dane_plus / README.md
KennethEnevoldsen's picture
Update README.md
09388ce
metadata
annotations_creators:
  - expert-generated
  - machine-generated
language_creators:
  - found
language:
  - da
license:
  - cc-by-sa-4.0
multilinguality:
  - monolingual
size_categories:
  - 1K<n<10K
source_datasets:
  - dane
  - extended|other-Danish-Universal-Dependencies-treebank
  - DANSK
task_categories:
  - token-classification
task_ids:
  - named-entity-recognition
  - part-of-speech
paperswithcode_id: dane
pretty_name: DaNE+
dataset_info:
  features:
    - name: text
      dtype: string
    - name: ents
      list:
        - name: end
          dtype: int64
        - name: label
          dtype: string
        - name: start
          dtype: int64
    - name: sents
      list:
        - name: end
          dtype: int64
        - name: start
          dtype: int64
    - name: tokens
      list:
        - name: dep
          dtype: string
        - name: end
          dtype: int64
        - name: head
          dtype: int64
        - name: id
          dtype: int64
        - name: lemma
          dtype: string
        - name: morph
          dtype: string
        - name: pos
          dtype: string
        - name: start
          dtype: int64
        - name: tag
          dtype: string
  splits:
    - name: train
      num_bytes: 7886693
      num_examples: 4383
    - name: dev
      num_bytes: 1016350
      num_examples: 564
    - name: test
      num_bytes: 991137
      num_examples: 565
  download_size: 1627548
  dataset_size: 9894180

DaNE+

This is a version of DaNE, where the original NER labels have been updated to follow the ontonotes annotation scheme. The annotation process used the model trained on the Danish dataset DANSK for the first round of annotation and then all the discrepancies were manually reviewed and corrected by Kenneth C. Enevoldsen. A discrepancy include notably also includes newly added entities such as PRODUCT and WORK_OF_ART. Thus in practice a great deal of entities were manually reviews. If there was an uncertainty the annotation was left as it was.

The additional annotations (e.g. part-of-speech tags) stems from the Danish Dependency Treebank, however, if you wish to use these I would recommend using the latest version as this version here will likely become outdated over time.

Process of annotation

  1. Install the requirements:
--extra-index-url pip install prodigy -f https://{DOWNLOAD KEY}@download.prodi.gy
prodigy>=1.11.0,<2.0.0
  1. Create outline dataset
python annotate.py
  1. Review and correction annotation using prodigy: Add datasets to prodigy
prodigy db-in dane reference.jsonl
prodigy db-in dane_plus_mdl_pred predictions.jsonl

Run review using prodigy:

prodigy review daneplus dane_plus_mdl_pred,dane  --view-id ner_manual --l NORP,CARDINAL,PRODUCT,ORGANIZATION,PERSON,WORK_OF_ART,EVENT,LAW,QUANTITY,DATE,TIME,ORDINAL,LOCATION,GPE,MONEY,PERCENT,FACILITY

Export the dataset:

prodigy data-to-spacy daneplus --ner daneplus --lang da -es 0
  1. Redo the original split:
python split.py