Semantic drift, metaphor vulnerability, and emotional meaning in health translation: A corpus-based comparison of English–French and English–Japanese subtitles

Health translation is often evaluated through terminological and propositional accuracy, yet health-related discourse also depends on emotional force, figurative framing, and register. This study asks how typologically different target languages differ in semantic drift, metaphor preservation, emotional intensity, and register in translated health-related discourse about illness, pain, suffering, trauma, and recovery. A custom extraction pipeline searched the TED2020 parallel corpus, which contains approximately 4000 TED Talk transcripts and more than 750,000 aligned subtitle lines, and identified 217 aligned English–French–Japanese segments meeting health-relevance and trilingual alignment criteria. Each target-language segment was manually coded against the English source using a four-dimensional framework: semantics, intensity, metaphor, and register. Japanese translations showed substantially more meaning change than French translations, with at least one coded shift in 67.7% of Japanese segments versus 41.9% of French segments, a statistically significant difference. Japanese translations also showed more multidimensional change, indicating that divergence often affected several layers of meaning at once. Metaphor was especially vulnerable: when the English source contained an active metaphor, 60.9% of Japanese translations and 37.7% of French translations modified it. A 25% validation subset, independently re-rated by two additional native-speaker reviewers, showed 96.5% exact agreement with the primary coding. The findings do not prove that typological distance causes this pattern, but they support it as a plausible explanation for why meaning is reconstructed differently across language systems. These findings support assessing translated health-related discourse through semantic, emotional, figurative, and register-sensitive criteria.