Wrong outputs from ru_core_news_sm model with hyphened words

[riZZZhik]

Hi, got strange outputs on lemmatizing Russian words with hyphen:

Initialization:

import spacy
nlp = spacy.load('ru_core_news_sm')

Test 1:

print([(x.lemma, x.lemma_) for x in nlp('по-любому')]

Output: [(12047934663327436226, 'по'), (9153284864653046197, '-'), (1520167342668405783, 'любому')]

Test 2:

print([(x.lemma, x.lemma_) for x in nlp('любому')]

Output: [(11783268970549905777, 'любой')]

Question:
Why the word 'любому' in second test lemmatized as 'любой' - what is correct, but in first as 'любому'?

[thomashacker]

Hello,
the lemmatizer (pymorphy2) depends on POS annotations from the tagger. I'd suggest checking the POS tags first, maybe it makes it more clear why the lemmatization breaks in your first example.

[riZZZhik]

Yeah, different POS tags, I will research about them and return if there would be any other questions.
Thank you very much!

[(x.pos_, x.lemma_) for x in nlp('по-любому')] # [('ADV', 'по'), ('ADV', '-'), ('ADV', 'любому')]
[(x.pos_, x.lemma_) for x in nlp('любому')]    # [('DET', 'любой')]