This paper describes a robust parsing algorithm for spoken language understanding. Comparing with the other work in robust parsing, we focus on building a parser that is robust to not only ill-formed spontaneous spoken language inputs but also under-specified grammars. Preliminary experiment results show that the parsing performance deteriorates more gracefully than another parser we have used when the grammar is more under-specified.