The encoder processes the input sequence and encodes it into contextualized representations through stacked self-attention layers. In the article, self-attention is explored primarily within the encoder. Each encoder layer refines token representations based on the entire input. These outputs are then passed to the decoder in full Transformer models. Encoders highlight how self-attention builds deep, hierarchical understanding of input text. They form the basis of models like BERT. Knowing how encoders apply self-attention grounds the reader in the model's operation.