AI Interpretation Is Advancing. Some Languages Can't Follow Yet

Equiti Just Made a Big Bet on AI-Human Hybrid Interpretation

Equiti, the company behind Martti's video remote interpretation network, acquired Lexi AI to build what it calls a "connected modality" model for healthcare language access. The idea is straightforward: use AI for routine, lower-stakes encounters in supported languages, and route complex or high-risk situations to qualified human interpreters. Lexi covers 35-plus languages and dialects. Martti's interpreter network covers more than 250.

This is a serious development. Equiti is not dabbling. Martti pioneered video remote interpretation in healthcare over two decades ago, and combining that infrastructure with purpose-built AI — integrated directly into EHR and phone workflows — gives health systems a genuinely useful tool for managing language access at scale.

But the announcement also reveals something worth paying attention to: the gap between what AI can do and what patients in certain communities actually need.

The 35-Language Ceiling and Who Falls Through It

Lexi supports 35-plus languages and dialects. That sounds like a lot. It is not.

There are roughly 7,000 living languages. Of the ones spoken by patients in US hospitals, a significant number remain entirely outside the reach of any commercially available AI interpretation tool. That includes Chuukese and Pohnpeian — two languages spoken by tens of thousands of Micronesian migrants living across Hawaii, Guam, Arkansas, and other US communities.

Chuukese speakers are disproportionately represented in emergency departments. Compact of Free Association agreements give citizens of the Federated States of Micronesia the right to live and work in the United States without a visa, which means Chuukese and Pohnpeian-speaking patients regularly walk into American hospitals expecting language access they are legally entitled to under Title VI of the Civil Rights Act and Section 1557 of the Affordable Care Act.

AI cannot serve them. Not today, and not in any near-term roadmap any vendor has announced.

Why AI Lags on These Languages

AI language models are only as good as the training data behind them. Languages like Spanish, Mandarin, Arabic, and French have billions of text tokens available for training. Chuukese has a fraction of that. Pohnpeian has even less.

Beyond data scarcity, these languages present structural challenges that trip up models trained on Indo-European patterns. Chuukese is a Micronesian language with complex verb morphology. Small errors in interpretation — a negation missed, a tense misread, a medical term approximated — can produce a clinically dangerous misunderstanding.

For a post-discharge medication instruction or an appointment reminder, an AI approximation might be acceptable with proper disclosure. For informed consent, mental health screening, or an oncology diagnosis conversation, it is not. That is exactly the context where Chuukese and Pohnpeian speakers most need accurate interpretation, and exactly where no AI tool can currently provide it.

What the Equiti Model Gets Right — and What It Assumes

Equiti's "Modality Matters" framework is genuinely thoughtful. The principle — match each encounter to the appropriate modality — is correct. The problem is that this framework assumes your patient's language is available in the AI modality at all.

For a Spanish-speaking patient, a health system using Equiti's combined platform can dynamically shift between Lexi's AI and a Martti interpreter depending on clinical complexity. That flexibility is valuable.

For a Chuukese-speaking patient, there is no AI modality. Every encounter routes to a human interpreter. That is fine — it is the right outcome — but it means health systems serving Micronesian communities cannot reduce interpreter demand through AI the way they can for higher-resource languages. Budget models built around AI deflection will not apply equally across their patient populations.

Health system administrators planning their language access programs around hybrid AI-human models need to account for this. If your service area includes Micronesian communities, your per-encounter interpreter costs for those patients are not going to fall in line with your Spanish or Vietnamese costs. Plan accordingly.

The Compliance Picture Hasn't Changed

Nothing in the Equiti-Lexi deal changes the federal compliance requirements for limited English proficiency patients. Under Title VI and Section 1557, healthcare organizations must provide meaningful access to qualified interpreters — at no cost to the patient — for any language the patient needs.

"Meaningful access" for a Chuukese speaker means a qualified human interpreter, full stop. There is no AI shortcut, and there is no waiver for languages that are difficult to support.

The table below summarizes how the options break down for rare Pacific languages right now:

For Chuukese and Pohnpeian, every row in that table is the same. That is not a criticism of Equiti's model. It is a reality that health systems serving these populations need to build into their language access planning.

What This Means for Your Language Access Program

If you run language access for a health system, school district, or government agency serving Pacific Islander communities, the practical takeaway from this acquisition is not about AI at all. It is about making sure your human interpreter pipeline for rare Micronesian languages is reliable and compliant, because that pipeline is not getting replaced anytime soon.

Vet your vendors on rare-language depth, not just total language count. A vendor boasting 250-plus languages means little if Chuukese is a gap they fill with a bilingual staff member who has no medical interpreter training.

If you need a qualified partner for Chuukese or Pohnpeian interpretation, TXLOC can help you assess what your current setup is missing.