Precision Communication: A Lexical Routing Approach to Community-Specific Clinical Translation for Spanish-Speaking Patients

Language barriers in emergency medicine represent a persistent source of health inequity for the 25 million Americans with limited English proficiency (LEP). Existing bilingual clinical communication tools treat Spanish as a monolithic language, failing to account for significant lexical variation across Spanish-speaking communities that can produce clinically consequential communication failures. A Dominican patient asked "¿Tiene diabetes?" may not recognize the term, because they use "azúcar" — a regional vocabulary difference that may contribute to misunderstandings and documentation inaccuracies. This paper describes a technical prototype — the Medical Voice Translator — implementing a precision communication framework that reconceptualizes clinical translation as a routing problem rather than a language generation problem. The system maps 53 clinical phrases to six distinct Spanish-speaking community variants — Dominican Republic, Puerto Rico, Mexico, Colombia, Cuba, and General Spanish — using a two-stage Jaccard similarity fuzzy matching algorithm and a bidirectional phrase consistency dictionary. Translation, speech recognition, and speech synthesis are performed in-process by default, so that no patient input need be transmitted to a third-party vendor. The implementation retains fallback paths to external services when a local component is unavailable; these are enumerated in Section 4.5. The architecture therefore supports fully local, self-hosted deployment, while the public demonstration instance described here runs on a hosted platform, retains the external fallback paths enabled, and is not intended for use with protected health information. Preliminary deployment analytics record 92 active users across 10+ US cities within 30 days of deployment. These counts are unfiltered and include automated traffic; they describe request volume against a public endpoint and do not constitute evidence of clinical adoption or effectiveness. A comparative evaluation across 53 preset phrases demonstrates that general-purpose translation tools produce zero community-specific regional variants, while the routing system retrieves the stored regional variant for all dictionary-covered phrases. For preset clinical communication, deterministic lexical routing avoids probabilistic generation for dictionary-covered phrases, reducing output variance for underrepresented Caribbean Spanish varieties.