Rama Ramakrishnan continues MIT's Hands-On Deep Learning course (15.773, Spring 2024) with a lecture on how machines represent words and text numerically. He covers stand-alone embeddings, the fixed vector representations produced by methods like word2vec, and contextual embeddings, which shift a word's representation depending on surrounding text, the innovation behind models like BERT. The lecture builds on prior sessions in the course's natural language processing sequence, walking through why raw text needs to become numbers before a neural network can use it, and how embedding spaces end up capturing semantic relationships between words. Ramakrishnan keeps the discussion grounded in practical deep learning workflow rather than pure theory, consistent with the course's hands-on framing for MIT Sloan students. Runtime runs about 78 minutes, delivered as a standard classroom lecture with slides and instructor narration.