Healthcare

Machine Translation for Vital Documents: When Human Review Is Required

Under the HHS Section 1557 rule, 45 CFR 92.201(c)(3), a covered health program that uses machine translation must have a qualified human translator review it when the text is critical to a person's rights, benefits or meaningful access, when accuracy is essential, or when the language is complex or technical. Other public agencies face fewer explicit rules.

Key takeaways

Section 1557 does not ban machine translation.
A qualified human translator must review machine translation of critical, technical or complex content.
Being bilingual does not make someone a qualified translator by default.
A full post-editing workflow with independent review and records protects accuracy and privacy.

In this guide

3 triggers

That require qualified human review

647 sentences

Tested in a 2019 study of discharge instructions

8 steps

In a sound post-editing workflow

What Does Section 1557 Say About Machine Translation?

The 2024 Section 1557 final rule took effect on July 5, 2024. The eCFR shows 45 CFR 92.201 unchanged as of October 1, 2026.

Section 92.201(c)(3) sets three triggers for human review. Review by a qualified human translator is required when:

The underlying text is critical to the rights, benefits or meaningful access of a person with limited English proficiency.

Accuracy is essential.

The source contains complex, non-literal or technical language.

The rule defines machine translation as automated, text-based translation “without the assistance of or review by a qualified human translator” (45 CFR 92.4).

The rule does not ban machine translation. In the preamble, HHS Office for Civil Rights (OCR) said machine translation “still carries significant potential for error.” OCR gave one exception example. An emergency medical technician with no other language help may use machine translation while a qualified interpreter is found. OCR said that output must be checked by a qualified human translator “as soon as practicable.”

OCR also recommended warning patients when machine translation is used without human review. OCR repeated the human review requirement in a Dear Colleague letter dated December 5, 2024.

Who Counts as a Qualified Translator?

Under 45 CFR 92.4, a qualified translator must meet three conditions:

Demonstrated proficiency in writing and understanding written English and at least one other written language.

The ability to translate accurately and impartially, using specialized terms “without changes, omissions, or additions,” while preserving tone and sentiment.

Adherence to generally accepted translator ethics, including confidentiality.

Being bilingual does not meet this definition by default. The rule’s definition of qualified bilingual staff covers oral language help, not written translation.

What Other Guidance Addresses Machine Translation?

U.S. Department of Justice. An Attorney General memorandum dated July 14, 2025, implements Executive Order 14224. It encourages federal agencies to consider “responsible use of artificial intelligence and machine translation” for people with limited English proficiency. It also states that information an agency deems mission critical “should be translated accurately.” The memo suspended LEP.gov pending an internal review.

California courts. The Judicial Council of California Translation Protocol (updated October 2024) states that machine translation “should not be used as the sole mechanism” for Judicial Council materials. It allows machine translation for a first draft “as long as a qualified human translator then carefully and fully reviews and edits the translation.” Where raw machine translation appears on its website, users must see disclaimers in their primary language.

ISO 18587. ISO 18587:2017 sets requirements for “full, human post-editing of machine translation output and post-editors’ competences.” ISO lists the standard as due for revision.

What Can Go Wrong When Machine Translation Is Not Reviewed?

A 2019 study in JAMA Internal Medicine tested Google Translate on 647 sentences from emergency department discharge instructions. It found 92% of Spanish sentences and 81% of Chinese sentences accurate. It also found 2% of Spanish and 8% of Chinese sentences had potential for significant harm. The study reported one example in which “hold the kidney medicine” (meaning stop it) became an instruction to keep taking it. Machine translation engines have changed since 2019, so these rates may not match current tools.

Errors with the highest stakes tend to fall into a few groups:

Dosage and instructions: units, frequency, “hold,” “as needed” and negation.
Deadlines and dates: response windows, date formats and “days” versus “business days.”
Legal terms: appeal, fair hearing, grievance, guardian and power of attorney.
Names and numbers: case numbers, phone numbers and addresses.
Reading level and tone: output that is harder to read than the English source.

