LAUSR

Standard digital communication techniques allow us to set aside the meaning of the messages to concentrate on the transmission of bits efficiently and reliably. However, with the integration of artificial intelligence into communications technology and the merging of communication and computation within devices, increasing evidence suggests that the semantic aspect of communication cannot be set aside. We propose a part-of-speech-based encoding strategy and context-based decoding strategies, in which various deep learning models are presented to learn the semantic and contextual features as background knowledge. With the background knowledge, our strategies can be applied to some non-jointly-designed communication scenarios with uncertainty. We compare the performances of two proposed decoding strategies, the deep learning models of which are different, to provide model-choice design guidelines in accordance with specific communication conditions. Further, we discuss the impact of several parameters on the performance of our strategies, such as the size of the context window and the size of the feature window. Simulation results indicate the effectiveness and the reliability of our strategies in terms of decreasing the number of bits used to transmit messages and increasing the semantic accuracy between transmitted messages and recovered messages.