DEV Community

CourtGPT
CourtGPT

Posted on

Translation Services for Courts: The Case for Bilingual Self-Help Portals

#legaltech #ai #translation #courts

Translation Services for Courts: The Case for Bilingual Self-Help Portals

Roughly 8.6% of U.S. residents speak English "less than very well" according to U.S. Census data. For state court systems, this presents a specific challenge: legal information must be reachable in the user's primary language. This article summarizes patterns from court translation deployments and self-help portals.

Background

U.S. Census Bureau estimates (2023 ACS) report:

  • 78.6 million U.S. residents speak a language other than English at home.
  • 8.6% speak English "less than very well."
  • The top 10 non-English languages: Spanish, Chinese (incl. Cantonese, Mandarin), Tagalog, Vietnamese, Arabic, French, Korean, Russian, Haitian, Polish.

State courts increasingly require or recommend multilingual access to:

  • Self-help portals
  • Court forms
  • Procedural instructions
  • Translation services for hearings

Source: https://www.census.gov/programs-surveys/acs

Current state

As of early 2026, observed deployments vary widely:

  • California: Multilingual self-help portals across multiple counties (Spanish, Chinese, Vietnamese, Tagalog, Korean, Armenian, Punjabi, Russian). Court interpreter programs at the state level.
  • Texas: Texas Law Help (Spanish, English). Limited-court-interpreter coverage in rural counties.
  • New York: NYC Civil Court multilingual portal (Spanish, Chinese, Russian, Bengali, Haitian Creole, Korean, Arabic).
  • Florida: Multilingual court interpreter registry; partial Spanish + Haitian Creole self-help.
  • Illinois: Cook County Circuit Court multilingual; statewide expansion in progress.

Source: NCSC and individual court websites.

Why translation is hard for legal text

Legal translation has challenges beyond simple language conversion:

  1. Terminology: Legal terms of art don't always have direct equivalents. "Habeas corpus" is used in many jurisdictions unchanged; "Voir dire" too.
  2. Procedural references: Cross-references to statutes and rules may not exist in the target language.
  3. Form layout: Many court forms are pre-formatted in English; bilingual translation can break layout.
  4. Authority conflict: Translated statutes may diverge from the legal authority of the source.
  5. Plain language: Legal English contains archaic structures; translation should produce plain language.

Working pattern: parallel corpus

California's multilingual self-help portals use a parallel-corpus approach:

  • Source statutes and forms in English (authority document)
  • Translated references maintained as glossaries (terminology alignment)
  • AI-assisted drafting in target language, with legal review by bilingual lawyers
  • Final translated documents cite back to authoritative English sources

Working stack:

  • Source docs (English): authoritative
  • Glossary (target lang): maintained by bilingual legal reviewers
  • Drafting: bilingual reviewers verify terminology, citation, plain language
  • Final form: parallel corpus

AI-assisted translation

Large language models can draft first-pass translations faster than human translators. But:

  • Hallucination rates are high in translation contexts (Stanford 2024 measured 12% terminology error in legal translation).
  • Citation fidelity is a specific concern (cited code sections, case names).
  • Cultural framing varies (Spanish legal terminology differs between Spain and Latin America).

Two-pass workflow:

  1. AI generates first-pass translation.
  2. Bilingual reviewer verifies terminology, citations, plain language.
  3. AI regenerates from edited draft for consistency.

This cuts human hours by ~50-70% versus pure manual translation.

Source: Stanford RegLab 2024 translation study.

Authenticated interpreter services

Federal and state courts maintain interpreter rosters. For 2026:

  • Federal: USCA certified interpreters in Spanish, French, Portuguese, Mandarin, Cantonese, Arabic, Haitian Creole, Russian, Polish, Korean, Vietnamese.
  • State: California Judicial Branch interpreter program (~17 languages); Texas (~12); New York (~30).
  • Court interpreter exams: Consortium-based, Federal Court Interpreter Certification Exam.

Language access plans:

  • All federal courts must provide interpreter services at no cost to LEP (limited English proficiency) parties in civil cases (Title VI).
  • 41 states have interpreter statutes for civil proceedings (NCSC counts).

Source: https://www.ncsc.org/data-and-research/access-to-justice

Self-help portal design

For court systems deploying bilingual self-help portals:

  1. Identify top non-English languages in jurisdiction (US Census ACS data).
  2. Translate top-50 common questions across chosen languages.
  3. Translate top-20 court forms.
  4. Maintain glossary of jurisdiction-specific terms.
  5. Test multilingual UX with native speakers.
  6. Track usage and satisfaction by language.

Eval framework

Track:

  • Pages translated and updated monthly
  • Accuracy (legal terminology, citation, plain language)
  • Bilingual reviewer time per page
  • User satisfaction by language
  • File-and-dispose rate by language (no different from English baseline)
  • Time-to-resolution by language

Open-source tools

  • docassemble multilingual support
  • LibreTranslate (machine translation)
  • Hugging Face translation models (NLLB, M2M-100)
  • Open-language-corpora-initiatives

Acknowledgments

This article summarizes public sources as of early 2026. Specific deployments vary by jurisdiction.

Dillon Deutsch has built multilingual systems serving users across 50 US states. https://courtgpt.ai

Top comments (0)