Content-based Models of Quotation: Datasets

data_content-based_models_of_quotation.zip contains two of the datasets (KJB-CA and LAT-EJC) from our paper Content-based Models of Quotation (EACL-21). Unfortunately, we are unable to share any data from the JSTOR Understanding Series (KJB-JA, SHAK-JA, ABL-JA). Please contact JSTOR to discuss getting access to this data.

If you use either of our datasets, please cite our paper:

@inproceedings{maclaughlin-smith-2021-content,
  author={Ansel MacLaughlin and David A. Smith},
  title={Content-based Models of Quotation},
  booktitle={EACL},
  year={2021}
}

kjb-ca.jsonl: King James Bible - Chronicling America:

• JSON Lines file (each line contains a JSON record)
• Each record contains information for a single verse and has the following fields:
  • text: the text of the verse (e.g. “In the beginning God created the heaven…”)
  • testament: Old Testament (OT) or New Testament (NT)
  • book: the corresponding book (e.g. Genesis)
  • chapter: the verse’s chapter in the book (e.g. 1)
  • verse: the verse number in the chapter (e.g. 4)
  • quote: number of times the corresponding verse was quoted in the Chronicling America collection
• This data is simply a processed version of the America’s Public Bible dataset

lat-ejc.jsonl: Latin Text - JSTOR Early Journal Content:

• JSON Lines file (each line contains a JSON record)
• Each record contains information for a single quote and has the following fields:
  • book: the URN (e.g. urn:cts:greekLit:tlg2042.tlg028.1st1K-lat1)
  • quoted_text: text of the quoted source passage
• How to use:
  • We are unable to share the full text of the original Latin works due to copyright restrictions.
  • The file latin-works-metadata.jsonl contains a list of section URNs for each Latin work.
  • Use the Perseus CTS API and the list of URNs to download the full text of each work with an associated quote (there are 329 works total).
  • For each quote in lat-ejc.jsonl, match the quoted string back to its original location in the full text.
  • Process each full text with the Stanza Model trained on UD Latin Perseus
  • Label each sentence in each full text Latin work with the number of quotes that overlap with it.