When a patient and clinician do not share a common language, a medical interpreter can appear to be doing something straightforward: listening to one person and conveying the same words in another language.
But in a medical setting, interpretation is rarely just a matter of converting sentences.
A trained medical interpreter may notice that a patient does not understand a diagnosis, recognize hesitation that a physician has missed, explain a cultural misunderstanding to the clinical team, preserve the emotional meaning of a patient’s words and help prevent a communication problem from becoming a medical error.
New research is highlighting just how important those less visible responsibilities can be. A 2026 study published in the Journal of General Internal Medicine identified 11 safety-related behaviors performed by medical interpreters, organized around the resilience-engineering concepts of responding, monitoring, learning and anticipating. The researchers concluded that today’s artificial-intelligence interpretation tools do not yet reproduce these broader human functions.
The findings arrive at a critical moment. AI-powered translation is rapidly entering healthcare, promising faster service, lower costs and greater availability. But the central question is no longer simply whether AI can translate a sentence accurately.
The more important question is:
Can an AI system recognize when a translation is medically dangerous, culturally inappropriate or misunderstood by the patient?
For now, there is considerable evidence that it cannot reliably do so.
Medical Interpretation Is a Patient-Safety Function
Language barriers can have serious consequences in healthcare.
Patients who prefer a language other than English can face greater risks of misunderstanding diagnoses, treatment instructions and follow-up requirements. Research has linked language barriers with problems including misdiagnosis, poorer comprehension, lower medication adherence and reduced participation in shared decision-making.
That makes interpretation fundamentally different from ordinary translation.
If an automated system translates a travel website incorrectly, the consequences may be inconvenient.
If it incorrectly translates a medication instruction, allergy, symptom or consent discussion, the consequences can be much more serious.
A medical interpreter therefore has to think about more than linguistic equivalence. The interpreter is part of a communication system in which misunderstandings can affect diagnosis and treatment.
The 11 Safety Behaviors Identified by Researchers
The new study involved 17 current or former certified medical interpreters with five to 24 years of experience. Participants had experience with in-person, video and telephone interpretation. The researchers used resilience engineering to examine how interpreters help clinical encounters continue safely despite unexpected problems.
The interpreters described 11 behaviors across four broad capabilities.
1. Responding
Interpreters help manage the flow of conversation.
Medical communication can become difficult when doctors use long explanations, patients give complicated answers or multiple people speak at once. Interpreters may encourage participants to speak in shorter segments so information can be conveyed accurately.
They can also preserve tone and emotion.
This is particularly important when patients are frightened, embarrassed, angry or uncertain. A technically accurate translation that removes emotional meaning can still produce an incomplete clinical picture.
Interpreters may also strengthen a patient’s voice by ensuring that the patient’s concerns are fully communicated rather than reduced to a simplified summary.
2. Monitoring
Interpreters continuously observe what is happening around them.
They may notice that a patient appears confused even though the patient has said they understand. They may detect tension between members of a medical team or recognize that a technology being used during the encounter is interfering with communication.
One particularly important observation from the study was that patients sometimes told clinicians they had no questions but later asked the interpreter about important medical information.
That is a warning sign.
A translation engine hears words.
A human interpreter can notice the gap between what a patient says and what the patient appears to understand.
3. Learning From Culture and Experience
Culture can influence how people describe symptoms, understand illness and interact with healthcare professionals.
A patient may have a different understanding of what causes disease, how much information a doctor should provide or when it is appropriate to question medical authority.
These differences do not mean that every patient from a particular culture behaves the same way. But trained interpreters can recognize potential cultural misunderstandings and help clinicians investigate them rather than making assumptions.
The researchers found that interpreters sometimes drew on cultural knowledge to identify information that might otherwise remain undisclosed.
That can be especially important in sensitive areas such as reproductive health, mental health, sexual health, domestic violence or chronic disease.
4. Anticipating Problems Before They Happen
Perhaps the most sophisticated role of a medical interpreter is anticipating problems.
An experienced interpreter may recognize that a patient is unlikely to return for follow-up, for example, because the patient interprets the healthcare system differently from the clinician.
The study described situations in which cultural beliefs could influence whether a patient returned for additional testing or treatment. An interpreter who recognizes that possibility can alert the clinical team and encourage clearer communication before the patient disappears from care.
