Take the output and convert it into a list of dicts. Each entry is a sentence. Each key is a pair of span indices for that sentence,

Take the output and convert it into a list of dicts. Each entry is a sentence. Each key is a pair of span indices for that sentence,

A list of sentences (a document) to json.

Parameters ---------- tag: `bs4.element.Tag` The <div/> element of a xml document produced by Grobid. Returns ------

Given an entity from the lex attribute of cons, and the list of possible constituents (with their character spans), return the list of character spans

Take data SUTD-formatted documents and convert to our JSON format.

Format the label fields, making the following changes: 1. Span indices should be with respect to sentence, not document. 2. Return dicts

Extracts annotations from a sentence annotation

Add coref to json, filtering to only keep coref roots and `coref_types`.

Get list of files that had random off-by-1-errors and will be excluded.

Return the index of the sentence that the span is part of.

Returns the list of tokens from a sentence This method oversplits the tokens, so that, as much as possible, all entities can be composed by j

Get unique NER labels.

Returns a dict whose keys are spans, and values are the ID of the cluster of which the span is a member.

Need to match the ner data and the coref data by sentence. Create dictionaries where the keys are the document ID's and the values are the

Add coref field to json, one fold.