Semantic communication has become a popular research area due its high spectrum efficiency and error-correction performance. Some studies use deep learning to extract semantic features, which usually form end-to-end semantic communication systems and are hard to address the varying wireless environments. Therefore, the novel semantic-based coding methods and performance metrics have been investigated and the designed semantic systems consist of various modules as in the conventional communications but with improved functions. This article discusses recent achievements in the state-of-art semantic communications exploiting the conventional modules in wireless systems. We demonstrate through two examples that the traditional hybrid automatic repeat request and modulation methods can be redesigned for novel semantic coding and metrics to further improve the performance of wireless semantic communications. At the end of this article, some open issues are identified.