This is fundamentally different from translation.
It involves prediction, contextual judgment and knowledge of how communication affects behavior.
Why AI Translation Is Still Valuable
None of this means that AI translation is useless.
Quite the opposite.
AI could provide significant benefits to healthcare systems if deployed carefully.
AI tools can potentially provide rapid language assistance when an interpreter is temporarily unavailable. They can assist with routine communication, translate lower-risk written information and potentially expand access in settings where professional interpreters are difficult to obtain.
A 2026 research agenda in npj Digital Medicine argues that AI language-access services deserve serious investigation precisely because they could expand access, particularly in under-resourced settings and for less common languages.
The problem arises when assistance becomes substitution.
An AI system can be available 24 hours a day.
That does not mean it understands what is happening in the room.
Translation Accuracy Is Not the Same as Clinical Safety
One of the biggest mistakes healthcare organizations could make is evaluating AI interpretation solely through linguistic accuracy.
A system might translate individual sentences impressively while still failing at the overall clinical task.
Consider a patient who says:
“I understand.”
A translation system can correctly convert those words.
But a human interpreter may have noticed that the patient looked confused, hesitated before answering and then quietly asked the interpreter what the physician meant.
The literal translation was correct.
The communication was not successful.
This distinction is at the center of the emerging debate over AI medical interpretation.
Researchers have warned that the evidence base for patient-facing AI interpretation remains limited. A 2026 npj Digital Medicine review notes that relatively little research has examined whether AI interpretation improves comprehension, affects patient trust or changes disparities in care.
High-Stakes Conversations Are Especially Difficult
AI translation may be more appropriate for simple, low-risk interactions than for complex medical conversations.
Routine directions such as where to find a laboratory or how to locate a clinic may be relatively straightforward.
But consider:
- Receiving a cancer diagnosis
- Discussing an operation
- Explaining a serious prognosis
- Obtaining informed consent
- Discussing end-of-life decisions
- Explaining psychiatric treatment
- Discussing pregnancy complications
- Taking a detailed medication history
- Reporting abuse or domestic violence
These situations require more than word substitution.
They require emotional sensitivity, precise terminology, confidentiality, cultural awareness and confirmation that the patient actually understands the information.
A 2026 review found that AI interpretation tools perform better in simpler interactions, while more complex clinical situations remain insufficiently evaluated. Another analysis found that approximately 5% of discharge-instruction translations examined contained at least one error with potential for clinically significant or life-threatening harm.
That does not mean every AI translation is dangerous.
It means that healthcare organizations cannot assume that a generally impressive translation system is automatically safe enough for every clinical circumstance.
Low-Resource Languages Create Another Problem
AI performance is not necessarily equal across languages.
Major commercial AI systems tend to have much more training data available for widely spoken languages than for many regional, minority and low-resource languages.
The result can be uneven quality.
An AI system might perform extremely well when translating between English and a heavily represented language while struggling with regional dialects, uncommon languages, code-switching or culturally specific expressions.
Researchers have also identified concerns about AI interpretation for low-resource languages, where inaccuracies can compound existing healthcare inequalities.
This creates an uncomfortable possibility: a technology introduced to improve healthcare access could inadvertently provide poorer communication to communities that already experience disadvantages.
Cultural Interpretation Cannot Be Reduced to a Dictionary
Medical language is filled with expressions that carry cultural meaning.
A patient may describe symptoms through metaphor. They may use a culturally specific concept of illness. They may avoid directly discussing a sensitive subject.
A literal translation may preserve the words while losing the meaning.
Human interpreters can ask for clarification within the boundaries of their professional role and communicate relevant cultural context to the healthcare team.
This is often called cultural brokering.
The interpreter does not simply act as a dictionary. The interpreter helps two people understand each other when their assumptions about illness, authority, family roles and communication may differ.
AI systems are increasingly capable of identifying context, but reliable cultural mediation in high-stakes healthcare remains an unsolved problem.
Confidentiality and Privacy Matter Too
Healthcare interpretation involves extremely sensitive information.
Patients may discuss diagnoses, medications, reproductive health, mental health, family relationships, financial circumstances or traumatic experiences.
