Why does SpaCy change the entire dependency tree because of a capital letter?

user_11323873 0
Viewed 100

The problem:

spaCy will produce a different parse tree if a single word is capitalized.

Examples:

  • Ketone is negative, protein is negative, Leucocytes is negative
  • Ketone is negative, Protein is negative, Leucocytes is negative

When Protein is capitalized, spacy produces 2 dependency trees and when protein only one tree, see the attached photo.

enter image description here

Environment:

windows server 2019,spaCy:2.1.4,spaCy-model: en_core_web_md

Questions:

  • Why does SpaCy change the entire dependency tree because of a capital letter?
  • Is there any way to make this more consistent?
spacy
0 Answers
Related