Abstract: ‍ ‍Semantic communication is a new communication paradigm， which can improve the reliability of the communication process from the semantic level and solve the problem of limited communication bandwidth and spectrum resources. Aiming at the problem of semantic importance division in semantic communication， this paper proposes a layered semantic communication system based on dependency parsing. First， to get the dependency syntax relationship of the transmission sentence， this paper designs a dependency parsing model based on graph decoding， and extracts the dependency syntax tree of the transmission sentence. Secondly， based on the extracted dependency syntax tree， this paper proposes a semantic layering method and selectively transmits different layers’ semantic information according to the channel quality， so as to ensure the accurate transmission of key semantics. Moreover， this paper improves the semantic recovery ability of the receiver by introducing the ERNIE language model and the dependency syntactic relationship. The simulation results show that the semantic layering method proposed in this paper can effectively extract the key semantic information of the transmission sentence. Compared with the traditional communication system， the system proposed in this paper significantly improves the communication reliability under low signal-to-noise ratio.