A Massive Blind Spot in Healthcare Language Access
The following is a guest article by Jeffrey J. Munks, Co-Founder of LanguageLine, and Jeff Burrow, Founder of 3Birds Strategy
AI-Assisted Machine Translation is Uncovering Conversations That Were Not Happening Before
Late one morning, a bilingual patient experience director stopped by the room of a patient whose preferred language was Spanish. Remembering the patient’s anxiety from the day before, she noticed that the patient was scheduled for surgery in a couple of hours. This check-in was simply to make sure she was comfortable and included a straightforward question: Was she ready for her surgery?
The patient’s eyes grew large. “What surgery?”
The director immediately ran to find the physician, who was adamant that the procedure had been explained in detail and that the patient fully understood what was going to happen. “She nodded her head in agreement and smiled the entire time.”
The physician was genuinely convinced that the patient understood. Yet she had not.
Now, there are obvious questions about what should have happened, including the appropriate use of a qualified medical interpreter. But there is another, less obvious question worth asking:
What if that second conversation had never happened?
That question points toward what may be one of the largest unmeasured gaps in healthcare language access.
The Conversations We Don’t See
Healthcare organizations have spent decades building increasingly sophisticated systems to provide language assistance to patients with limited English proficiency. We can measure interpreter encounters, minutes of use, languages requested, and cost. We can track whether an interpreter was available and how quickly that interpreter joined the conversation.
The one thing all of these data points have in common is that they exist only when someone initiates a conversation.
What they miss are the conversations that never begin.
How often does a nurse weigh the importance of a simple question or greeting against the time required to connect to an interpreter? How often does someone from patient transport, environmental services, or the dietary team simply perform a task and move on? How often does a leader skip an informal visit to the room? And how often does a patient or family member have a question, concern, or piece of information that is never expressed?
For an English-speaking patient, many of these interactions are so routine that we barely recognize them as communication events. From the parking lot to the cafeteria, emergency department, information desk, hallways, and even the gift shop, they happen everywhere. For a patient who does not speak English, however, missing these seemingly insignificant interactions can have a cumulative effect.
The Beryl Institute defines the patient experience as “the sum of all interactions, shaped by an organization’s culture, that influence patient perceptions across the continuum of care.” Not having one conversation may seem trivial. What about three? Ten? Twenty? At some point, what is not being said may become as consequential as what is.
Research suggests those numbers may not be as hypothetical as they sound. In one 24-hour observational study of general hospital wards, patients interacted with a median of 14 people across 23 contact episodes. Other observational research has also documented frequent, short interactions involving clinical staff, nonclinical staff, and visitors. Regrettably, these studies do not tell us how many contacts became meaningful conversations, with or without a language barrier, or how many possible conversations never began.
We know a great deal about the conversations for which an interpreter is requested. We know remarkably little about those for which no request is ever made. Early experience with AI-assisted machine translation (AI/MT) is beginning to give us some idea of how large that gap may be.
Measuring the Demand
In the first quarter of 2025, Augusta Health purchased 25 HIPAA-oriented AI translation devices. Although several use cases were considered and adopted, the primary goal was to support nurses and registration staff while also helping volunteers with wayfinding around the 255-bed regional hospital in southern Virginia. With relatively low linguistic diversity, there was little reason to expect extraordinary utilization.
Three months later, the organization had logged more than 11,000 translations—an average of approximately 440 per device.
Soon afterward, one hospital within a large Midwestern academic health system deployed 100 devices and recorded approximately 30,000 uses in two months. At another hospital in that system, 12 devices assigned specifically to environmental services generated more than 6,000 uses.
Around the same time, a two-hospital regional system in the East, serving a highly diverse patient population, placed devices in nearly every department and recorded almost 200,000 uses in its first 30 days. It reached roughly the same volume the following month.
These figures come from operational deployment data, not controlled research. They do not tell us how many of these interactions were new, how many replaced another language resource, or whether they improved clinical outcomes.
What the figures do demonstrate is substantial demand. When immediately accessible translation became available, staff used it thousands of times, including in interactions rarely captured by traditional language-access metrics. Healthcare may have been underestimating the demand for communication all along.
