Models

In this section, we cover the list of supported human languages and models that are available for download in Stanza, the performance of these models, as well as how you can contribute models you trained to the Stanza community.

Packages

Packages for download can be specified with the Pipeline. The default packages include the tokenizer, mwt if appropriate, POS, lemma, dependencies, NER, sentiment, and constituencies, depending on availability. The default_accurate package adds coref and uses transformers if we have trained transformer based models for that language.

Other packages available depend on the treebank used to train the models, with recent results listed on the performance page. When multiple UD treebanks had compatible annotation schemes, the default package uses combined models.

Table of contents

• Available Models & Languages
• Model Performance
• Past Performance 1.0
• Past Performance 1.3
• Combined models
• NER Models