In order to extract catchwords, new words and other kinds of words for language monitoring, this paper propose the temporal-spatial modeling, which seemed the sort of words as the representation of the different word's distribution in a uniform temporalspatial system. The steady and dynamic features of the different distribution are given. Therefore various kinds of words could be identified and extracted according to the same feature set. The results of the experiment on catchword extraction show the modeling is effective and robustness for language monitoring.