kardosdrur's picture
Updated Readme
0471b7d
metadata
license: mit
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
dataset_info:
  features:
    - name: da
      dtype: string
    - name: en
      dtype: string
    - name: sv
      dtype: string
  splits:
    - name: train
      num_bytes: 620348322.4
      num_examples: 1304296
    - name: test
      num_bytes: 155087080.6
      num_examples: 326074
  download_size: 488376564
  dataset_size: 775435403

Europarl Scandinavian Languages

The data originates from the Europarl parallel corpus, where English transcriptions of parliamentary discussions were aligned with a number of other languages algorithmically.

In order to align Danish and Swedish corpora in the dataset, English entries were hashed with 128bit Murmurhash3, and the Danish and Swedish transcriptions were joined on the obtained hash values. Entries that had more than one pair in the other dataset were removed, this ensures that no false positives due to hash collisions got into the dataset. Source code is available in the repository.

The dataset was created for aiding the training of sentence transformer models in the Danish Foundation Models project.