Translation Anatomy · 2026-05-04 · Reading time ~ 10 min
Why Iekitsu Is Wrong — A Kabuki Reading Most AI Translators Miss
One Japanese character, two readings, and how Pagera's twin-Opus reviewer system caught what DeepL and Google Translate missed in Orikuchi Shinobu's kabuki memoir.
Pagera Editorial
The kanji 「家」 has two standard Japanese readings: ie and ka. Lookup any dictionary and you'll get those two. But when this character appears inside a kabuki actor's stage name, both readings are wrong — and DeepL, Google Translate, and Pagera's own first-pass translation all got it wrong in exactly the same way.
One character, one wrong reading
Orikuchi Shinobu (折口信夫, 1887–1953) was a Japanese folklorist, poet, and scholar of classical literature. His essay 「春永話」 (Haru Naga Banashi, 1935) recalls the Kyoto kabuki kaomise (顔見世) season — the annual year-end showcase — with the mix of scholarly precision and self-deprecating Osaka wit that defined his prose style. More than ten kabuki actor names appear in the text.
One of them is 市村家橘 (Ichimura Kakitsu).
The character 「家」 in that name looks like ie or ka. DeepL rendered it as Ichimura Iekitsu. Google Translate produced Ichimura Kakitsu — closer in phonology, but still the wrong romanization convention for the stage name system. Pagera's Pass 1 translation used Ichimura Iekitsu as well.
The correct reading is kakitsu (かきつ). This is the fixed reading documented in kabuki performance records, the National Theatre of Japan database, and specialized theatrical reference works. In the kabuki hereditary name system, 「家」 in this particular name is invariably read as ka. The general-purpose reading ie does not apply here.
A translation that renders the actor as Iekitsu produces a name that cannot be verified, looked up, or cross-referenced with any existing record. The actor effectively disappears.
What the Pagera review log shows
Pass 1 — specialist-reviewer flagged:
A05 −1 — Proper noun mismatch. 市村家橘 standard reading is かきつ (kakitsu). Kabuki reference required. Glossary violation.
A05 is the deduction code for proper noun inconsistency: one point off. That single deduction pushed the Pass 1 total below 98, triggering a FAIL and sending the translation to the targeted-fixer.
After fixer:
「…市村家橘(いちむら かきつ)が右手に…」 — the kanji retained alongside the corrected reading, so readers can verify the original form.
Pass 2 — both reviewers passed: specialist 99 / re-reviewer 98. Published.
Why this category of error exists
Korean and Japanese both draw on classical Chinese characters, but they developed independent reading systems over centuries. For most vocabulary, the mapping is reasonably predictable. For personal names — especially hereditary stage names — the predictability breaks down.
Kabuki families have used a fixed hereditary name system for hundreds of years. The reading of a given character inside a stage name is frozen by tradition, not derivable from standard on-readings or kun-readings. An AI translation model trained on general-purpose text has little reason to have encountered the specific reading kakitsu for 「家橘」 in sufficient density to override the default reading.
The result is a translation that looks fluent and passes surface-level review. The error is not a mistranslation of meaning — it's a misidentification of the person being named.
Two more traps found in the same text
「我当」 (Gatō) and 「我童」 (Gadō) — separating brothers
In Korean Han character pronunciation, 「我当」 reads as a-dang and 「我童」 as a-dong. In the kabuki name system, these are two distinct stage names: Gatō and Gadō. In Orikuchi's memoir, the two men are brothers — 「我当」 is the younger name of the actor who became the 11th Kataoka Nizaemon, and 「我童」 is his elder brother's name.
Confusing the readings collapses the familial relationship Orikuchi is recalling. The Pagera specialist-reviewer verified both names against kabuki records and standardized the text to Gatō's elder brother Gadō, making the identification explicit.
「弄花」 — a gambling term hidden in plain sight
The characters 「弄花」 translate literally as "playing with flowers." In Meiji- and Taisho-era Japanese, however, this was slang for hanafuda gambling — 「花 (hana)」 being the hanafuda card game, 「弄 (rō)」 meaning to play or handle. The social context Orikuchi gestures at — the gambling culture around kabuki districts — vanishes if this is rendered as a decorative flower metaphor.
Pagera's translation annotated the term as 「弄花 (hanafuda gambling)」, preserving the period-specific connotation.
How the system catches this
Pagera uses two independent Opus-model reviewers on every translation. The specialist-reviewer does a full line-by-line comparison against the source. The re-reviewer scores blind — no access to the previous score, deduction list, or fixer history — from a reader's perspective. Both must pass for publication.
In this translation, the specialist-reviewer caught 「家橘.」 The 「我当/我童」 separation and the 「弄花」 annotation were flagged by the two reviewers at different positions independently. That distribution of catches is exactly why two independent reviewers matter: one reviewer's blind spot is not the system's blind spot.
The 13 kabuki actor names resolved in this translation — kakitsu, gatō, gadō, nizaemon, uzaemon, tokizō, karouku and others — are logged into a per-author glossary. The next Orikuchi Shinobu kabuki essay goes in with those readings pre-loaded. The same trap does not repeat.
Orikuchi Shinobu left more than 100 untranslated works. A significant portion involve kabuki and traditional performing arts. One glossary, reused across all of them.
Next in the series
The next Translation Anatomy post examines 「師匠」 rendered as sajang — the Korean homophone for "company president." A kabuki woodcarver's memoir about his master becomes a text about a CEO. This happened in Takamura Kōun's Bakumatsu Ishin Kaikōdan (Recollections of the Meiji Restoration) and was caught before publication.
Browse 71 Korean translations in the Pagera catalog — each passed both Opus reviewers