A Spanish-speaking patient walks in. The options a health system has depend on its size and resources. Larger systems keep one or more in-house Spanish interpreters on staff. Those interpreters speak a neutral Spanish, not granular to every dialect. Smaller systems pull a proficient staff member from their primary role. Across the board, phone interpretation fills the gaps: widely available for Spanish, rarely configurable to the dialect level, with no meaningful choice between Mexican Spanish and Cuban Spanish. The structural limits of these approaches, primarily dialect coverage and scalability, explain why language access in healthcare is shifting from a task model to an infrastructure model.
Language access treated as a task to complete is different from language access treated as infrastructure to build. The first produces workarounds. The second produces flows. Dr. Gezzer Ortega, Assistant Professor of Surgery and Lead Faculty for Research and Innovation for Equitable Surgical Care at Brigham and Women's Hospital, Harvard Medical School, has spent his career at that intersection. On Care Culture Talks, he brought both the research and the biography.
1. How Many Americans Face Language Barriers in Healthcare?
"There are 31 million people in the United States with diabetes," Dr. Ortega said on the podcast. "We think that is a big problem. Twenty-six million is not far from that." The 26 million Americans who speak English less than very well face elevated rates of misdiagnosis, avoidable readmissions and reduced participation in shared decision-making across the surgical continuum: history-taking, informed consent, post-operative instructions and complication recognition.
Dr. Ortega also named the provider side of the equation. Physician burnout from inadequate interpretation infrastructure is documented, not burnout from overwork but from the specific weight of knowing a patient is not receiving the care they need because communication has broken down. "There are a lot of workarounds that exist in this space," he said. "That tells you the problem is real and the system has not solved it."
A health system running on workarounds has already acknowledged the gap.
The workarounds are the evidence.
2. What Types of Dialect Errors Actually Occur in Spanish Medical Interpretation?
Dialect mismatch errors in Spanish medical interpretation fall into three types: semantic inversion (the same word carrying opposite meanings across regional variants), register mismatch (clinical vocabulary that does not map cleanly across dialects) and false familiarity (a clinician or interpreter assuming shared meaning because the language is nominally the same). Dr. Ortega documented all three in his own clinical practice before he had language to name them.
Dr. Ortega grew up in Brooklyn with Dominican immigrant parents and served as their informal interpreter from childhood. When he trained in Washington, D.C., he encountered the Central and South American communities that predominate there and discovered the gaps firsthand.
2.1 Semantic inversion: ahorita
In his New York Dominican household, ahorita meant later. He used it at the bedside: "I will be back ahorita." In D.C., patients were confused. To them, ahorita meant right now. Same word. Opposite clinical meaning. A patient expecting their surgeon back now behaves differently from one told to wait until later.
2.2 False familiarity: bicho
In Argentina, bicho is an affectionate term for a toddler. In Mexico, it means insect. In Puerto Rico and the Dominican Republic, it is slang for male genitalia. Place that word in a post-operative conversation mediated by an interpreter trained in a different regional variant and the communication gap is immediate.
He noted that dialect mismatches between patients and medical interpreters can reduce shared terminology to roughly 50% overlap. A conversation that appears to be happening may not be.
Most language access RFPs do not measure this. Contracts are evaluated on language coverage breadth, not regional dialect depth. A health system that cannot distinguish between Mexican Spanish and Cuban Spanish at the interpreter level is operating a general-Spanish program, not a dialect-aware one. That distinction matters most in post-operative instructions and informed consent, where precision is a clinical requirement. The 2026 NEJM Catalyst study co-authored by Dr. Ortega was designed, in part, because those gaps produce measurable consequences in surgical outcomes.
3. What Did the NEJM Catalyst Study Find About AI vs. Human Interpretation in Surgery?
The February 2026 NEJM Catalyst study enrolled 23 adult Spanish-speaking surgical patients at Brigham and Women's Hospital and asked what patients actually want from interpretation technology, and when.
The findings did not produce a winner. Patients did not favor any single modality. AI interpretation was preferred for speed, privacy and time-sensitive interactions. Human remote video interpretation was preferred for emotionally complex encounters: goals of care, bad news delivery, family meetings and informed consent. The study produced a hybrid interpretation framework that maps modality to encounter type rather than mandating one technology across every scenario.
Dr. Ortega named the high-stakes context directly: "It is helpful to have that person there with you serving as the interpreter and being able to read the room. What are the emotions? What's happening?" He also named the positive case for AI: patients reported an increased sense of autonomy when using AI tools because the technology is in their control. "It's in their pocket, it's in their hands, and they can guide the conversation."
The hybrid framework reflects something more specific than a research finding. It reflects how interpretation actually needs to work across a patient's full journey: a scheduling call before the visit, a telehealth pre-op consultation, the in-person surgical encounter, the discharge conversation and the follow-up. Each touchpoint carries a different level of emotional complexity, a different urgency and a different format requirement. A program designed for the encounter room alone leaves every other touchpoint unaddressed. What the study describes as patient preference is also, structurally, what an infrastructure-first approach requires: interpretation available at every point of contact, in the right modality for that moment, without the patient or provider having to negotiate access each time.
4. Why Does Interpretation Infrastructure Determine Whether Patients Trust Their Providers?
"We as clinicians have dropped the ball sometimes on making a good first impression," Dr. Ortega said. For LEP patients, having interpretation resources ready before the encounter, rather than scrambling during it, is a clinical signal. "Having an interpreter ready or having a resource or technology ready for them to use, but then also providing them with that resource throughout their entire care, lets them know that you want to hear what they have to say."
He then identified what is lost when clinical teams become transactional: "I truly believe that patient care happens in the small talk. When we work with an interpreter, we become very transactional. You don't tend to ask about the dog or the cat or who lives at home."
That small talk is how a clinician learns that a patient lives alone and needs home nursing rather than standard discharge. It is how readmission risk surfaces before it becomes a readmission. When interpretation is a logistics workaround rather than an embedded clinical function, those questions stop getting asked and the social determinants that shape outcomes disappear from the record.
No Barrier covers 295+ language access options and is built to support interpretation at every touchpoint of the patient journey. Reach out to understand the platform we offer.
5. What Is CMS Currently Requiring for AI Medical Interpretation, and Where Is Policy Headed?
CMS currently requires that AI-translated clinical content be audited by a qualified human linguist before use. Dr. Ortega described this as a reasonable starting point and named informed consent as the area where regulatory complexity will be hardest to resolve: a legal act, a clinical act and a communication act simultaneously.
His concern was with how the next rules get made. "We need the technology sector and the healthcare sector and the clinicians to work together on these policies from the very beginning. For everyone to be in the same room." Healthcare regulation has historically been reactive. The programs being built today, with documented human oversight architecture and modality-mapped routing, will become the reference architecture when formal standards crystallize. Building to that standard now avoids a retrofit later.
6. The Bottom Line on Language Access and Dialect in Surgery
The gap is not in intention. It is in infrastructure: scheduling calls handled without interpretation, telehealth visits where the patient navigates alone, discharge instructions delivered in a language the patient reads at a lower level than they speak. Every touchpoint in the patient journey carries a language access requirement. Most programs cover one or two of them.
"It is vital to provide optimal care to be able to communicate a diagnosis, a treatment plan and what to do moving forward," Dr. Ortega said. "When you cannot do that, everything else suffers."
Ahorita can mean later or right now, depending on where someone grew up. In a surgical corridor that is not a linguistic curiosity. It is a clinical variable. And it is one data point in a patient journey that has dozens of them.
.svg){kind=small}