Add text to get the translation that 22 AI models agree on
Who counts as a qualified human reviewer of AI translation?
A fluent AI translation hides its own errors. A sentence reads smoothly, a number is quietly dropped, and the error surfaces only when someone relies on the document. That is why laws about AI translation keep reaching for one safeguard, human review, and why the safeguard fails when nobody says what counts as review.
On September 19, 2026, California enacted SB 1360, which presumes that translated election materials produced solely by automated translation services are insufficient to convey intent and essential meaning (California Elections Code section 14035, 2026). The position I hold is that a human review requirement works only when it names who reviews and what that person must be able to show.
Federal healthcare law defines the reviewer in three parts. California's election text, in the sentence that creates the presumption, does not.
This article compares the two with California's own court protocol, then proposes four tests for deciding whether human review of AI translation is real. It covers written translation only.
Key takeaways
- SB 1360, signed on September 19, 2026, presumes that election translations produced solely by automated translation services are insufficient to convey intent and essential meaning.
- Under Section 1557, a qualified translator has demonstrated written proficiency, translates effectively, accurately, and impartially, and follows translator ethics, and a qualified human translator must review machine translation of critical content.
- A bilingual employee's self-identified proficiency alone is insufficient under Section 1557 (HHS Office for Civil Rights, 2024), and California's court protocol calls for a second independent professional translator.
- The sentence that creates the SB 1360 presumption does not say what overcomes it, so the definition of human review is still open in California election law.
- The four tests for human review of AI translation ask whether the reviewer is qualified, whether the review is independent, whether a record documents it, and whether a named person is accountable.
What does California's SB 1360 say about AI translation?
SB 1360 says translated election materials must convey the intent and essential meaning of the English original, and that materials produced solely by automated translation services are presumed insufficient to do so. The text does not say what human involvement overcomes that presumption.
The governor signed the bill on September 19, 2026, and it became Chapter 339 of the Statutes of 2026. It covers election materials such as official and sample ballots, voter registration forms, and election notices and instructions, and it requires them to be of equal quality to the English versions and distributed at the same time.
SB 1360 on election materials is a different bill from AB 1360 on English learner reclassification. As I read the sentence, it sets a default and leaves the exception to be defined. This article is general information and no substitute for legal advice.
Who counts as a qualified human translator under federal law?
Under the federal Section 1557 rule, a qualified translator has demonstrated proficiency in writing and understanding English and at least one other written language, translates effectively, accurately, and impartially using any necessary specialized vocabulary, and follows generally accepted translator ethics, including confidentiality (45 CFR 92.4, current).
The same rule requires a qualified human translator to review machine translation when the text is critical to a person's rights, benefits, or meaningful access, when accuracy is essential, or when the source contains complex, non-literal, or technical language.
When an emergency makes immediate review impossible, HHS says the machine translation must be checked by a qualified human translator as soon as practicable. As far as I understand, the rule's language access provisions remain in force as of November 2026, so check the current text at the eCFR before relying on this summary.
Does a bilingual employee's review count as human review?
No, not by itself, because reading a language does not make someone a qualified translator. HHS says an individual's self-identified proficiency alone is insufficient to show the proficiency Section 1557 requires (HHS Office for Civil Rights, 2024), and California's court protocol calls for a second independent professional translator (Judicial Council of California, 2024).
What I look for in a review step is a reviewer whose qualification can be shown, since fluent AI translation gives a reader with partial skills little reason to disagree.
California election law already names qualified translators elsewhere. Its candidate statement provision lists, among others, a certified and registered interpreter on the Judicial Council Master List and an interpreter categorized as certified or professionally qualified by the Administrative Office of the United States Courts.
The court protocol adds a distinction that applies here. Not all certified interpreters are qualified to translate legal documents, and the second reviewer must meet the same qualification requirements as the first.
What does SB 1360's presumption leave open?
