BRAT Connector

BRAT is currently the only supported in-text annotation editor at EDS. BRAT annotations are in the standoff format. Consider the following document:

Le patient est admis pour une pneumopathie au coronavirus.
On lui prescrit du paracétamol.

It could be annotated as follows :

T1  Patient 4 11    patient
T2  Disease 31 58   pneumopathie au coronavirus
T3  Drug 79 90  paracétamol

The point of the BRAT connector is to go from the standoff annotation format to an annotated spaCy document :

import spacy
from edsnlp.connectors.brat import BratConnector

# Instantiate the connector
brat = BratConnector("path/to/brat")

# Instantiate the spacy pipeline
nlp = spacy.blank("fr")

# Convert all BRAT files to a list of documents
docs = brat.brat2docs(nlp)
doc = docs[0]

doc.ents
# Out: [patient, pneumopathie au coronavirus, paracétamol]

doc.ents[0].label_
# Out: Patient

The connector can also go the other way around, enabling pre-annotations and an ersatz of active learning.