Is it possible to have entity linking and rel extraction in a single pipeline based on coreference results?

[tomasonjo]

I've developed a simple pipeline:

# Start with english model
nlp = spacy.load('en_core_web_sm')

# Add coreference resolution
nlp.add_pipe(
    "xx_coref", config={"chunk_size": 2500, "chunk_overlap": 2, "device": -1})


# Add opentapioca entity linking
nlp.add_pipe('opentapioca')

# Add Rebel relationship extraction
nlp.add_pipe("rebel", after="senter", config={
    'device':-1, # Number of the GPU, -1 if want to use CPU
    'model_name':'Babelscape/rebel-large'} # Model used, will default to 'Babelscape/rebel-large' if not given
    )


doc = nlp("Christian Drosten works in Germany. He likes to work for Google.")

print(doc._.resolved_text)

for span in doc.ents:
    print((span.text, span.kb_id_, span.label_, span._.description, span._.score))

for value, rel_dict in doc._.rel.items():
    print(f"{value}: {rel_dict}")

However, it seems that opentapioca and rebel steps don't use coreference resolution as their input. Is it possible to somehow fix this without introducing two separate pipelines?

[polm]

What is xx_coref? How are coreference relations annotated?

It looks like you're using OpenTapioca and Rebel, which are third-party modules for spaCy and not developed by us - you'd prohbably be off contacting their devs. I don't think it should be too hard to reflect your coreference relations for them, but not being familiar with them it's hard to say.