This　paper　proposed　an　interactive　query　mode　for　an　OWL-oriented　question-answering　system,　named　Agile,　to　refine　answers.　The　key　problem　for　interactive　query　modes　in　QAs　is　producing　natural　language　sub-questions.　This　paper　used　a　two-phases　semantic　mapping　method　to　produce　sub-questions.　The　first　phase　builds　RDF　triples　based　on　existing　OWL　knowledge　and　given　original　questions.　The　second　phase　maps　RDF　triples　into　natural　language　sub-questions　through　some　lingual　common　sense　and　pre-defined　rules.　To　validate　effect　of　interactive　queries,　we　have　conducted　some　experiments　involving　two　different　domain　of　knowledge　bases　and　accepted　some　promising　results.