Skip to content

Changelog#

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

3.7.0 - 2023-10-25#

Changed#

  • spaCy 3.6.x compatibility

Fixed#

3.6.1 - 2023-08-21#

Changed#

  • Fine-tuned models

3.6.0 - 2023-08-03#

Changed#

  • Updated dependencies
  • Updated resource list in model description
  • Pinned NerKor version

3.5.2 - 2023-05-19#

Changed#

  • Updated dependencies
  • Removed lemma smoothing

3.5.1 - 2023-03-22#

Changed#

  • Using the backported EditTreeLemmatizer for speedup inference explosion/spaCy#12017
  • Updated dependencies

3.5.0 - 2023-01-31#

Changed#

  • spaCy 3.5.x compatibility

3.4.4 - 2023-11-10#

Fixed#

  • Fixed dependencies of lemmatizer components

3.4.3 - 2022-10-25#

Fixed#

  • Improved hyperparameters on dependency parsing

3.4.2 - 2022-09-30#

Changed#

  • Improved NER: using beam_ner with update_prob=1.0
  • Using the r2.10 version on UD corpus (GitHub Changelog)
  • Using the v3 version on Szeged-Corpus (GitHub Changelog)
  • Improved lemmatization
    • Replaced spacy's trainable_lemmatizer with our own port of it (trainable_lemmatizer_v2) which improves on the lemmatization of uppercase tokens
    • Added lemma smoothing step for improving on common errors of the trainable lemmatizer

3.4.1 - 2022-08-12#

Changed#

  • Improved lookup lemmatizer: bugfix and morph. feats usage for indexing lemma

3.4.0 - 2022-07-29#

Changed#

  • spaCy 3.4.x compatibility
  • improved tagging performance

3.3.1 - 2022-05-11#

Added#

  • Added lookup lemmatizer before edit tree lemmatizer in the pipeline
  • Added lemma smoother after edit tree lemmatizer in the pipeline

3.3.0 - 2022-05-11#

Changed#

  • Replaced Lemmy lemmatizer w/ edit tree lemmatizer

3.2.2 - 2022-04-25#

Changed#

  • Replaced static word vectors w/ char n-gram based floret ones
  • Added multistep training to mitigate non-deterministic training behaviour

3.2.1 - 2022-05-06#

Changed#

  • Using the Szarvas-Farkas split for SzegedNER
  • Learning lemmata w/o "+" characters
  • hunnerwiki is no longer used to train the NER
  • Started using spacy's model numbering convention

0.4.2 - 2022-04-05#

Fixed#

  • Better integration of the lemmatizer
  • Updated the project's documentation

0.4.1 - 2022-04-01#

Added#

Changed#

  • NER model is built on NerKor and SzegedNER
  • Improved lemmatization for numbers and sentence starting tokens
  • Improved lemmatization by using the whole Szeged Corpus
  • Improved PoS tagging by pretraining on a silver standard corpora
  • Improved Dependency parser by using pretraining on silver standard corpora
  • Improved sentence splitter by using the multitask neural model

Fixed#

  • Compatibility w/ Spacy 3.x

0.3.1 - 2019-10-03#

Fixed#

  • Compatibility w/ Spacy 2.2.x

0.3.0 - 2019-09-26#

Added#

  • Named Entity Recognition

0.2.0 - 2019-06-02#

Fixed#

  • Compatibility w/ Spacy 2.1.x

Added#

  • Minor sentence segmentation improvements
  • Minor improvements in PoS tagging

0.1.0 - 2019-01-04#

Added#

  • Lemmatization support using lemmy
  • Rule based lemmatizer
  • Multi-task CNN-based dependency parser

Changed#

0.0.1 - 2017-06-11#

Added#

  • Experimental support for spaCy 1.x
  • PoS Tagger model with word vectors trained on an unreleased automatically transcribed Szeged Korpusz version
Last update: January 3, 2024