Bug: Contextual Matcher excluded terms

[0dey]

Description

When patterns to exclude from the context are contained in the pattern of the main matcher, entities are always discarded, even if it's not in the right window. This happens when we want to extract "ASA 5" but not "5-ASA" for instance. In this example, we would put a negative window on the excluded pattern but the contextual matcher will however discard the entity. This is the case when the pattern to exclude (here 5) is included in the main pattern as we are looking for an integer after ASA. If we would like to discard "other-ASA", the pipe works, as "other" is not in the main pattern.

To sum up, window seems to be counting from the end_char of the entity whereas in this case, we would expect to be counting from start_char.

How to reproduce the bug

import edsnlp
import edsnlp.pipes as eds

asa_pattern = r"\basa\b ?:? ?([1-5]|[A-Z]{1,3})"
exclude_asa_ttt=r"5"    
asa = dict(source="asa",
          regex=asa_pattern,
          regex_attr="NORM",
          exclude=dict(
                regex=exclude_asa_ttt,
                window=-5)
          )

nlp = edsnlp.blank("eds")
nlp.add_pipe(eds.sentences())
nlp.add_pipe(
    eds.contextual_matcher(patterns=[asa], label="asa")
)

text="ASA 5"
docs=nlp(text)
docs.ents

Your Environment

• EDS-NLP Version Used: 0.15

[percevalw]

Hi @0dey, thank you for your detailed issue!
If I understand your problem correctly, in the example above, "ASA 5" should match, but not "5 ASA 5" ?
Would a similar problem be to exclude "anti" in the surroundings of anchor terms such as "antibiotics" ?

[percevalw]

In any case, the #289 should solve your problem.

[percevalw]

Closing as #289 was merged

[0dey]

Yes exactly ! Thank you !