Colombian Medical Semantic Translator
The problem
Colombian Spanish has a strong dialectal dimension: each region names illness its own way. “Rasquiña” on the coast, “baldado” in Bogotá, “estómago prendido” in Antioquia. None of those terms exist in international clinical coding systems, which only recognize technical vocabulary. The result is clinical information lost in translation, right at the moment the patient tries to describe what is wrong.
The gap isn’t one of intention — it’s infrastructure: electronic health record systems expect SNOMED CT, and people speak the way they speak.
What the system does
The system takes a colloquial expression and returns the correct SNOMED CT code, ready to be integrated into a clinical record. It’s not keyword search: it’s semantic translation. The system understands meaning in context, not just surface form.
Some examples from the validated dictionary:
| Expression | Clinical code | Region |
|---|---|---|
| “guayabo” | Headache | Bogotá |
| “rasquiña” | Pruritus | Coast |
| “estómago prendido” | Gastritis | Antioquia |
| “me pega en el pecho” | Chest pain | Coast |
| “baldado” | Fatigue | Bogotá |
| “zumba el oído” | Tinnitus | Bogotá |
Tech stack
- Language models fine-tuned for the medical domain in Colombian Spanish
- NLP pipeline for normalization and semantic mapping
- SNOMED CT as the standardized output layer
- Automatic capture queue for unresolved terms, with batch medical validation
Current status
In production. The dictionary currently holds 145 validated terms, 70 documented Colombian dialectalisms, and 38 terms under active medical review. Every expression the system doesn’t recognize goes into a research queue that feeds future versions. Refinement is ongoing, in collaboration with doctors from multiple regions of Colombia.
You can read the reflection on why this problem matters in this blog post.