metadata

license: apache-2.0
task_categories:
  - token-classification
  - text-classification
language:
  - en
  - es
pretty_name: meta4xnli
size_categories:
  - 1K<n<10K
configs:
  - config_name: det_es_finetune
    data_files:
      - split: train
        path: detection/splits/es/meta4xnli_train.jsonl
      - split: dev
        path: detection/splits/es/meta4xnli_dev.jsonl
      - split: test
        path: detection/splits/es/meta4xnli_test.jsonl
  - config_name: det_en_finetune
    data_files:
      - split: train
        path: detection/splits/en/meta4xnli_train.jsonl
      - split: dev
        path: detection/splits/en/meta4xnli_dev.jsonl
      - split: test
        path: detection/splits/en/meta4xnli_test.jsonl
  - config_name: det_es_eval
    data_files:
      - split: esxnli_prem
        path: detection/source_datasets/es/esxnli_prem.jsonl
      - split: esxnli_hyp
        path: detection/source_datasets/es/esxnli_hyp.jsonl
      - split: xnli_dev_prem
        path: detection/source_datasets/es/xnli_dev_prem.jsonl
      - split: xnli_dev_hyp
        path: detection/source_datasets/es/xnli_dev_hyp.jsonl
      - split: xnli_test_prem
        path: detection/source_datasets/es/xnli_test_prem.jsonl
      - split: xnli_test_hyp
        path: detection/source_datasets/es/xnli_test_hyp.jsonl
  - config_name: det_en_eval
    data_files:
      - split: esxnli_prem
        path: detection/source_datasets/en/esxnli_prem.jsonl
      - split: esxnli_hyp
        path: detection/source_datasets/en/esxnli_hyp.jsonl
      - split: xnli_dev_prem
        path: detection/source_datasets/en/xnli_dev_prem.jsonl
      - split: xnli_dev_hyp
        path: detection/source_datasets/en/xnli_dev_hyp.jsonl
      - split: xnli_test_prem
        path: detection/source_datasets/en/xnli_test_prem.jsonl
      - split: xnli_test_hyp
        path: detection/source_datasets/en/xnli_test_hyp.jsonl
  - config_name: int_finetune
    datafiles:
      - split: train_no_met
        path: interpretation/splits/train_no_met.jsonl
      - split: train_met
        path: interpretation/splits/train_met.jsonl
      - split: train_nonrelevant
        path: interpretation/splits/train_nonrelevant.jsonl
      - split: dev_no_met
        path: interpretation/splits/dev_no_met.jsonl
      - split: dev_met
        path: interpretation/splits/dev_met.jsonl
      - split: dev_nonrelevant
        path: interpretation/splits/dev_nonrelevant.jsonl
      - split: test_no_met
        path: interpretation/splits/test_no_met.jsonl
      - split: test_met
        path: interpretation/splits/test_met.jsonl
      - split: test_nonrelevant
        path: interpretation/splits/test_nonrelevant.jsonl
  - config_name: int_eval
    datafiles:
      - split: esxnli_met
        path: interpretation/source_datasets/esxnli_met.jsonl
      - split: esxnli_no_met
        path: interpretation/source_datasets/esxnli_no_met.jsonl
      - split: esxnli_nonrelevant
        path: interpretation/source_datasets/esxnli_nonrelevant.jsonl
      - split: xnli_dev_met
        path: interpretation/source_datasets/xnli_dev_met.jsonl
      - split: xnli_dev_no_met
        path: interpretation/source_datasets/xnli_dev_no_met.jsonl
      - split: xnli_dev_nonrelevant
        path: interpretation/source_datasets/xnli_dev_nonrelevant.jsonl
      - split: xnli_test_met
        path: interpretation/source_datasets/xnli_test_met.jsonl
      - split: xnli_test_no_met
        path: interpretation/source_datasets/xnli_test_no_met.jsonl
      - split: xnli_test_nonrelevant
        path: interpretation/source_datasets/xnli_test_nonrelevant.jsonl

Dataset Card for Dataset Name

Meta4XNLI is a parallel dataset with annotations in English and Spanish for metaphor detection at token level (13320 sentences) and metaphor interpretation framed within NLI the task (9990 premise-hypothesis pairs). It is a collection of existing NLI datasets manually labeled for both metaphor tasks.

Repository: data available also in .tsv format at https://github.com/elisanchez-beep/meta4xnli
Paper: Meta4XNLI: A Crosslingual Parallel Corpus for Metaphor Detection and Interpretation

Dataset Sources

Meta4XNLI is a collection of XNLI and esXNLI datasets with metaphor annotations.

Dataset Structure

The dataset is divided according to detection and interpretation tasks.

Detection: labels at token level.
- splits: train, dev and test files for fine-tuning and evaluation.
- source_datasets: splits by original source dataset and premises and hypotheses for evaluation.
Intepretation: set of sentences split by metaphor occurrence. Non-relevant cases include sentences with metaphors, however, their literal interpretation is not necessary to extract the inference label.
- splits: train, dev and test files for fine-tuning and evaluation.
- source_datasets: splits by original source dataset and metaphor presence.

Dataset Fields

Detection:
- "id": example id
- "tokens": list of text split.
- "tags": list of metaphor annotations for each token.
  - 0: literal
  - 1: metaphor
Interpretation:
- "language": Spanish (es) or English (en)
- "gold_label": inference label: entailment, neutral or contradiction
- "sentence1": premise
- "sentence2": hypothesis
- "promptID": premise id
- "pairID": premise and hypothesis pair id
- "genre": text domain
- "source_dataset": original dataset: {xnli.dev, xnli.test, esxnli}

Citation [optional]

If you use Meta4XNLI, please cite our work:

@misc{sanchezbayona2024meta4xnli,
      title={Meta4XNLI: A Crosslingual Parallel Corpus for Metaphor Detection and Interpretation}, 
      author={Elisa Sanchez-Bayona and Rodrigo Agerri},
      year={2024},
      eprint={2404.07053},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Dataset Card Contact

{elisa.sanchez, rodrigo.agerri}@ehu.eus