Introducing AI into such conversations raises questions about what happens to the audio, transcripts and other information used by the system.
Privacy therefore cannot be treated as an afterthought.
Healthcare organizations need to know:
- Where the data is processed
- Whether conversations are stored
- How long information is retained
- Whether recordings are used to improve models
- Who can access transcripts
- How vendors protect patient information
- Whether data can be used for purposes beyond interpretation
These questions become even more important when AI systems are supplied by outside technology companies.
Privacy concerns have been highlighted alongside accuracy and equity as major ethical issues in AI-mediated medical interpreting.
U.S. Language-Access Rules Already Recognize the Importance of Qualified Interpreters
In the United States, language access is not simply a matter of customer service.
Section 1557 of the Affordable Care Act establishes protections against discrimination in covered health programs and requires meaningful access for individuals with limited English proficiency. HHS guidance says language assistance should be accurate and timely and protect privacy and independent decision-making.
HHS also describes a qualified interpreter as someone who can interpret effectively, accurately and impartially, including specialized vocabulary, while preserving the tone, sentiment and emotional level of the original statement and following accepted confidentiality principles.
That definition is revealing.
It describes interpretation as substantially more than replacing words in one language with words in another.
The National CLAS Standards likewise call for language assistance from competent individuals and emphasize culturally and linguistically appropriate care.
Why Family Members Are Not Always a Good Substitute
When a professional interpreter is unavailable, healthcare workers sometimes turn to family members or friends.
That can create serious problems.
Family members may unintentionally filter information, misunderstand medical terminology or decide that certain details are too embarrassing to repeat.
A patient may also be unwilling to disclose sensitive information when a relative is interpreting.
This is one reason professional interpretation exists in the first place.
A qualified interpreter is expected to provide accurate, impartial communication while protecting confidentiality and the patient’s ability to make independent decisions.
AI Could Become a Medical Interpreter’s Assistant
The most realistic future may not be humans versus machines.
It may be humans with machines.
AI could potentially handle some of the mechanical workload while trained interpreters concentrate on the tasks that require human judgment.
For example, AI might:
- Generate preliminary transcripts
- Identify unfamiliar medical terminology
- Assist with routine written translation
- Provide language support while a professional interpreter connects
- Help organize multilingual patient information
- Flag phrases that require clarification
- Support interpreters working across multiple languages
The human interpreter could then remain responsible for high-stakes communication, cultural mediation, emotional context and safety monitoring.
The researchers behind the new study specifically suggest considering AI as a teammate rather than automatically treating it as a replacement.
That approach could offer the efficiency advantages of AI without eliminating the human safety layer.
Hospitals Need Clear Rules for When AI Is Appropriate
If healthcare organizations adopt AI interpretation, they will need more than a software license.
They will need policies.
A sensible framework could classify communication into different risk levels.
Low-risk communication
AI might be appropriate, with safeguards, for tasks such as:
- General hospital directions
- Basic appointment information
- Non-sensitive educational materials
- Routine administrative information
Moderate-risk communication
Human review could become important for:
- Medication instructions
- Detailed medical histories
- Follow-up instructions
- Chronic disease education
- Pre-procedure information
High-risk communication
Qualified human interpreters should remain central to:
- Informed consent
- Serious diagnoses
- Complex treatment decisions
- Emergency care
- End-of-life discussions
- Psychiatric crises
- Sensitive reproductive or sexual-health discussions
- Abuse and safeguarding situations
The exact boundaries would need to be established through evidence, regulation and clinical experience.
Patients Should Have a Voice in the Technology Debate
Another weakness in the current evidence base is the lack of patient-centered research.
The 2026 npj Digital Medicine paper emphasizes that research into patients’ experiences with AI interpretation is still limited. Researchers need to understand not only whether a system produces accurate translations but whether patients trust it, understand the information and feel comfortable using it.
A technically accurate system that patients distrust may fail.
Likewise, a system that saves money but makes patients less willing to disclose sensitive information could ultimately make healthcare worse.
The people most affected by language barriers should therefore have a meaningful role in deciding how AI interpretation is designed and deployed.
Clinicians Also Need Training
AI cannot solve a communication problem if clinicians do not know how to work with interpreters or AI systems.
