v3.1.0 Lemmatizer

Pretrained neural lemmatizer available.

m6

Updating readme to show examples for lemmatization.

v2.0.4 PyTorch version stick and initial token mask bug fix

• Changing how the initial token mask works in order to avoid bugs
• Stuck the version of torch to 1.7.* (since we ship the model with code).

v2.0.3 Sub-word chunking bug fix

Added a bug fix in which we now chunk sentences which contain too many sub-words for the BERT-like model.

v2.0.2 bug fixes

Some bug fixes to initial release.

v2.0.0 BERT-like language model

This version includes:

• A major rewrite of most components of the system.
• API exposed to user mostly unaffected, but the changes do break it.
• A new model file is required to be able to run the tool.
• Better accuracy for tagging.
• Supports lemmatization, but the module is not included in the released model file on CLARIN.

v1.0.1 - Python library improvements

• Bug fix in tag_bulk function.
• Configured the logger using standard conventions.
• Improved developer documentation

Public release

The first public release of the ABLTagger.
This version uses

• pretrained word embeddings
• characters-to-token model
• DMII token lookup

This version achieves an accuracy of 95.59% on MIM-Gold (cross-validation)

DyNet ABLTagger

The original DyNet ABLTagger