How to Meet the DOJ’s AI Translation Mandate with Speed, Accuracy, and Control

On July 14, the Department of Justice (DOJ) issued a memo calling on federal agencies to adopt artificial intelligence and machine translation to improve the efficiency of multilingual government communications.

Framed as part of a broader effort to implement a March executive order declaring English the official language of the United States, the DOJ memo sets a clear mandate for how government agencies are expected to approach translation work moving forward. When translations are required, the DOJ writes, “technological advances in translation services will permit agencies to produce cost-effective methods for bridging language barriers and reducing inefficiencies with the translation process.”

For the state, local, and federal agencies that administer public benefits programs, one of the most resource-intensive areas of translation is client-facing communications such as eligibility determinations, verification notices, appeals instructions, benefit change notices, and other essential documents sent directly to program participants. These materials number in the hundreds to thousands, and must be updated and retranslated frequently to reflect policy or program changes. Today, this work is almost entirely human-led, making it slow, costly, and well-suited for automation.

The DOJ memo explicitly encourages the “responsible use of artificial intelligence and machine translation.” For agencies serving the more than 8% of Americans with Limited English Proficiency (LEP), the stakes are high. Research shows that programs that don’t have accurately translated materials experience higher rates of unnecessary churn and ultimately spend more to reprocess applications when participants inevitably return.

The challenge most agencies face is finding an approach to AI translation that delivers the efficiency gains the DOJ calls for, without compromising the accuracy that client-facing communications demand.

Navigating the Landscape of AI Translation Solutions

AI translation has made remarkable progress in recent years, with some models achieving near-human performance in many contexts. Some studies show that “even experts cannot reach a consensus on which translation is better.” Research from the University of Washington has found that AI, when paired with human oversight, can deliver high-quality translations in less time and at just 5% of the cost of traditional human-only methods.

For government agencies seeking to modernize their translation processes in line with DOJ guidance, there are several solution types on the market to be aware of. Each has distinct strengths and limitations, and choosing the right fit requires understanding how these tools align with your agency’s priorities, constraints, and specific use case.

1. Standalone AI Platforms

This refers to tools like DeepL, Google Translate, and ChatGPT, which enable users to manually input content and receive translations instantly. They’re inexpensive and easy to use, but not built for secure or compliant workflows. Content must be copied in and out of these systems, which introduces risk, especially if personally identifiable information (PII) or protected health information (PHI) is included. Generative AI tools like ChatGPT require carefully crafted prompts to ensure reliable output, which can compromise efficiency and lead to inconsistent results across users.

Recommendation: These tools can supplement one-off translation workflows involving low-risk non-personalized content, but should only be used with enterprise licenses that enforce data protections and prohibit AI model training on agency data.

2. Fully automated translation pipelines

These systems translate communications automatically after they’ve been generated, often just before delivery. Examples include Amazon Translate, Google Cloud Translation, and SYSTRAN configured for batch processing. They are well-suited to high-volume, templated content like appointment reminders or generic notices, offering cost and speed advantages at scale.

But because translation occurs after the communication is personalized, live client data—names, case numbers, eligibility status—may be processed by third-party AI models. This raises meaningful concerns around data exposure and compliance. These systems also offer limited visibility or control over translation quality and may struggle to preserve formatting and layout in translated versions.

Recommendation: These solutions are appropriate for translating simple, generic communications. Avoid using them for complex, regulated communications and those involving PII, PHI, or other sensitive information unless data redaction and audit controls are in place.

3. Communication platforms with integrated AI translation modules

Some communications platforms, such as CCM, CMS, or CRM tools, now include built-in AI translation features, enabling agencies to manage content creation, translation, and approval in the same environment. This ensures efficient and secure workflows by eliminating the need to move content between tools. These systems support human-in-the-loop review, allowing agencies to balance efficiency with control over the final product. Some tools also offer AI translation accuracy checks that validate meaning, structure, and formatting in translated content, reducing the need for line-by-line QA.

Recommendation: Ideal for government agencies managing translation of regulated, personalized client-facing communications with a high degree of complexity and variation, such as notices that differ by program type or eligibility criteria. Look for platforms that support content reuse across communications and provide integrated glossaries to drive efficiency, accuracy, and consistency at scale.

4. Specialized Translation Management Systems (TMS)

Tools like Smartling, Memsource (Phrase TMS), KantanMT, and SYSTRAN offer advanced translation platforms tailored to high-compliance environments. These systems combine AI-powered translation with human review, quality assurance, glossaries, and detailed audit logs. Many support both interactive use (like standalone tools) and deep integration with other systems via API or plugins. Though historically designed for professional translation teams, many TMS platforms now offer more accessible interfaces and managed service options. Still, they generally assume a higher degree of translation expertise than other solutions.

Recommendation: Ideal for translation of highly regulated or sensitive communications such as standardized forms, legal correspondence, and multilingual web portals. These tools offer the most comprehensive controls over translation quality.

The DOJ’s call for agencies to use AI to streamline translation doesn’t have to come at the expense of clarity, safety, or control. Organizations in highly regulated industries like financial services and healthcare are already using AI to translate sensitive, personalized constituent-facing communications, achieving faster turnaround times and significant cost savings in the process. With the right tools and approach, there’s no reason government agencies can’t achieve those same results.

Originally published on Government Technology Insider