German lemmatizer confused by capitalization

[giopina]

I'm having issues with the lemmatizer for German (both in v3.0.6 and v2.3.2, using both de_core_news_lg and de_dep_news_trf).
Basically, the lemmatizer gets confused by the capitalization of the verb, and can't assign the right lemma to it (while the POS tagger is actually correct).

Code:

import spacy
nlp = spacy.load("de_core_news_lg")
doc = nlp('Meldet sich keiner')
for token in doc:
    print(token,token.lemma_,token.pos_,token.tag_)

Output:

Meldet Meldet VERB VVFIN
sich sich PRON PRF
keiner kein PRON PIS

(the lemma of meldet should be melden)

Is there some general solution/fix to this issue?

Originally posted by @giopina in #2668 (comment)

[adrianeboyd]

The current extremely simple lookup lemmatizer is just not very good, a closely related discussion: #8695 (comment). It doesn't know anything about POS, casing, or spelling variation.

We do have some good news to report, though: we have internal work-in-progress on a statistical lemmatizer that should be much, much better than the lookup lemmatizer, which was only meant to be a stopgap solution and has been the default for German for way too long at this point. With the new lemmatizer, the accuracy on TIGER is ~97%. Keep an eye out for an official announcement soon!