This article aims to determine which specific linguistic feature more effectively distinguishes translated texts (TTs) from non-translated texts (NTTs) in the corporate reporting genre by adopting representative and comparable corpora with the help of conditional inference trees (CIT) and random forests (RF). Based on the widely explored translation theory of translation universals (TUs) and ample empirical evidence in this area, we selected linguistic features that were proven effective in showing the differences between TTs and NTTs. Among the numerous features known to represent the four categories of TUs— simplification, explicitation, normalization, and leveling out—we selected ten factors to find out how effective they are in distinguishing TTs from NTTs. After encoding ten factors in two monolingual, specialized corpora that consist of a total of 58 English management forewords of sustainability reports including TTs and NTTs, we analyzed the encoded data with conditional inference trees (CIT) and random forests (RF). The results show that, unlike our expectations, only simplification plays a significant role in identifying TTs from NTTs. This contrasts sharply with previous studies on the Korean-English language pair. The results not only confirm the considerable impact of genre variations on TU behaviors but also present the effectiveness of CIT and RF in corpus-based translation studies.