The spotting-based method is thought to be an effective fundamental techniques for understanding the natural chain of human speech, so we applied it to two real-world tasks. The first is to automatically extract the intent of speakers in various contexts such as meetings and unstructured private conversations. The second task is to retrieve speech data. It is difficult to deal with these two tasks by using simple word spotting. We believe connected word spotting, also called concept spotting, is a more effective method. This paper describes a series of preliminary experiments of keyword spotting using TV news and talk programs as raw material. We explain the necessity and the methodology of connected word spotting. The results of the experiments show the effectiveness of connected word spotting.