AI-assisted translation is emerging as a promising tool to help courts better serve individuals with limited English proficiency. While language access has long been a challenge in the justice system, advances in artificial intelligence are creating new opportunities to make legal information more accessible. Yet alongside these opportunities come important questions about accuracy, fairness, and the role of human oversight.
Image générée par ChatGPT.
These questions were the focus of a recent webinar co-hosted by the National Center for State Courts (NCSC) and the Thomson Reuters Institute. Held on May 21, 2025, the session, titled “AI-Assisted Translation in the Courts: Exploring Opportunities and Navigating Risks”, examined how courts are beginning to use AI tools for translation, what risks need to be managed, and how thoughtful implementation can help support both efficiency and justice.
A Step-by-Step Approach Inspired by Orange County
A helpful example of how courts can introduce AI-assisted translation carefully comes from the Orange County Superior Court in California. Recognizing that many commercially available tools failed to capture court-specific terminology, the court decided to develop its own system, known as the Court Application for Translation (CAT). This platform is powered by Microsoft Azure’s Translator but was trained using legal glossaries created by the Judicial Council of California, along with actual court forms and notices in Spanish and Vietnamese, the two most widely used non-English languages in the region.
To manage risk and allow time to improve the tool gradually, the court adopted a rollout in three phases:
- Phase one: focused on educational materials and video scripts
- Phase two: now underway, is centered on juvenile court reports
- Phase three: will involve collaborative court essays, including translations from Vietnamese or Spanish back into English
This structure was designed to gradually increase the level of complexity. Educational content provided a manageable starting point. Most documents in Phase 1 were short, with the exception of one 300-page handbook that was intentionally included to test how the system would handle longer formats. This turned out to be an important experiment. The tool performed well overall but showed some difficulty maintaining structure and consistency in longer documents. In response, the technical team is now working on ways to improve the system’s ability to manage extended texts.
Phase two introduces a more demanding use case. Juvenile reports are often lengthy, sensitive, and time-consuming to translate manually. The final phase, reverse translation, presents its own challenges, as it tests the system’s ability to capture meaning going in both directions.
During the first phase, which ran from October 2024 through April 2025, 113 translated sections were reviewed by certified Spanish and Vietnamese language professionals. The results offered useful insight:
- Spanish:
• 80% accurate with no changes needed
• 17% needed minor corrections
• 3% had major errors
- Vietnamese:
• 57% fully accurate
• 39% had minor issues
• 4% had major errors
This gap was largely due to the limited availability of high-quality training materials in Vietnamese, compared to the more extensive and standardized Spanish resources.
In addition to human review, the team also tracked BLEU scores, a widely used metric that compares AI-generated translations to expert human ones by checking for overlap in words and phrases. The scores range from 0 to 100, with higher numbers indicating greater accuracy and fluency. Spanish translations scored 66.21 and Vietnamese 63.71, both considered very high, though the difference echoed what reviewers saw in practice.
The court has recently received formal approval to move forward with Phase 2 and is currently collecting juvenile reports to begin the next round of testing and evaluation. These reports typically exceed 100 pages and can take interpreters between three and four hours to translate manually. The court sees this phase as an opportunity to reduce that burden while still ensuring quality. At the same time, the team continues to assess risks and make improvements where needed. Ongoing input from judicial officers has helped guide the project, keeping it aligned with broader goals around access to justice and procedural fairness.
Since then, the court has continued refining the system. Translator feedback is used to make ongoing adjustments, including grammatical fixes and improvements in handling longer documents. With each step, the focus remains on careful expansion, not just to improve efficiency, but to ensure that the translated materials meet legal standards and can be trusted by the people who rely on them.
Challenges in the Courtroom: Shortages, Delays, and Risk
Language access remains a long-standing challenge in the justice system. Many courts struggle with:
- A limited supply of certified interpreters, especially for less common languages
- High costs and logistical delays in human translation processes
- Uneven availability of services between urban and rural regions
These barriers can cause delays, reduce the quality of communication with litigants, and erode confidence in the legal process. AI tools have emerged as one possible way to ease these constraints, but they come with challenges of their own.
The Promise and the Limits of AI Translation
AI translation tools can enhance court operations by making informational materials more accessible. Practical uses include:
- Drafting translations of forms, public-facing materials, or video scripts
- Supporting staff in providing basic guidance where no interpreter is available
- Reducing the time needed to produce first drafts of large documents
However, these benefits are paired with important limitations:
- Difficulty handling legal nuance or culturally specific meaning
- Greater likelihood of error in high-stakes or sensitive content
- Risk of bias, particularly in gendered or less-resourced languages
These tools are most effective for low-risk content and must be complemented by qualified human translators.
Keeping Humans in the Loop: A Non-Negotiable Safeguard
In legal contexts, accuracy is essential. Courts that adopt AI translation systems are encouraged to:
- Require human post-editing for any translated material shared with litigants
- Establish strong internal quality control processes
- Include clear notices indicating when AI tools have been used
Some systems, like the one in Orange County, are designed to operate within court-controlled environments, reducing privacy concerns. Human reviewers operate under confidentiality agreements, just as they would in traditional workflows. These safeguards help ensure the integrity of the translation process.
Ethical Issues and Responsible Implementation
As with any technology in the justice system, courts must be mindful of how AI is introduced and used. Ethical concerns include:
- Transparency: Letting users know when AI tools are involved
- Accountability: Ensuring responsibility for errors remains clear
- Privacy: Protecting sensitive information from exposure
- Access: Giving users the ability to request human translation where needed
Clear internal policies and regular reviews are key to navigating these questions responsibly. Courts are encouraged to test systems under real conditions and involve legal and linguistic professionals in evaluating outcomes.
Moving Forward: Combining Innovation with Caution
AI-assisted translation has real potential to improve language access in the courts. But that potential will only be realized if the technology is deployed with care.
A strong implementation approach includes:
- Starting with non-judicial documents and public education materials
- Building the capacity to assess and refine AI tools over time
- Putting safeguards in place to ensure quality and privacy
- Providing staff with the training and support they need
With this kind of foundation, AI tools can support courts in serving diverse communities more effectively, without compromising fairness or trust.
Key Takeaways
AI translation tools can help courts reach more people, faster. But they must be supported by human expertise, clear policies, and ongoing evaluation. With a careful approach, courts can use these tools to expand access to justice while protecting the values at the heart of the legal system.
For those interested in learning more:
This content has been updated on 10/29/2025 at 15 h 39 min.