Pos tag bug when processing contiguous emojis

[charx7]

Problem

After loading the es_core_news_sm library for for an nlp pipeline, and trying read the POS of a document that contains contiguous emojis the detected pos_ tag changes depending on the number of contiguous emojis

How to reproduce the behaviour

import spacy
import es_core_news_sm

# init spacy nlp
nlp = es_core_news_sm.load()

s1 = "que tenga un excelente fin de año líder y que este y los que sigan sean mejores 👏🙏"
s2 = "que tenga un excelente fin de año líder y que este y los que sigan sean mejores 👏🙏🙏"

doc = nlp(s1)
for word in doc:
  print(f"The word: {word} has a pos tag {word.pos_}")
# >> observe that the POS tag of the 🙏 is PUNCT

doc = nlp(s2)
for word in doc:
  print(f"The word: {word} has a pos tag {word.pos_}")
# >> observe that the POS tag of the 🙏 is PROPN

We can observe that depending the number of contiguous 🙏 emojis the POS tag changes :(

Your Environment

• spaCy version: 2.2.4
• Platform: Linux-5.4.144+-x86_64-with-Ubuntu-18.04-bionic
• Python version: 3.7.12
• Models: en
• Environment Information: google-collab

[polm]

Please see #3052 - the statistical models make mistakes sometimes, and we can't address individual errors.

Also note that spaCy v2.2.4 is pretty old at this point - v3 has been out over a year, and I'd recommend you give it a look.

[charx7]

Thanks, I will bump the version of spaCy and try again