Knowledge Graph Completion (KGC) aims to predict missing nodes or edges in a knowledge graph. Embedding-based methods effectively represent facts in a knowledge graph through entity embeddings. However, these methods often lack generalization ability when handling semantically rich entities due to the limitations of structural information. Pre-trained Language Models (PLMs) contain rich latent knowledge and can effectively model unseen entities through implicit representations. Based on this, we propose a novel KGC method, PCL-KGC, which combines pre-trained language models with contrastive learning. In this method, entities and relations are connected into sequences to capture their textual semantic information. Contrastive learning, along with negative sampling, is used to train the model, thereby enhancing its ability to represent positive and negative samples. Experimental results show that PCL-KGC achieves improvements of 6.5%, 10.3%, and 11.2% in Hits@K (k = 1, 3, 10) on the WN18RR dataset, while also demonstrating time efficiency. To promote further research and exploration, the model code is available at https://github.com/yw3l/PCL-KGC .