Age as date?

[cnayan]

I took a text example from https://opennlp.apache.org
and tested with the sample code presented on landing page of https://spacy.io/

Pierre Vinken, 61 years old, will join the board as a nonexecutive director Nov. 29. Mr. Vinken is chairman of Elsevier N.V., the Dutch publishing group. Rudolph Agnew, 55 years old and former chairman of Consolidated Gold Fields PLC, was named a director of this British industrial conglomerate.

# pip install -U spacy
# python -m spacy download en_core_web_sm
import spacy

# Load English tokenizer, tagger, parser and NER
nlp = spacy.load("en_core_web_sm")

# Process whole documents
text = "Pierre Vinken, 61 years old, will join the board as a nonexecutive director Nov. 29. Mr. Vinken is chairman of Elsevier N.V., the Dutch publishing group. Rudolph Agnew, 55 years old and former chairman of Consolidated Gold Fields PLC, was named a director of this British industrial conglomerate."
doc = nlp(text)

# Find named entities, phrases and concepts
for entity in doc.ents:
    print(entity.label_, " => ", entity.text)

Output:

PERSON  =>  Pierre Vinken
DATE  =>  61 years old
DATE  =>  Nov. 29
PERSON  =>  Vinken
ORG  =>  Elsevier N.V.
NORP  =>  Dutch
PERSON  =>  Rudolph Agnew
DATE  =>  55 years old
ORG  =>  Consolidated Gold Fields PLC

Should age not be represented as age and not date? Or is there a way to tweak rules somehow?

https://corenlp.run/ is able to suggest "61 years old" as DURATION. Close, but still not age.

I am new to spaCy. Please help!

Thanks

[polm]

The spaCy NER models are not rule based, and you cannot "tweak" them. It's not like a function in code where you can go in and change a line.

The models have a bunch of numbers they use to make decisions, and we give them example documents and they learn the optimal numbers to reproduce the labels we put on the example documents.

For English we use OntoNotes for the training data. Looking at OntoNotes documents it looks like they use the DATE label for dates or periods of time, so that's how the spaCy models work. (in the pdf, look for "absolute or relative dates or periods")

If you want to label things as AGE, you can try training your own model, or using rules to re-label DATE entities as appropriate.

Also see #3052 about the model making errors.