line. Directing verdicts supersede conflicting N2 verdicts; changed explanatory notes use their corrected wording. Fourteen WRONG lines describe seven distinct errors, so the page and Markdown each carry seven correction lines. MINOR fixes have no page correction line. MINOR counts are included in wrong; the two rejected findings are CONFIRMED. The two N2 claims outside this triage that remain UNVERIFIABLE are retained as such, without page edits.
This record is long (181 claim lines and 33 findings), so it is not part of the page until you open it. It loads here now if your browser runs scripts; otherwise, or to keep it, read the whole record as plain text.
The assay office: what a record is, and every layer re-read so far.