---
language:
- en
license: mit
size_categories:
- 10K<n<100K
pretty_name: Semantically-augmented FEVER for NLI
dataset_info:
  features:
  - name: id
    dtype: string
  - name: premise
    dtype: string
  - name: hypothesis
    dtype: string
  - name: label
    dtype: string
  - name: wsd
    struct:
    - name: premise
      list:
      - name: index
        dtype: int64
      - name: text
        dtype: string
      - name: pos
        dtype: string
      - name: lemma
        dtype: string
      - name: bnSynsetId
        dtype: string
      - name: wnSynsetOffset
        dtype: string
      - name: nltkSynset
        dtype: string
    - name: hypothesis
      list:
      - name: index
        dtype: int64
      - name: text
        dtype: string
      - name: pos
        dtype: string
      - name: lemma
        dtype: string
      - name: bnSynsetId
        dtype: string
      - name: wnSynsetOffset
        dtype: string
      - name: nltkSynset
        dtype: string
  - name: srl
    struct:
    - name: premise
      struct:
      - name: tokens
        list:
        - name: index
          dtype: int64
        - name: rawText
          dtype: string
      - name: annotations
        list:
        - name: tokenIndex
          dtype: int64
        - name: verbatlas
          struct:
          - name: frameName
            dtype: string
          - name: roles
            list:
            - name: role
              dtype: string
            - name: score
              dtype: float64
            - name: span
              sequence: int64
        - name: englishPropbank
          struct:
          - name: frameName
            dtype: string
          - name: roles
            list:
            - name: role
              dtype: string
            - name: score
              dtype: float64
            - name: span
              sequence: int64
    - name: hypothesis
      struct:
      - name: tokens
        list:
        - name: index
          dtype: int64
        - name: rawText
          dtype: string
      - name: annotations
        list:
        - name: tokenIndex
          dtype: int64
        - name: verbatlas
          struct:
          - name: frameName
            dtype: string
          - name: roles
            list:
            - name: role
              dtype: string
            - name: score
              dtype: float64
            - name: span
              sequence: int64
        - name: englishPropbank
          struct:
          - name: frameName
            dtype: string
          - name: roles
            list:
            - name: role
              dtype: string
            - name: score
              dtype: float64
            - name: span
              sequence: int64
  splits:
  - name: train
    num_bytes: 357653267
    num_examples: 51086
  - name: validation
    num_bytes: 15794078
    num_examples: 2288
  - name: test
    num_bytes: 15736002
    num_examples: 2287
  download_size: 77623798
  dataset_size: 389183347
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---
# Semantically-augmented FEVER for NLI

This dataset is a random downsample of the [FEVER dataset adapted for NLI](https://github.com/easonnie/combine-FEVER-NSMN/blob/master/other_resources/nli_fever.md).
We downsampled the training and development set to 25% of the original, and recovered labels for the development set from the original [FEVER dataset](https://huggingface.co/datasets/fever/fever).

The dataset is also augmented with semantic annotations such as Word Sense Disambiguation (WSD) and Semantic Role Labeling (SRL) information.
We annotated the whole downsampled dataset (both `premise` and `hypothesis`) with [AMuSE-WSD](https://aclanthology.org/2021.emnlp-demo.34) (Orlando et al., EMNLP 2021) and [InVeRo](https://aclanthology.org/2020.emnlp-demos.11) (Conia et al., EMNLP 2020). 


## Dataset Creation

The idea was to curate a version of the FEVER dataset adapted to the NLI task for Homework 2 of the Multilingual Natural Language Processing 2024 course at Sapienza University of Rome.

We sourced the data following the instructions in [this repo](https://github.com/easonnie/combine-FEVER-NSMN/blob/master/other_resources/nli_fever.md), modifying the labels to the following schema:

```json
{
  "id": ..., # the FEVER dataset ID
  "premise": ..., # the context in FEVER
  "hypothesis": ..., # the query in FEVER
  "label": ..., # mapped version of FEVER, where 'supports' -> 'entailment', 'refutes' -> 'contradiction' and 'not enough info' -> 'neutral'
}
```

We filtered out any sample with an empty premise (a minority of the data) and downsampled both `train_fitems.jsonl` and `dev_fitems.jsonl` to 25% of their total size.
We then recovered the labels for the development set from the `labelled_dev` split in the [original FEVER](https://huggingface.co/datasets/fever/fever) and split the development set into the final `dev.jsonl` and `test.jsonl`.

Finally, we ran both AMuSE-WSD and InVeRo to augment our samples with WSD and SRL annotations, respectively.