This study proposes a multidimensional metric—Semantic Leakage Indicator (SLI) —to quantify meaning loss in AI-generated translations of Taiwanese into Mandarin. Drawing on a corpus of 26,046 Taiwanese sentences sourced from diverse spoken and written contexts, the analysis identifies five morpho-pragmatic features highly susceptible to semantic erosion: aspect markers, sentence-final particles, reduplication, kinship terms, and directional compounds. The SLI integrates entropy reduction, back-translation divergence, and semantic embedding shifts to detect untranslatable constructions that evade surface-level fluency metrics. Results reveal that even when lexical equivalents are present, high-SLI instances exhibit pragmatic distortion and affective dissonance. Cluster analysis delineates patterns of leakage across grammatical categories, while qualitative examples illustrate genre-specific vulnerabilities. The study advances translation studies by offering a scalable, language-sensitive diagnostic tool for evaluating AI performance on low-resource, morphologically rich languages. It contributes empirically to discussions on untranslatability, pragmatics, and meaning preservation in AI translation, with direct implications for algorithm design, language preservation, and corpus-informed evaluation in underrepresented languages.