Using machine translation for patient materials?

Our qualified linguists post-edit and independently review machine output for vital content.

When Is Machine Translation With Full Post-Editing Acceptable?

Use this table as a starting point. Your legal counsel and language access plan decide the final policy.

Content	Example	Recommended approach
Internal understanding only	Reading an incoming letter written in another language	Raw machine translation for gist, then qualified review before any action
General information	Office hours, parking, general program FAQs	Machine translation with full post-editing by a qualified translator
High-volume, repetitive text	Updates to a large knowledge base with an approved glossary	Machine translation with full post-editing and terminology checks
Vital documents	Consent forms, eligibility and denial notices, appeal rights	Human translation plus independent review; full post-editing only with a second qualified reviewer
Clinical instructions	Discharge and medication instructions	Human translation with clinical terminology review
Legal and court content	Hearing notices, court forms	Human translation; machine output only as a draft for a qualified translator
Languages with weak engine output	Languages your vendor flags for low machine quality	Human translation
Emergencies with no other help	Urgent care before an interpreter arrives	Machine translation as a stopgap, with qualified human review as soon as practicable

How Does Machine Translation Post-Editing Work?

Machine translation post-editing (MTPE) means a qualified linguist corrects machine output against the source text. Full post-editing aims for a result that reads as a complete, accurate translation. Light post-editing aims only for understandable text. Light post-editing does not fit vital content.

A sound MTPE workflow for important documents follows these steps:

Prepare the source. Fix errors and use plain language before translation.

Protect privacy. Section 1557 requires language services that protect privacy. Use a secure engine that does not retain your text.

Load terminology. Apply an approved glossary for program names and legal terms.

Run the engine. Generate a draft in the target language.

Post-edit in full. A qualified translator checks every segment against the source for meaning, numbers, dates and tone.

Review independently. A second qualified linguist reviews vital content.

Check the final layout. Confirm text, links and formatting in the finished file.

Keep records. Note which documents used machine translation and who reviewed them.

Frequently Asked Questions

Does Section 1557 ban machine translation?

No. The rule allows machine translation. It requires review by a qualified human translator when content is critical to rights, benefits or meaningful access, when accuracy is essential, or when language is complex, non-literal or technical. For content outside those triggers, OCR recommends telling patients the translation may contain errors.

Can our bilingual staff review machine translation?

Only if they meet the qualified translator definition in 45 CFR 92.4. That means demonstrated written proficiency in both languages, the ability to translate accurately and impartially, and adherence to translator ethics. Speaking two languages fluently does not establish written translation skill. Test and document staff skills before assigning review work.

Do non-health agencies have a machine translation rule?

We did not find a federal rule outside Section 1557 that sets a specific human review standard. The July 2025 Attorney General memo encourages “responsible use” of machine translation. Some state bodies set their own policies, such as the Judicial Council of California protocol. Check your state law, contracts and grant terms.

Should we tell readers when content is machine translated?

Yes, when no qualified human has reviewed it. OCR recommends warning patients that unreviewed machine translation may contain errors. The Judicial Council of California requires disclaimers in the user’s primary language for its machine-translated web content. A short notice in each language helps readers decide when to ask for help.

How AsianText Helps

Human Review for Every Vital Document

AsianText combines machine translation with full post-editing by qualified linguists under ISO 17100 and ISO 9001 certified processes. We assess which content suits AI Translation & Post-Editing and which needs human Translation with independent review. We work across 200+ language combinations with 3,000+ native linguists. We can also build these decisions into your Language Access Plan. Contact us to review your document mix.

This article provides general information, not legal advice. Last updated October 5, 2026.

Written by Asyiah Abdullah

Part of the AsianText team, writing about language access, translation and interpreting for government, healthcare and life sciences.

Sources

Related Services

Serving Patients in Every Language?

Tell us the languages, the content and the deadline. We reply with a fixed quote within one business day.

+1 123-456-7890