While semantic communication is expected to bring unprecedented communication efficiency in comparison to classical communication, many challenges must be resolved to realize its potential. In this work, we provide a realistic semantic network dubbed seq2seq-SC, which is compatible to 5G NR and can work with generalized text dataset utilizing pre-trained language model. We also utilize a performance metric (SBERT) which can accurately measure semantic similarity and show that seq2seq-SC achieves superior performance while extracting semantically meaningful information.