Aiming at providing full-scale support to the study of language acquisition, this article takes the Dynamic Corpus on HSK Compositions as the key object and analyzes the systems of error classification of the established inter-language corpora. A series of problems have been found, which includes: focusing on syntax while sematic and pragmatic analysis being neglected; over-generalization of errors; ambiguity of the content which causes fuzzy boundaries between different error classes; ill principles of error annotation which result in some “micro-constructs” being covered by “macro-constructs”, etc.