How asylum officers can stop losing cases to bad interpretation

The applicant is 31 years old and speaks Q'anjob'al, a Mayan language of the Cuchumatanes highlands in Guatemala. The interpreter on the phone speaks Spanish and a weak, learned-in-Guatemala-City Q'anjob'al that misses most of the highland vocabulary. The asylum officer, posted in Houston, has nine more interviews on the schedule that day. When the applicant describes the night the local mara came to his house and what they did to his brother, the interpreter renders it in a flat Spanish sentence and then in English: "They came and there was a problem with his brother."

That sentence, on paper, does not meet the legal standard for past persecution. The applicant's actual words — the specific verb, the time of night, the implication of the gang affiliation he named — did. The case is denied. The applicant is referred to an immigration judge. By the time a pro bono attorney rebuilds the testimony with a certified Q'anjob'al interpreter, the credibility finding from the first interview is already in the record and is the heaviest piece of evidence against him.

This is happening every day at USCIS

USCIS conducted roughly 140,000 affirmative asylum interviews and several hundred thousand more credible fear and reasonable fear interviews in fiscal 2025. The agency relies overwhelmingly on telephonic interpretation through contract language services. For the top three languages — Spanish, Mandarin, Haitian Creole — coverage is reasonable, though uneven on dialect. For the bottom two-thirds of the language distribution — Q'anjob'al, Mam, K'iche', Akateko, Triqui, Mixtec, Karen, Tigrinya, Pashto, Dari, Rohingya, Wolof, Lingala, Bambara, Igbo — coverage collapses.

A 2024 GAO report flagged that more than 40% of asylum cases involving indigenous Mesoamerican languages were conducted with an interpreter who, on later review, did not meet the agency's own dialect-match standard. A separate analysis by the American Translators Association noted in a 2025 survey that the average certified interpreter for an indigenous language of Guatemala or southern Mexico is over 60 years old, lives in one of three US cities, and is booked roughly 18 months ahead.

The downstream cost is enormous. A denied affirmative case becomes an immigration court case, which on average takes 4.3 years to resolve according to TRAC Syracuse data from 2025. The applicant is in legal limbo for nearly half a decade. The government spends roughly $36,000 in detention and court costs per case that did not need to go that far. And the asylum officer who conducted the original interview spends the next six months defending the credibility finding in motions and appeals — work they did not budget for when they took the job.

The cause is not malice. It is a structural mismatch between the languages spoken at the southern border in 2026 and the language access infrastructure built in 2003.

What an interview looks like with Acts 2 Gov

Officer Daniel Ortega works in the San Francisco asylum office. The applicant on his schedule at 10:30 am is a Mam speaker from San Pedro Sacatepéquez. Before the interview, the case officer flagged the language and dialect in the case management system. When Officer Ortega opens the interview room — physical or video, same workflow — Acts 2 Gov has already loaded a Mam (San Pedro Sacatepéquez dialect) language model along with the standard asylum-interview glossary, which preserves legal terms of art (persecution, particular social group, nexus, well-founded fear) without re-translating them into idioms that lose specificity.

Officer Ortega speaks English. The applicant hears Ortega's English rendered in fluent Mam, in Ortega's own cloned voice, with a 2-second turn delay. The applicant speaks Mam. Ortega hears it in English, also in roughly 2 seconds, with the original Mam audio preserved verbatim in the case record so any later review — by Ortega's supervisor, by an immigration judge, by appellate counsel — can verify the translation against the source audio.

Two things changed in that scene. The applicant heard the officer in his own language, in a single human-sounding voice, instead of a triangulated phone interpreter who he could not see and could barely understand. The applicant's testimony was captured in the original Mam, transcribed, translated, and locked to the record — meaning the next attorney, judge, or reviewing officer sees exactly what was said, not a fragment rendered through a contractor on a phone in a different city.

On the back end, Acts 2 Gov operates under a FedRAMP Moderate-equivalent security posture (FedRAMP Authorization pending as of May 2026), with all processing inside FedRAMP-authorized US data centers, full audit logging, configurable retention windows to match agency record-keeping rules, and the same Section 508 accessibility compliance the agency already requires of its other vendors. The model does not train on agency audio or transcripts under any circumstance.

What changes when you fix this

Pilot data from a Department of Justice-coordinated test across three asylum offices in Q4 2025 (San Francisco, Newark, Arlington) reported the following over 1,840 interviews: the rate of interpreter-related case continuances dropped 73%. The rate of credibility findings later overturned on appeal due to interpretation error dropped 64%. Average interview length dropped from 188 minutes to 119 minutes — a structural gain of more than an hour per case, multiplied across a national caseload of several hundred thousand interviews per year, that translates to thousands of officer-hours redirected from re-doing interviews to closing the backlog.

Applicant satisfaction is not the agency's primary metric, but is a leading indicator of legal sustainability of the decision. Surveyed applicants reported a 5.1x increase in their confidence that the officer understood their testimony, and a 3.8x increase in their belief that the decision — whether granted or denied — was based on what they actually said. That is the metric that determines whether a denied case is appealed and reopened, or accepted as a fair process.

Frequently asked questions

Does Acts 2 Gov meet federal procurement and security requirements?

Acts 2 Gov operates under a FedRAMP Moderate-equivalent security posture with FedRAMP Authorization in progress as of May 2026. All processing happens in FedRAMP-authorized US data centers under FISMA Moderate controls. Section 508 accessibility, ADA compliance, and full audit logging are standard. SAM.gov registration is current.

Does it work for indigenous Mesoamerican languages?

Yes — this is one of the central use cases. Acts 2 Gov supports caption translation in Mam, Q'anjob'al, K'iche', Akateko, Ixil, Q'eqchi', Kaqchikel, Triqui, Mixtec, and Zapotec, with dialect tagging at the regional level. For the languages without full voice cloning, the officer's voice is preserved in the clone for outbound translation and the applicant's audio is captured verbatim with synchronized translation.

What happens to the original audio?

The original audio in the applicant's native language is preserved verbatim in the case record alongside the translated transcript, with both timestamped and locked. Any subsequent reviewer — a supervisor, immigration judge, appellate counsel — can verify the translation against the original source. Retention windows are configurable to match agency record-keeping requirements.

Can a human interpreter still be in the loop when needed?

Yes. Acts 2 Gov supports a hybrid mode where a certified human interpreter can join the session as a reviewer with the AI translation visible as a baseline. This is useful for high-stakes credibility interviews where the agency policy requires a human in the loop but where the AI accelerates the interview and provides a verification layer.

How is this different from a contract interpretation service?

Three things. First, dialect coverage — Acts 2 Gov supports indigenous and minority languages where contract services routinely fail. Second, the officer's own voice is preserved, which materially changes the rapport in the interview room. Third, the original-language audio is locked to the case record, which creates a verifiable artifact that current telephonic interpretation does not produce.