How do embeddings evolve in contextual models like BERT?

Contextual embeddings change dynamically based on surrounding words.

Unlike static embeddings, models like BERT compute token representations using bidirectional transformers, allowing word meaning to shift based on context.