Pipes overview

EDS-NLP provides easy-to-use pipeline components (aka pipes).

Available components

You can add them to your pipeline by simply calling add_pipe, for instance:

import edsnlp

nlp = edsnlp.blank("eds")
nlp.add_pipe("eds.normalizer")
nlp.add_pipe("eds.sentences")
nlp.add_pipe("eds.tnm")

Basic architecture

Most components provided by EDS-NLP aim to qualify pre-extracted entities. To wit, the basic usage of the library:

1. Implement a normaliser (see eds.normalizer)
2. Add an entity recognition component (eg the simple but powerful eds.matcher component)
3. Add zero or more entity qualification components, such as eds.negation, eds.family or eds.hypothesis. These qualifiers typically help detect false-positives.

Extraction components

Extraction components (matchers, the date detector or NER components, for instance) keep their results to the doc.ents and doc.spans attributes directly.

By default, some components do not write their output to doc.ents, such as the eds.sections matcher. This is mainly due to the fact that, since doc.ents cannot contain overlapping entities, we filter spans and keep the largest one by default. Since sections usually cover large spans of text, storing them in ents would remove every other overlapping entities.

Entity tagging

Moreover, most components declare extensions, on the Doc, Span and/or Token objects.

These extensions are especially useful for qualifier components, but can also be used by other components to persist relevant information. For instance, the eds.dates component declares a span._.date extension to store a normalised version of each detected date.