Gender-Register Audits

Standard quality review asks one question of a translation: is it faithful to the source? Gender-register failure lives in a different question entirely — is it right for the reader? — and a text can pass the first while failing the second completely. The maternal-health pamphlet whose terminology lands as clinical-cold, or worse, as shaming; the screening invitation in a register no one would use with the women it addresses; the protection-services intake whose tone forecloses the disclosure it exists to receive. None of this is a mistranslation. All of it is failure.

And it is a uniquely silent failure. The woman it fails does not file a complaint about register; she reads it once and does not return — and the absence shows up nowhere. Outcomes data records low uptake without a cause; the vendor’s QA shows a clean pass; the institution concludes the population is “hard to reach.” Meanwhile the machine layer compounds the problem at scale: translation and generation systems trained on skewed corpora default to masculine forms and flatten register, quietly writing women out of content that was commissioned for them.

In settings where the stakes run as high as maternal health and protection services, this is not a polish issue. The audit exists because the failure produces no signal of its own — someone qualified has to go looking for it, before the absence begins.

The failures that never filed a complaint, found before they cost you the reader.

The audit changes what your quality pass can see. Content is judged against its actual reader, not just its source text — which surfaces the failures accuracy review is structurally blind to. The machine layer is checked before it scales, so a generation system’s masculine default becomes a finding instead of a fleet of published errors. Every register decision arrives documented with its rationale, so when a reviewer, a community advisory board, or your own leadership asks why this term for this audience, the answer exists in writing. And where your programs already use gender-configured interpretation or liaison protocol, the audit is the standard those configurations answer to — one regime, end to end.