The paper introduces a deep learning-enabled semantic communication system (DeepSC) designed to enhance the transmission of text by maximizing system capacity and minimizing semantic errors. Inspired by advancements in natural language processing (NLP) and end-to-end (E2E) communication systems, DeepSC aims to recover the meaning of sentences rather than focusing on bit or symbol errors. The system leverages the Transformer architecture to extract semantic information from text, ensuring robustness to noise and channel distortion. Two loss functions, cross-entropy and mutual information, are used to optimize the transmitter and receiver, with a new metric, sentence similarity, introduced to evaluate performance at the semantic level. Transfer learning is employed to adapt DeepSC to different communication environments, accelerating model training. Extensive simulations demonstrate that DeepSC outperforms traditional communication systems, particularly in low signal-to-noise (SNR) regimes, by improving robustness and achieving better performance in semantic communication.The paper introduces a deep learning-enabled semantic communication system (DeepSC) designed to enhance the transmission of text by maximizing system capacity and minimizing semantic errors. Inspired by advancements in natural language processing (NLP) and end-to-end (E2E) communication systems, DeepSC aims to recover the meaning of sentences rather than focusing on bit or symbol errors. The system leverages the Transformer architecture to extract semantic information from text, ensuring robustness to noise and channel distortion. Two loss functions, cross-entropy and mutual information, are used to optimize the transmitter and receiver, with a new metric, sentence similarity, introduced to evaluate performance at the semantic level. Transfer learning is employed to adapt DeepSC to different communication environments, accelerating model training. Extensive simulations demonstrate that DeepSC outperforms traditional communication systems, particularly in low signal-to-noise (SNR) regimes, by improving robustness and achieving better performance in semantic communication.