All natural language interface systems are likely to receive inputs for which they are irregular, i.e., ungrammatical or extra-grammatical. So, the system must be able to respond to such ill-formed inputs by clearly indicating the reasons the given input could not be understood, so that the user will have precise information for retrying to type the input. If natural language communication to expert systems is to be understood and accepted naturally by others than computer specialists, this is an essential necessity. In my paper, I will outline 3 levels of ill-formedness in natural language interface systems and corresponding recovery strategies, and based on these strategies, I designed and implemented an experimental system RUl(Robust User Interface) as a tool for achieving an efficient natural language sentence processing in a user interface of an expert system. RUl, an interface for HAPS2 expert system, can detect and recover the ill-formedness in the input sentence at the lexical-level. Finally, I suggested further improvements for recognizing and processing of the ill-formed input in an expert system.