This repository contains pretrained models for Pie (A Framework for Joint Learning of Sequence Labeling Tasks).

More on Pie: https://github.com/emanjavacas/pie.

Models are arranged by language. TODO: add a json documentation file per model.

german-ren.model.tar: Lemmatizer pretrained on a subset of the Referenzkorpus Mittelniederdeutsch/Niederrheinisch: https://www.slm.uni-hamburg.de/ren.html

spanish-AnCora.model.tar: Lemmatizer pretrained on the AnCora corpus for Spanish (part of the Universal Dependencies)

french-Geste.model.tar: Lemmatizer pretrained on the Geste corpus

fro-poslemmes_cat-lemma-2019_01_22-02_34_11.tar: lemmatizer and POS-tagger trained on the Geste corpus, and other Old French data from the École des chartes.

Target task: lemma. 
Accuracy on test data
  lemma: 0.9383
  pos: 0.9473

fro-poslemmes_cat-lemma-2019_01_23-00_34_12: same as the previous one, but using pre-trained word embeddings from a large unlabelled corpus.

Target task: lemma. 
Accuracy on test data
  lemma: 0.9409
  pos: 0.9468

fro-poslemmes_cat-lemma-2019_01_24-00_05_57.tar: same as the previous one, but using convolutions (cnn) for the character embeddings.

Target task: lemma. 
Accuracy on test data
  lemma: 0.9462
  pos: 0.9509

model_fro_poslemmesmorph.tar: POS-tagger, lemmatizer and morphological analyzer trained on the Geste corpus

capitula.model.tar: Lemmatizer pretrained on a non-open source dataset of medieval latin

turkish-IMST.model.tar: Lemmatizer pretrained on the IMST corpus for Turkish (part of the Universal Dependencies)

lemma.config.json is an example config file for training a lemmatizer to reasonable good accuracy.

For more information check the repo at , but in short:

virtualenv env -p python3.7
source env/bin/activate
pip3 install -r requirements.txt