An error-correcting parsing method for context-sensitive languages is described, which finds all sentences ω's in a language with the minimum weighted Levenshtein distance WLD (ω, T) for input string T. The method is obtained by adding an error-correction mechanism to the Adachi-Kamata's parsing method for context-sensitive languages. By introducing the notion of a WLD-graph we can split the error-correction mechanism from the derivation of sentence i.e., the construction of derivation graphs. This makes the algorithm conceptually simple and easy to understand.

A loop detection is inherent in a context-sensitive parsing. A new detection method is presented for a top-down context-sensitive parsing using a grammar GE. This method requires only two input items, while the previous one requires three.