Book Indexing and Artificial Intelligence (Presentation Slides)

Book Indexing and Artificial Intelligence Elizabeth Bartmess

Readers and authors expect a book’s index to accurately reflect the book’s contents. An index that fails to do so is unethical: it misrepresents the author's intellectual property to readers and can fail to guide readers to the information they seek, breaking the trust between the reader and the book. For an index to reflect the book’s contents, it must be at minimum complete (providing access to all indexable information in the book), navigable (guiding the reader to subtopics and related topics via cross-references), and accurate (containing no false or invented information and reflecting the author’s perspective and terminology). We asked large language models (LLMs) to index out-of-copyright books and found they failed to meet these criteria of completeness, navigability, and accuracy. Attendees will learn about necessary criteria for an adequate index and will learn where LLMs fall short in attempting to meet those criteria.

Elizabeth Bartmess is the chair of the American Society for Indexing’s AI Committee. She is an award-winning freelance indexer specializing in back-of-book and embedded indexes for scholarly, trade, and technology and design books. She also develops software utilities for indexers. Her academic background is in research psychology and information science.