In July 2025, the U.S. Department of Justice (DOJ) released updated language access guidance encouraging federal agencies to consider using AI translation tool when providing multilingual services. This reflects a growing effort to modernize public sector communication, aiming to improve efficiency while maintaining inclusivity for individuals with limited English proficiency.
With more than 350 languages spoken in the U.S., the challenge is real: agencies must deliver timely, accurate translations of public-facing content, ranging from health and housing notices to legal documents and emergency alerts, without compromising on legal, ethical, or accessibility standards.
But as public agencies explore AI tools to meet this demand, a critical question emerges:
How Can Teams Evaluate And Implement AI Translation Tools Responsibly, Without Compromising Accuracy, Trust, Or Civil Rights Compliance?
This guide outlines essential criteria for selecting AI translation tool suited to public sector use. It also introduces MachineTranslation.com, an AI translation platform developed by Tomedes, which enables neutral, side-by-side evaluation of leading AI engines, helping agencies make informed, risk-conscious decisions before adopting a solution. With features such as Secure Mode, which limits processing to SOC 2-compliant AI engines, and anonymized text handling for sensitive data, the platform helps teams meet strict data protection and compliance requirements.
Why AI Translation Matters More Than Ever in Government
Agencies should also be aware that civil rights laws requiring language access in federally funded programs remain in effect, including Title VI of the Civil Rights Act. Although the DOJ’s July 2025 memo rescinds prior internal guidance, it does not override these legal obligations. Ensuring meaningful access for individuals with limited English proficiency remains a compliance requirement.
Modern AI translation tool, powered by neural machine translation (NMT) and large language models (LLMs), can speed up routine translation tasks, especially when combined with oversight and custom glossaries. When used responsibly, they can help agencies promote equity and accessibility without sacrificing quality or compliance.
Important clarification: While the DOJ memo encourages AI-assisted translations, it notably does not mention the need for human oversight. However, longstanding federal guidance from Digital.gov, run by the General Services Administration (GSA), clearly states that “all translations should be checked by a competent human translator.” Agencies should therefore ensure AI outputs are reviewed by qualified linguists, especially for high-stakes content.
“You don’t want to just take a generative AI translation tool and just flip it on and assume everything is perfect.”
— Michael Boyce, U.S. Digital Response
(on the risks of skipping human oversight when using AI translation)
5 Key Criteria for Evaluating AI Translation Tools
1. Translation Accuracy and Contextual Relevance
Public sector materials often deal with benefits, legal rights, or healthcare, areas where mistranslations can cause real harm. Not all AI tools handle nuance or domain-specific language well.
What to look for:
- Support for legal, medical, or technical terminology
- Ability to understand complex sentence structure and tone
- Context-aware language modeling
- Built-in quality scoring or accuracy insights
Tip: Use comparison tools like MachineTranslation.com to test a sample document across multiple AI engines (e.g., DeepL, Google Translate, Microsoft Translator) and evaluate which performs best for meaning, tone, and clarity.
2. Formatting and Document Fidelity
Official documents, like court summons, grant applications, and policy memos, often include structured formatting such as tables, bullet points, headings, or page numbers. When formatting breaks during translation, even accurate text can become unusable.
What to prioritize:
- Upload support for DOCX, PDF, XLSX, and scanned files
- Retention of original layout and visual structure
- Download options in editable formats
Preserving formatting is crucial, especially for documents containing deadlines, case numbers, or procedural instructions.
3. Language Coverage and Terminology Control
Federal and state agencies often need to translate into less commonly supported languages such as Amharic, Haitian Creole, Tagalog, or Punjabi (Shahmukhi). Consistency across terminology is equally important.
Checklist:
- Support for 270+ languages and dialects
- Glossary upload and term management
- Formal or informal tone selection based on audience
Some tools let agencies predefine key phrases, such as program names and benefits terminology, reducing inconsistency and improving public understanding.
4. Security, Compliance, and Privacy Controls
AI tools used by public agencies must adhere to strict data protection rules, especially when translating sensitive or personally identifiable information (PII).
Minimum requirements:
- End-to-end encryption for uploads and downloads
- No persistent storage or backend logging
- Compliance with ISO 27001, SOC 2, or Fed RAMP
- Anonymization tools for added privacy
Note: Avoid using consumer-grade or free tools for official documents. Enterprise platforms with secure modes and compliance-grade architecture are better suited for government use.
5. Scalability and Workflow Compatibility
Translation needs vary across departments and use cases, from bulk data exports to one-time form updates. Choose a tool that fits your operational scale.
Ideal features:
- Batch translation and file queuing
- Pay-as-you-go or tiered pricing
- API access for CMS or database integration
- Multi-user support for collaborative teams
A scalable tool reduces operational bottlenecks, enabling faster multilingual publishing and better resource allocation.
Simplifying the Evaluation Process
Testing multiple AI tools internally can be time-consuming. That’s where neutral comparison platforms like MachineTranslation.com come in.
With this platform, agencies can:
- Upload a sample document
- Run side-by-side translations with multiple engines
- See quality scores, formatting retention, and tone differences
- Evaluate terminology handling in real time
- Request Human Verification for sensitive or high-stakes content
MachineTranslation.com also offers a Human-in-the-Loop option for public sector users. For critical communications, such as court notices, healthcare guidance, or legal disclaimers, agencies can request a certified human linguist to review and refine AI-generated translations. This hybrid approach ensures that AI efficiency is paired with professional-level accuracy and civil rights compliance, especially where language precision impacts legal access or public safety.
By combining AI-powered translation with optional human review, the platform gives public sector teams a responsible path forward. It supports speed and cost-efficiency without compromising on quality or accountability, an essential balance in light of DOJ guidance and ongoing obligations under Title VI of the Civil Rights Act.
Example Use Case: Translating Court Appearance Notices
Imagine a state court needing to issue appearance notices in Spanish, Vietnamese, and Russian. These documents include case numbers, legal terms, and deadlines, all of which must be accurately translated without disrupting the layout.
Using MachineTranslation.com, the court:
- Uploads a redacted sample notice
- Compares output from the four top translation engines
- Reviews how each handles terms like “arraignment” or “failure to appear”
- Assesses formatting consistency and form layout
- Selects the most reliable engine for ongoing use
This fast, low-risk process supports compliance while preserving translation quality.
Getting Implementation Right
Selecting the right AI tool is only step one. Agencies should establish best practices to guide adoption and mitigate risks.
Recommendations:
- Pilot before scaling:
Start with internal or low-risk documents
- Define key terms early:
Upload glossaries for programs, departments, and legal language
- Include human oversight:
Assign bilingual staff or third-party reviewers to verify output
- Train internal users:
Educate teams on when AI tools are appropriate versus when human translators are required
- Maintain documentation: Ensure that all workflows are traceable and audit-ready
Final Thoughts: Access Through Accuracy
The DOJ’s memo encourages efficiency, but public trust depends on accuracy, fairness, and accessibility. Translation errors, especially in legal or health content, can lead to exclusion, misunderstandings, or even harm.
When used responsibly, AI translation tool can improve equity and reduce barriers, but only with proper safeguards and evaluation.
Platforms like MachineTranslation.com help ensure that agencies make informed, side-by-side comparisons before choosing a tool. By prioritizing accuracy, compliance, and transparency, public sector teams can better serve all communities, no matter what language they speak.
