This study employs two multi-million-word corpora of comparable translated and non-translated/original English texts to examine the “source language shining-through” effect. Utilizing machine learning techniques to analyze 96 linguistic features, it compares translations from various source languages across multiple registers. The results reveal that (1) translated English consistently differs from original English, regardless of source language; (2) language status exerts a greater influence on the shining-through effect than typological distance; and (3) language status and typological distance interact competitively across registers. These findings advance our understanding of translation universals and provide theoretical guidance for translators seeking to adapt effectively to source language typology, language status, and register-specific demands.