End of Japanese community

Original Article ↗ Hacker News Discussion ↗

What Happened

  • Mozilla enabled an AI “SumoBot” / MT workflow on Japanese support KB articles.
  • The bot auto-generated translations, auto‑approved them after ~72 hours, and appears to have overwritten or sidelined ~20 years of volunteer human translations, in production rather than staging.
  • The long‑time Japanese locale leader publicly resigned, listing issues: violation of translation guidelines, ignoring existing Japanese localization choices, loss of control, lack of prior consultation, and objections to using their work as AI training data.

Reaction to Mozilla’s Reply

  • The official response – “sorry for how you feel” + “hop on a call so we understand what you’re struggling with” – is widely read as:
    • A classic non‑apology that frames the problem as volunteers’ feelings rather than Mozilla’s actions.
    • Tone‑deaf and corporate (“hop on a call” for someone whose work you just trashed).
    • An attempt to move the issue off‑record into a private channel rather than address it publicly and concretely.
  • A minority argue it’s a reasonable first response by a community manager with limited power, trying to gather details across language and time zones.

Machine Translation, Language, and Culture

  • Many multilingual commenters say MT into Japanese is particularly bad: unnatural, wrong contexts, inconsistent terminology, and no awareness of local style guides.
  • Even for European languages, several say “good enough” MT still produces atrocious UI text and documentation.
  • Broader concern: LLM/MT flattens cultural nuance into an en‑US‑flavored generic style; people increasingly encounter auto‑translated threads and videos that “feel off.”

Process, Power, and Volunteers

  • Core anger is about process and respect, not just quality:
    • No nemawashi / consensus building with locale teams; change was imposed top‑down.
    • Bot can overrule volunteers instead of serving them as an optional tool.
    • Doing this to unpaid contributors is seen as especially insulting; many say quitting is entirely justified.
  • Some propose a better model: MT only as opt‑in suggestions, or for untranslated pages, with humans in control.

Licensing and AI Training

  • Dispute over Creative Commons:
    • Some argue the contributor can’t now “prohibit” AI training on CC‑licensed text and that CC is irrevocable.
    • Others point to moral rights and unsettled law around whether models are derivative works, and argue legalities aside, Mozilla has clearly broken trust.

Bigger Pattern Noted

  • Many tie this to a larger pattern: Mozilla chasing AI buzz, rolling out half‑baked features on live users, poor internal governance, and historic disregard for community feedback while marketing itself as community‑driven and “not like big tech.”