quantmod dependency issue: maximum of

[dsenkyr]

How to reproduce the behaviour

doc = nlp("Each project should be supervised by a maximum of 2 teachers.")

token_a = doc[6]
print(f"{token_a} <-{token_a.dep_}- {token_a.head}")

token_maximum = doc[7]
print(f"{token_maximum} <-{token_maximum.dep_}- {token_maximum.head}")

token_of = doc[8]
print(f"{token_of} <-{token_of.dep_}- {token_of.head}")

Current output

a <-quantmod- 2
maximum <-quantmod- 2
of <-quantmod- 2

Expected output

Only the maximum token has the quantmod dependency.

Your Environment

• spaCy version: 3.4.1
• Platform: Windows-10-10.0.19044-SP0
• Python version: 3.10.6
• Pipelines: en_core_web_trf (3.4.0)

[polm]

Thanks for the report, that does look kind of weird! We will check to see if this is not a wider issue. For the record, with the 3.3.0 trf or the 3.4.0 large pipeline this is the output:

a <-det- maximum
maximum <-pobj- by
of <-prep- maximum

[adrianeboyd]

I checked and the training data is just inconsistent here. I'm not sure whether it's from original OntoNotes annotation or related to the dependency conversion, but that's why's it's inconsistent.