Ben and I have been brainstorming things that could be done with AI tools once you have transcriptions – crowdsourced or HTR’d – of documents. There’s a LOT! I thought it would be fun to share that list with you, to spark ideas and possibilities. Reply and let me know what ideas you have and what we missed.
- Traditional metadata activities
- Language detection
- Summarization
- Assigning subject headings
- Detecting sender / recipient
- Detecting place of composition or place sent to
- Determining the place the document is about
- Detecting document type
- Traditional metadata data
- Create finding aids
- Associate with EAC-CPF
- Associate with SNAC records
- Associate with Library of Congress Subject Headings
- Linking to wikidata and other linked open data authorities
- Non-traditional metadata activities
- Extracting entities (including families, companies, or other non-person entities)
- Identifying relationships between mentioned entities
- Detecting slavery, conflict, lists of names, etc. (This is what I think of as “Answering questions about a page or document”)
- Emotional valences (positive or negative; strongly or weakly emotive) This would be fun to graph as a discovery interface for documents.
- Computer Vision-y features
- Classifying documents as text vs. handwritten
- Detecting handwriting on printed/typed pages
- Determining whether pages contain photos or diagrams
- Getting rid of “Queen Victoria’s Birthday” (or other text preprinted on diary pages)
- Identifying blank pages
- Identifying meaningful pages
- Long-form derivatives
- Text optimized for screen readers (modernized punctuation and spelling, expanded abbreviations)
- Translations
- Modernized spelling for full-text search.
- “Explain it Like I’m Five” versions (and other “junior reader” versions)
- Audio version of the text
- Images or video inspired by/to go along with the text
Yes, some of these are crazy – but many of them are not. And I think they’re all within reach in the coming years.