We Have Seen This Before
For Jeffrey Munks, Co-Founder of LanguageLine, the accessibility question has a familiar ring. In the early 1980s, Munks was working as a police officer in San Jose when tens of thousands of Vietnamese, Laotian, Hmong, and Cambodian refugees arrived in the area. Seemingly overnight, police and local hospitals faced interactions in languages they were unequipped to manage. People could suffer grave consequences for reasons rooted in an inability to communicate.
As a Spanish-speaking bilingual officer, Munks understood the problem firsthand. When Southeast Asian refugees arrived, however, he was as unable to communicate as other officers and healthcare workers.
Interpreters were an invaluable resource, but they were in fixed locations while the need could arise almost anywhere. Munks’s solution came to him in five words: “Put it in the network.” By organizing qualified interpreters and connecting them through the telephone network, they could be available wherever and whenever they were needed. AT&T acquired the service in 1989, and the business subsequently grew by more than 4,000 percent over five years.
More than four decades later, healthcare faces a different version of the same problem. The resource exists. The question is whether it is accessible enough, quickly enough, and appropriately enough to reach all of the interactions in which language stands in the way.
The Next Access Problem
Network-based interpreting depends on qualified human interpreters, an essential but finite resource. As demand increases, adding interpreter capacity alone cannot address every circumstance involving a language barrier.
Then comes the issue at the center of this discussion: immediate accessibility. Ask a group of nurses how their language-access solution is working, and you will likely hear similar concerns: “It’s in another room.” “Nobody can find it.” “It’s never charged.” “I don’t have time to deal with it.”
A process that requires an employee to locate a cart or device, log in, navigate several steps, and then wait for an interpreter creates a natural deterrent to spontaneous interaction. The 6,000 environmental-services interactions described earlier are instructive. Those employees did not suddenly encounter thousands more people. They were given an immediate way to communicate with people they already encountered every day.
The accuracy and capabilities of AI/MT have improved dramatically. The opportunity is to create a tiered access model: immediately available AI/MT for appropriate, low-risk, and routine interactions, with rapid escalation to a qualified medical interpreter whenever the communication involves diagnosis, treatment, medication, informed consent, discharge instructions, legal rights, complex clinical information, uncertainty, patient preference, or any other circumstance in which accuracy and impartiality are essential.
The goal is not to replace qualified interpreters. It is to extend language access to far more of the interactions that shape the patient and employee experience.
Starting the Conversation
Patients who do not speak English need to communicate routine information dozens of times each day. They need to use the restroom. They want to order an item from the menu, request an extra blanket or ask to speak with a doctor or nurse. These interactions may seem insignificant individually, but together they are an important part of both the patient experience and the work of those caring for them.
For appropriate routine interactions, immediately accessible AI/MT can provide a way to begin communicating while preserving access to qualified medical interpreters. As one patient-experience leader put it, “Using the device justifies the use of an interpreter—not the other way around.”
The patient who asked “What surgery?” brings us back to where we started. Her misunderstanding was discovered because a bilingual employee happened to stop by and begin another conversation. The ability to communicate should not depend on a chance encounter. Imagine giving nurses, transporters, environmental-services staff, dietary teams, and others thousands of additional opportunities to help a patient feel seen and heard.
The value may be found in one conversation. At other times, it may emerge through the cumulative effect of more opportunities to communicate, building an environment of trust. With that trust comes a chance to uncover, good or bad, what might otherwise go unnoticed.
The challenge ahead is not simply improving the conversations we translate. It is making sure more of them happen in the first place.
Jeffrey J. Munks has been called the “Godfather of Network-Based Interpretation.” As a Co-Founder of LanguageLine, he helped create an industry that fundamentally changed how people overcome language barriers. Throughout his career, his work has focused on a deceptively simple problem: how to allow two people who do not share a language to communicate when and where they need to. He believes AI/MT represents the greatest leap forward in language access since the advent of network-based interpreting more than four decades ago. Jeff Munks can be reached at jeffmunks@gmail.com.
Jeff Burrow is the Founder of 3Birds Strategy and has spent the past several years working in healthcare and government to improve language access with AI-assisted machine translation. His work has included developing and supporting AI/MT strategies and deployments across health systems, with a particular focus on expanding communication beyond traditional interpreter encounters. After hearing some version of the “smile and nod” story dozens of times, he has made eliminating it a personal mission. Jeff Burrow can be reached at Jeff@3BirdsStrategy.com.