The sentence that creates the presumption does not say what human involvement overcomes it, who must provide that involvement, or what record shows it happened. Those three questions decide whether a translation produced with AI translation and then reviewed by a person counts as sufficient.
California already has an answer in its own courts. The Judicial Council's protocol says machine translation can help with an initial translation as long as a qualified human translator then carefully and fully reviews and edits it, and that machine translation should not be the sole mechanism for translating its materials.
That protocol governs court materials, so it does not bind election officials. It does show that a workable definition of human review exists in California and could be borrowed.
My concern is that those three questions will be answered one case at a time, after a mistranslated notice has already reached a voter.
What makes human review of AI translation count in practice?
Four tests make human review count. The reviewer is qualified and independent of both the machine step and the first translator, a record documents the review, and a named person is accountable for the result. A step that fails any of the four is a read-through.
The test I propose has four parts, and each can be answered with a document. The table shows how three sources handle them, and where the text I reviewed is silent.
| Source | Qualified | Independent | Documented | Accountable |
|---|---|---|---|---|
| Section 1557 rule, 45 CFR 92.4 and 92.201(c) (current) | Three-part definition of a qualified translator | Not stated in the provisions cited | Not stated in the provisions cited | Not stated in the provisions cited |
| Judicial Council translation protocol (2024) | American Translators Association certification, or qualification shown by experience, education, and references, in the right language and direction | A second independent professional translator reviews | Records of translator, reviewer, legal reviewer, date, and language | The vendor commits to corrections after delivery |
| SB 1360 presumption sentence (2026) | Not stated in the sentence | Not stated in the sentence | Not stated in the sentence | Not stated in the sentence |
Qualified means the reviewer meets a definition you can name, such as the three-part federal definition or a credential like American Translators Association certification. Independent means the reviewer did not produce the translation being reviewed, which is how Tomedes' human translation process works, with an independent second reviewer checking the work before delivery.
Documented means a record shows who translated, who reviewed, and when. Accountable means a named person corrects errors found after delivery.
How can a business check that human review of AI translation is real?
Ask in writing who reviews the AI translation output, what qualifies that person, who checks the review, and what record shows it happened, before you rely on the document. Then ask who is accountable for correcting an error found after delivery.
- Ask who reviews the AI translation output and what qualifies that person.
- Ask whether the reviewer is independent of the machine step and of any first translator.
- Ask what record shows the review happened, including who reviewed and when.
- Ask who is accountable for correcting an error found after delivery.
At Tomedes, AI post-editing is performed by domain-matched human post-editors under ISO 18587:2017, and Tomedes can provide the translator's qualifications for a project on request. The AI translation track has no human review step in its default workflow, and a buyer who needs one requests AI post-editing as an add-on or orders human translation, as this pricing explanation describes. How Tomedes combines AI translation with human review is covered in a separate guide.
For legal documents that go to a court or an agency, the legal translation service is built for that work. For any business reading this, my recommendation is to ask the four questions before the next filing.
FAQs
Q: What makes someone a certified translator?
A: A certified translator either holds a credential from a professional body, such as American Translators Association certification, or signs a statement that a specific translation is complete and accurate. Section 1557 asks for a qualified translator instead, defined by demonstrated proficiency, accuracy, and ethics, and it names no credential.
Q: Is there a certification for a qualified human translator?
A: No, not as a single credential, because the federal rule asks for demonstrated proficiency without naming one. California's court protocol recommends American Translators Association certification and accepts experience, education, and references where no certified translator exists.
Q: Is a qualified medical interpreter the same as a qualified translator?
A: No, because Section 1557 defines them separately, with a qualified interpreter working in spoken language and a qualified translator working in written language. California's court protocol adds that not all certified interpreters are qualified to translate legal documents.
Written by Rachelle Garcia
Connect on LinkedInRachelle leads product and AI at Tomedes, where she runs the experiments that turn internal data into better translation experiences. She writes about what actually happens when you build AI products such as MachineTranslation.com — the numbers, the surprises, and the parts that don't go to plan.
Share:
Post your Comment