This paper introduces the error corpus of Korean learner English and mal rules to detect the errors. Based on the corpus, we classified 42 error types. Our criteria for error classification are more general in order to enhance agreement rate and decrease errors. For generating mal rule, we testified two different grammars. One is Context Free Grammar (CFG) from Penn Treebank. The other is the typed feature structure grammars based on the Head-Driven Phrase Structure Grammar (HPSG), using Natural Language ToolKit (Bird et al. 2009). We advanced grammatical formalism from CFG to HPSG since CFG needs abundant phrasal markers causing over-generation, structural ambiguity and complexity.