[i18n] "recalled" mistranslated as "revoked/withdrawn" in ja-JP, de-DE, ko-KR

Status Closed — not planned
Maintainer reply None cached
Activity 1 comment · opened Aug 31, 2026 · closed Sep 3, 2026

Summary

The memory-chip label recalled (i18n key FyGldNBVoJ) is translated with the "revoke / withdraw / recall a product" sense instead of the "recall a memory" sense in three locales. In Japanese the result reads as the opposite of what happened: the chip says 取り消しました N件のメモリ (= "revoked N memories"), so users reasonably conclude their memory writes were rolled back. They were not — the files on disk were correct.

Where

/usr/lib/claude-desktop/resources/ion-dist/i18n/<locale>.json, key FyGldNBVoJ
(claude-desktop 1.40609.0, Linux .deb)

| locale | current | back-translation | ok? |
|---|---|---|---|
| en-US | recalled | — | — |
| ja-JP | 取り消しました | "revoked / cancelled" | :x: |
| de-DE | widerrufen | "revoked" | :x: |
| ko-KR | 회수됨 | "collected back / recalled (a product)" | :x: |
| es-ES / es-419 | recordado | "remembered" | :white_check_mark: |
| pt-BR | lembrado | "remembered" | :white_check_mark: |
| id-ID | diingat | "remembered" | :white_check_mark: |
| hi-IN | याद किया गया | "remembered" | :white_check_mark: |
| fr-FR | Rappelé | "recalled to mind" | :white_check_mark: |
| it-IT | richiamato | "called back" | ambiguous |

Every other 取り消し string in ja-JP.json maps to Revoke / Withdraw / Unpublish and is translated correctly. This key is the only one that leaked into that group, which is what makes it read as a real product behaviour rather than a wording quirk.

Steps to reproduce

  1. Set the app language to Japanese
  2. In a session, have Claude read existing memory files and then update them
  3. The chip above the response reads 取り消しました 2件のメモリ

Underlying behaviour is correct: the memories were read, and the files were updated on disk (verified by mtime and content). Only the label is wrong.

Impact

Low severity, but high confusion: the message tells the user in three languages that a write was undone when it succeeded. My first reaction was to go verify the files by hand, assuming data loss.

Suggested fix

  • ja-JP: 想起しました (as a chip label, 参照したメモリ also works)
  • de-DE: abgerufen
  • ko-KR: 불러옴
  • it-IT: recuperato, if the "retrieved from memory" sense is intended

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