Abstract: Creative achievements have been made in the research and development of China's Standards of English Language Ability (CSE), among which translation proficiency scales are of original significance, filling the gap in the language proficiency assessment systems both at home and abroad. Taking the English version of the translation proficiency scales as the research text, adopting text mining approach, the study analyzes the features of descriptors in the aforesaid scales by using the tools of word clusters, semantic co-occurrence network and correspondence analysis in the software KH Coder. The research findings are as follows:(i) The high-frequency word clusters highlight one standard:reproduction of both content and form, and emphasize two themes:language quality and communication field; (ii) The distribution of core words in the semantic co-occurrence network verifies the scientificity of translation ability constructs to a large extent; (iii) There is no significant difference between the descriptors of adjacent levels. Furthermore, the descriptors of some particular level are relatively scattered, thus the definition of the corresponding translation ability does not have distinguishing characteristics.