Doctors and nurses should understand how to:
- Speak directly to the patient
- Use short, interpretable segments
- Avoid unnecessary jargon
- Allow time for interpretation
- Confirm patient understanding
- Recognize signs of confusion
- Give interpreters opportunities to raise concerns
- Understand the limitations of AI translation
The new research highlights an especially practical idea: brief pre-appointment discussions can explicitly invite interpreters to identify safety concerns.
That simple step could strengthen the entire clinical team.
The Future of Medical Interpretation Will Probably Be Hybrid
The debate is often framed as a choice between human interpreters and artificial intelligence.
That may be the wrong question.
AI will almost certainly continue improving. Speech recognition, machine translation, multilingual language models and real-time voice technologies will become increasingly capable.
Some tasks that currently require human effort will probably become automated.
But healthcare communication has another dimension that is harder to automate: human judgment under uncertainty.
The strongest future model may therefore combine both technologies.
AI can provide speed, scalability and assistance.
Human interpreters can provide judgment, cultural awareness, emotional sensitivity, advocacy and safety monitoring.
Such a model would treat AI as an additional layer of healthcare infrastructure rather than as a shortcut around professional language services.
The Real Measure of AI Translation Should Be Patient Understanding
The ultimate goal of interpretation is not linguistic elegance.
It is safe communication.
A successful encounter occurs when the patient understands what is happening, can communicate important information, can ask questions, can make informed decisions and knows what to do next.
That means healthcare organizations should not judge AI systems only by benchmark translation scores.
They should ask:
Did the patient understand the diagnosis?
Did the patient understand the medication instructions?
Did the patient feel able to ask questions?
Did the clinician receive the information needed to make a safe decision?
Did the system preserve privacy and dignity?
Those measures are much closer to what medical interpretation is actually supposed to accomplish.
Conclusion: AI Can Translate Words, but Healthcare Requires Communication
The new research does not suggest that AI has no role in medical interpretation.
It suggests something more important: translation is only one part of the job.
Medical interpreters monitor conversations, identify confusion, preserve emotion, navigate cultural differences, recognize potential safety threats and sometimes anticipate problems before they occur.
Those responsibilities are easy to overlook precisely because much of the work happens quietly.
As healthcare organizations search for ways to reduce costs and expand language access, replacing trained interpreters with AI may appear attractive. But if the technology is judged only by whether it produces a plausible translation, hospitals risk overlooking the human safety functions that professional interpreters provide.
The better strategy is likely to use AI where it genuinely helps while retaining human expertise where misunderstanding could cause serious harm.
Medical interpretation is not simply about converting language.
It is about helping two human beings understand each other when the consequences of misunderstanding can be enormous.
5 Most Commonly Asked Questions
1. Can AI replace medical interpreters?
Not reliably in high-stakes healthcare today. AI can assist with certain translation tasks, particularly simple or lower-risk communication, but current evidence does not demonstrate that it can consistently reproduce the cultural, emotional, safety-monitoring and anticipatory functions of trained medical interpreters.
2. What does a medical interpreter do besides translate?
Medical interpreters can manage communication flow, preserve tone and emotion, monitor patient understanding, identify cultural misunderstandings, help surface information patients may otherwise withhold and alert clinicians to potential safety problems. Research has identified 11 such safety-related behaviors.
3. Is AI translation safe for medical use?
It can be useful in carefully selected circumstances, but it should not automatically be treated as equivalent to a qualified human interpreter. Research has identified accuracy problems and gaps in evidence, particularly for complex, high-stakes encounters and less-supported languages.
4. When should a human medical interpreter be used?
A qualified human interpreter is particularly important for high-stakes communication such as informed consent, serious diagnoses, complex treatment decisions, emergencies, end-of-life discussions and other sensitive encounters where errors or misunderstandings could cause significant harm. U.S. language-access guidance emphasizes accurate, timely and qualified interpretation for covered healthcare services.
5. What is the best future role for AI in medical interpretation?
A hybrid model may be the most practical approach. AI could assist with routine translation, transcription and language-access tasks while professional interpreters remain responsible for complex communication, cultural mediation, patient understanding and safety oversight. Researchers have specifically proposed viewing AI as a teammate rather than simply as a replacement for interpreters.
Sources Bioengineer

