During my mother's six years of treatment, I learned that translation in healthcare is not a linguistic act—it's an information system. And sometimes, two systems collide.
My mother and I relied on interpreters during many appointments. Some were excellent, careful, and patient. Others were hurried, dismissive, or visibly annoyed. Over time, I noticed a deeper pattern:
The interpreter and I were functioning as two different information systems—with different priorities, constraints, and fidelities.
The interpreter prioritized efficiency and flow. I prioritized accuracy and representation. These goals often collided.
Interpreters, especially in busy oncology settings, must keep appointments moving. Their incentives are structural:
• complete translation quickly • keep the conversation aligned with the doctor’s pace • avoid long explanations • use familiar, simple English terms • reduce culturally specific expressions
This makes practical sense—clinics run on tight schedules. But it introduces **semantic compression**.
For example, my mother might describe a sensation with cultural metaphor or layered nuance. The interpreter often rephrased it into:
"She feels pain."
That phrase was not wrong. But it was incomplete—and incompleteness becomes inaccuracy over time.
When I interpreted, my priorities were different:
• preserve nuance • preserve metaphor • clarify uncertainty • represent emotional context • ensure my mother's voice remained whole
I was not neutral—I was deeply invested. My goal was fidelity, not speed.
This often meant challenging the interpreter:
“That is not what she meant.” “She did not say sharp pain—she said pressure that moves.” “This is not anxiety—this is a physical symptom she cannot describe easily in English.”
These corrections annoyed some interpreters, who saw them as disruptions. But they were essential to preserving meaning.
Interpreter-driven notes tended to be:
• shorter • simpler • less descriptive • less culturally informed
My interpretations produced:
• longer explanations • culturally contextualized symptoms • more precise descriptions • clearer timelines and histories
The same patient produced two different notes depending on who interpreted. In the EHR, this becomes:
Interpreter System → compressed notes Family System → expanded notes
When these notes become data, the variation appears as “noise.” But it is not noise—it is representational inequity.
I witnessed many moments where interpreters influenced what my mother felt allowed to say:
• impatience discouraged her from asking questions • tonal shifts made her feel “stupid” • interruptions cut off descriptions mid-sentence • minimized symptoms changed how clinicians responded
These small moments mattered. They shaped her care.
A rushed interpreter can unintentionally narrow what gets recorded. A sensitive interpreter can expand it.
When multilingual patients enter datasets as “inconsistent,” “low-quality,” or “high-missingness,” those labels often reflect:
• interpreter variability • documentation drift • cross-cultural mismatch • semantic compression
Not the patient.
This is why I am drawn to studying:
• multilingual EHR representation • interpreter-mediated data quality • documentation inequity • upstream variation before data becomes data
My mother’s care taught me that translation is not a side task—it is a structural factor in data quality.