The following is a guest article by Jeffrey J. Munks, Co-Founder of LanguageLine, and Jeff Burrow, Founder of 3Birds Strategy
AI-Assisted Machine Translation is Uncovering Conversations That Were Not Happening Before
Late one morning, a bilingual patient experience director stopped by the room of a patient whose preferred language was Spanish. Remembering the patient’s anxiety from the day before, she noticed that the patient was scheduled for surgery in a couple of hours. This check-in was simply to make sure she was comfortable and included a straightforward question: Was she ready for her surgery?
The patient’s eyes grew large. “What surgery?”
The director immediately ran to find the physician, who was adamant that the procedure had been explained in detail and that the patient fully understood what was going to happen. “She nodded her head in agreement and smiled the entire time.”
The physician was genuinely convinced that the patient understood. Yet she had not.
Now, there are obvious questions about what should have happened, including the appropriate use of a qualified medical interpreter. But there is another, less obvious question worth asking:
What if that second conversation had never happened?
That question points toward what may be one of the largest unmeasured gaps in healthcare language access.
The Conversations We Don’t See
Healthcare organizations have spent decades building increasingly sophisticated systems to provide language assistance to patients with limited English proficiency. We can measure interpreter encounters, minutes of use, languages requested, and cost. We can track whether an interpreter was available and how quickly that interpreter joined the conversation.
The one thing all of these data points have in common is that they exist only when someone initiates a conversation.
What they miss are the conversations that never begin.
How often does a nurse weigh the importance of a simple question or greeting against the time required to connect to an interpreter? How often does someone from patient transport, environmental services, or the dietary team simply perform a task and move on? How often does a leader skip an informal visit to the room? And how often does a patient or family member have a question, concern, or piece of information that is never expressed?
For an English-speaking patient, many of these interactions are so routine that we barely recognize them as communication events. From the parking lot to the cafeteria, emergency department, information desk, hallways, and even the gift shop, they happen everywhere. For a patient who does not speak English, however, missing these seemingly insignificant interactions can have a cumulative effect.
The Beryl Institute defines the patient experience as “the sum of all interactions, shaped by an organization’s culture, that influence patient perceptions across the continuum of care.” Not having one conversation may seem trivial. What about three? Ten? Twenty? At some point, what is not being said may become as consequential as what is.
Research suggests those numbers may not be as hypothetical as they sound. In one 24-hour observational study of general hospital wards, patients interacted with a median of 14 people across 23 contact episodes. Other observational research has also documented frequent, short interactions involving clinical staff, nonclinical staff, and visitors. Regrettably, these studies do not tell us how many contacts became meaningful conversations, with or without a language barrier, or how many possible conversations never began.
We know a great deal about the conversations for which an interpreter is requested. We know remarkably little about those for which no request is ever made. Early experience with AI-assisted machine translation (AI/MT) is beginning to give us some idea of how large that gap may be.
Measuring the Demand
In the first quarter of 2025, Augusta Health purchased 25 HIPAA-oriented AI translation devices. Although several use cases were considered and adopted, the primary goal was to support nurses and registration staff while also helping volunteers with wayfinding around the 255-bed regional hospital in southern Virginia. With relatively low linguistic diversity, there was little reason to expect extraordinary utilization.
Three months later, the organization had logged more than 11,000 translations—an average of approximately 440 per device.
Soon afterward, one hospital within a large Midwestern academic health system deployed 100 devices and recorded approximately 30,000 uses in two months. At another hospital in that system, 12 devices assigned specifically to environmental services generated more than 6,000 uses.
Around the same time, a two-hospital regional system in the East, serving a highly diverse patient population, placed devices in nearly every department and recorded almost 200,000 uses in its first 30 days. It reached roughly the same volume the following month.
These figures come from operational deployment data, not controlled research. They do not tell us how many of these interactions were new, how many replaced another language resource, or whether they improved clinical outcomes.
What the figures do demonstrate is substantial demand. When immediately accessible translation became available, staff used it thousands of times, including in interactions rarely captured by traditional language-access metrics. Healthcare may have been underestimating the demand for communication all along.
We Have Seen This Before
For Jeffrey Munks, Co-Founder of LanguageLine, the accessibility question has a familiar ring. In the early 1980s, Munks was working as a police officer in San Jose when tens of thousands of Vietnamese, Laotian, Hmong, and Cambodian refugees arrived in the area. Seemingly overnight, police and local hospitals faced interactions in languages they were unequipped to manage. People could suffer grave consequences for reasons rooted in an inability to communicate.
