Book Sequels Decomposed as Geometric Operators in Sentence Embedding Space

Using all-mpnet-base-v2 paragraph embeddings from PG19, sequel transformations are modeled as displacement vectors decomposed via PCA into interpretable content axes. Across 13 author pairs from Project Gutenberg, sequels cluster into three types: formulaic (low-rank, displacement norm ~0.12), concentrated (single dominant axis), and compositional (many small axes).