DaCy: Sota spaCy pipeline for NER, POS, Dependency Parsing, Coreference Resolution and NED

[KennethEnevoldsen]

DaCy's large model utilises the dfm-encoder-large-v1 model for its latest release. This latest model have lead to notable improvements over the previous large model built using a the xlm-roberta-large.

This model:

• Half the errors rate in part-of-speech tagging
• Obtains new sota performance in dependency parsing from a LAS of 88.4 to a 90.8
• Notably improves lemmatisation from 84.91 to 0.95
• Obtains perfect sentence segmentation on the test set
• and much more (see more on the model card)