Cold never get classified as a noun in POS tagging.

[midnight-wonderer]

  const doc = nlp.readDoc('I have a cold.');
  doc.printTokens();

outputs

token      p-spaces   prefix  suffix  shape   case    nerHint type     normal/pos
———————————————————————————————————————————————————————————————————————————————————————
I                 0   I       I       X       2       0       word     i / PRON
have              1   ha      ave     xxxx    1       0       word     have / AUX
a                 1   a       a       x       1       0       word     a / DET
cold              1   co      old     xxxx    1       0       word     cold / ADJ
.                 0   .       .       .       0       0       punctuat . / PUNCT

In all cases I encountered, it never got classified as a noun, always an adjective.

P.S. wink-eng-lite-web-model v1.8.0

[rachnachakraborty]

Hi @midnight-wonderer

Thanks for sharing your observations!

POS tagging in models like wink-eng-lite-web-model employs probabilistic methods. While effective, these methods don't guarantee 100% accuracy. The quality of training data plays a crucial role in determining the probabilities.

We will explore possibilities for improvements.

Also, please note that .printTokens() is an undocumented, debugging-only API. For better reliability, it’s recommended to use documented methods from the API.

Best,
Rachna

[midnight-wonderer]

Thank you for the snappy response.

I kind of understand the machine learning bit, but the gist of my concern is that, from what I see, the model seems to have 0% accuracy on this word. (and potentially others)
It never got one right.

I don't actually know the details, but I just suspect the data tagging used for model training.