LitBank is an annotated dataset of 100 works of English-language fiction to support tasks in natural language processing and the computational humanities, described in more detail in the following publications:

• David Bamman, Sejal Popat and Sheng Shen (2019), "An Annotated Dataset of Literary Entities," NAACL 2019.

• Matthew Sims, Jong Ho Park and David Bamman (2019), "Literary Event Detection," ACL 2019.

• David Bamman, Olivia Lewke and Anya Mansoor (2020), "An Annotated Dataset of Coreference in English Literature", LREC.

LitBank currently contains annotations for entities, events, entity coreference, and quotation attribution in a sample of ~2,000 words from each of those texts, totaling 210,532 tokens.

LitBank is licensed under a Creative Commons Attribution 4.0 International License.

The entity annotation layer of LitBank covers six of the ACE 2005 categories in text:

• People (PER): Tom Sawyer, her daughter
• Facilities (FAC): the house, the kitchen
• Geo-political entities (GPE): London, the village
• Locations (LOC): the forest, the river
• Vehicles (VEH): the ship, the car
• Organizations (ORG): the army, the Church

The targets of annotation here include both named entities (e.g., Tom Sawyer) and common entities (the boy). These entities can be nested, as in the following:

For more, see: David Bamman, Sejal Popat and Sheng Shen, "An Annotated Dataset of Literary Entities," NAACL 2019.

The event layer in LitBank identifies events with asserted realis (depicted as actually taking place, with specific participants at a specific time) -- as opposed to events with other epistemic modalities (hypotheticals, future events, extradiegetic summaries by the narrator).

For more, see: Matt Sims, Jong Ho Park and David Bamman, "Literary Event Detection," ACL 2019.

The coreference layer in LitBank covers the six ACE entity categories outlined above (people, facilities, locations, geo-political entities, organizations and vehicles); while the entity tagging above only covers proper noun phrases (Tom Sawyer) and common noun phrases (the boy), the coref annotations also cover personal pronouns (he) as well.

We annotate three different categories of coreference phenomena -- coreference of identity (which links a mention in text to a discourse entity); copula (which links an attribute mention to another mention); and apposition (which links an appositive expression to another mention).

Phenomenon	Example	Source
Coreference	One may as well begin with [Helen]x's letters to [[her]x sister]y	Forster, Howard's End
Copula		Melville, Bartleby, the Scrivener
Apposition		Conrad, Heart of Darkness

Annotations largely follow OntoNotes guidelines (though singleton mentions are annotated here), with several departures for literary style (including the distinction between generic/specific mentions, near-identity and the revelation of identity). For more on the coreference criteria used in these annotations, see David Bamman, Olivia Lewke and Anya Mansoor (2020), "An Annotated Dataset of Coreference in English Literature", LREC.

The quotation layer in LitBank identifies all instances of direct speech in the text, attributed to its speaker.

This layer captures dialogue in quotation marks and other typographical markers (such as dashes in Joyce's Ulysses), and excludes strings wrapped in quotation marks that do not constitute dialogue, such as the use of scare quotes for emphatic use (for jargon, neologisms or irony), titles of works of art, and the mention of a term (as distinct from its use).

Speaker labels are identical to those in the coref/ section of LitBank, to enable linking the quotation and coreference layers.

For more on the quotation annotations, see this paper.

A trained tagger (which can be used to tag entities and events in new text) can be found in the tagger/ directory. (A model for coreference resolution is coming shortly.)

The corpus is drawn from the public domain texts on Project Gutenberg, and includes individual works of fiction (both novels and short stories) that include a mix of high literary style (e.g., Edith Wharton's Age of Innocence, James Joyce's Ulysses) and popular pulp fiction (e.g., H. Rider Haggard's King Solomon's Mines, Horatio Alger's Ragged Dick). We select approximately 2,000 words from each of 100 texts; the total annotated dataset contains 210,532 tokens.

The entity data is formatted in the original brat standoff annotation format and in the tab-separated layered format of https://github.com/meizhiju/layered-bilstm-crf.

We welcome contributions to LitBank in the form of annotations for new texts in the public domain and suggestions for new texts to include; please contact dbamman@berkeley.edu to get involved. For more information:

• See interannotator.com to explore the current entity annotations in LitBank through the brat annotation interface.
• Try annotating entities in O Pioneers! yourself (login by hovering over the "brat" icon in the upper right corner, using username "test" and password "annotate").

This work was supported by an Amazon Research Award ("Natural Language Processing for Literary Texts").