As a Spanish-speaking bilingual officer, Munks understood the problem firsthand. When Southeast Asian refugees arrived, however, he was as unable to communicate as other officers and healthcare workers.
Interpreters were an invaluable resource, but they were in fixed locations while the need could arise almost anywhere. Munks’s solution came to him in five words: “Put it in the network.” By organizing qualified interpreters and connecting them through the telephone network, they could be available wherever and whenever they were needed. AT&T acquired the service in 1989, and the business subsequently grew by more than 4,000 percent over five years.
More than four decades later, healthcare faces a different version of the same problem. The resource exists. The question is whether it is accessible enough, quickly enough, and appropriately enough to reach all of the interactions in which language stands in the way.
The Next Access Problem
Network-based interpreting depends on qualified human interpreters, an essential but finite resource. As demand increases, adding interpreter capacity alone cannot address every circumstance involving a language barrier.
Then comes the issue at the center of this discussion: immediate accessibility. Ask a group of nurses how their language-access solution is working, and you will likely hear similar concerns: “It’s in another room.” “Nobody can find it.” “It’s never charged.” “I don’t have time to deal with it.”
A process that requires an employee to locate a cart or device, log in, navigate several steps, and then wait for an interpreter creates a natural deterrent to spontaneous interaction. The 6,000 environmental-services interactions described earlier are instructive. Those employees did not suddenly encounter thousands more people. They were given an immediate way to communicate with people they already encountered every day.
The accuracy and capabilities of AI/MT have improved dramatically. The opportunity is to create a tiered access model: immediately available AI/MT for appropriate, low-risk, and routine interactions, with rapid escalation to a qualified medical interpreter whenever the communication involves diagnosis, treatment, medication, informed consent, discharge instructions, legal rights, complex clinical information, uncertainty, patient preference, or any other circumstance in which accuracy and impartiality are essential.
The goal is not to replace qualified interpreters. It is to extend language access to far more of the interactions that shape the patient and employee experience.
Starting the Conversation
Patients who do not speak English need to communicate routine information dozens of times each day. They need to use the restroom. They want to order an item from the menu, request an extra blanket or ask to speak with a doctor or nurse. These interactions may seem insignificant individually, but together they are an important part of both the patient experience and the work of those caring for them.
For appropriate routine interactions, immediately accessible AI/MT can provide a way to begin communicating while preserving access to qualified medical interpreters. As one patient-experience leader put it, “Using the device justifies the use of an interpreter—not the other way around.”
The patient who asked “What surgery?” brings us back to where we started. Her misunderstanding was discovered because a bilingual employee happened to stop by and begin another conversation. The ability to communicate should not depend on a chance encounter. Imagine giving nurses, transporters, environmental-services staff, dietary teams, and others thousands of additional opportunities to help a patient feel seen and heard.
The value may be found in one conversation. At other times, it may emerge through the cumulative effect of more opportunities to communicate, building an environment of trust. With that trust comes a chance to uncover, good or bad, what might otherwise go unnoticed.
The challenge ahead is not simply improving the conversations we translate. It is making sure more of them happen in the first place.
Jeffrey J. Munks has been called the “Godfather of Network-Based Interpretation.” As a Co-Founder of LanguageLine, he helped create an industry that fundamentally changed how people overcome language barriers. Throughout his career, his work has focused on a deceptively simple problem: how to allow two people who do not share a language to communicate when and where they need to. He believes AI/MT represents the greatest leap forward in language access since the advent of network-based interpreting more than four decades ago. Jeff Munks can be reached at jeffmunks@gmail.com.
Jeff Burrow is the Founder of 3Birds Strategy and has spent the past several years working in healthcare and government to improve language access with AI-assisted machine translation. His work has included developing and supporting AI/MT strategies and deployments across health systems, with a particular focus on expanding communication beyond traditional interpreter encounters. After hearing some version of the “smile and nod” story dozens of times, he has made eliminating it a personal mission. Jeff Burrow can be reached at Jeff@3BirdsStrategy.com.
Go to Source
2026-09-14 19:30
Except for the headline, this story has not been edited by Healthcare Central is published from a syndicated feed.
