README.md

metadata

license: cc-by-sa-4.0
language_creators:
  - machine-generated
dataset_info:
  features:
    - name: tokens_a
      sequence: string
    - name: tokens_b
      sequence: string
    - name: labels_a
      sequence: float64
    - name: labels_b
      sequence: float64
    - name: lang_a
      dtype: string
    - name: lang_b
      dtype: string
    - name: subset
      dtype: string
    - name: id
      dtype: string
    - name: alignments
      dtype: string
  splits:
    - name: train_en
      num_bytes: 1640900
      num_examples: 1506
    - name: train_de
      num_bytes: 1101404
      num_examples: 3012
    - name: train_es
      num_bytes: 1154765
      num_examples: 3012
    - name: train_fr
      num_bytes: 1206414
      num_examples: 3012
    - name: train_ja
      num_bytes: 838252
      num_examples: 3012
    - name: train_ko
      num_bytes: 829328
      num_examples: 3012
    - name: train_zh
      num_bytes: 796140
      num_examples: 3012
    - name: test_en
      num_bytes: 833900
      num_examples: 750
    - name: test_de
      num_bytes: 558624
      num_examples: 1500
    - name: test_es
      num_bytes: 580224
      num_examples: 1500
    - name: test_fr
      num_bytes: 610017
      num_examples: 1500
    - name: test_ja
      num_bytes: 425912
      num_examples: 1500
    - name: test_ko
      num_bytes: 424407
      num_examples: 1500
    - name: test_zh
      num_bytes: 403680
      num_examples: 1500
  download_size: 2569205
  dataset_size: 11403967
task_categories:
  - token-classification
language:
  - en
  - de
  - es
  - fr
  - ja
  - ko
  - zh
size_categories:
  - 1K<n<10K

Training and test data for the task of Recognizing Semantic Differences (RSD).

See the paper for details on how the dataset was created, and see our code at https://github.com/ZurichNLP/recognizing-semantic-differences for an example of how to use the data for evaluation.

The data are derived from the SemEval-2016 Task 2 for Interpretable Semantic Textual Similarity organized by Agirre et al. (2016). The original URLs of the data are:

• Train: http://alt.qcri.org/semeval2016/task2/data/uploads/train_2015_10_22.utf-8.tar.gz
• Test: http://alt.qcri.org/semeval2016/task2/data/uploads/test_goldstandard.tar.gz

The translations into non-English languages have been created using machine translation (DeepL).

Citation

@inproceedings{vamvas-sennrich-2023-rsd,
      title={Towards Unsupervised Recognition of Token-level Semantic Differences in Related Documents},
      author={Jannis Vamvas and Rico Sennrich},
      month = dec,
      year = "2023",
      booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing",
      address = "Singapore",
      publisher = "Association for Computational Linguistics",
}