metadata

annotations_creators:
  - expert-generated
language_creators:
  - crowdsourced
language:
  - en
license:
  - apache-2.0
multilinguality:
  - monolingual
pretty_name: probability_words_nli
paperwithcoode_id: probability-words-nli
size_categories:
  - 1K<n<10K
source_datasets:
  - original
task_categories:
  - text-classification
  - multiple-choice
  - question-answering
task_ids:
  - open-domain-qa
  - multiple-choice-qa
  - natural-language-inference
  - multi-input-text-classification
tags:
  - wep
  - words of estimative probability
  - probability
  - logical reasoning
  - soft logic
  - nli
  - verbal probabilities
  - natural-language-inference
  - reasoning
  - logic
train-eval-index:
  - config: usnli
    task: text-classification
    task_id: multi-class-classification
    splits:
      train_split: train
      eval_split: validation
    col_mapping:
      sentence1: context
      sentence2: hypothesis
      label: label
    metrics:
      - type: accuracy
        name: Accuracy
      - type: f1
        name: F1 binary
  - config: reasoning-1hop
    task: text-classification
    task_id: multi-class-classification
    splits:
      train_split: train
      eval_split: validation
    col_mapping:
      sentence1: context
      sentence2: hypothesis
      label: label
    metrics:
      - type: accuracy
        name: Accuracy
      - type: f1
        name: F1 binary
  - config: reasoning-2hop
    task: text-classification
    task_id: multi-class-classification
    splits:
      train_split: train
      eval_split: validation
    col_mapping:
      sentence1: context
      sentence2: hypothesis
      label: label
    metrics:
      - type: accuracy
        name: Accuracy
      - type: f1
        name: F1 binary

Dataset accompanying the "Probing neural language models for understanding of words of estimative probability" article

This dataset tests the capabilities of language models to correctly capture the meaning of words denoting probabilities (WEP, also called verbal probabilities), e.g. words like "probably", "maybe", "surely", "impossible".

We used probabilitic soft logic to combine probabilistic statements expressed with WEP (WEP-Reasoning) and we also used the UNLI dataset (https://nlp.jhu.edu/unli/) to directly check whether models can detect the WEP matching human-annotated probabilities according to Fagen-Ulmschneider, 2018. The dataset can be used as natural language inference data (context, premise, label) or multiple choice question answering (context,valid_hypothesis, invalid_hypothesis).

Code : colab

Accepted at Starsem2023 (The 12th Joint Conference on Lexical and Computational Semantics). Temporary citation:

@inproceedings{sileo-moens-2023-probing,
    title = "Probing neural language models for understanding of words of estimative probability",
    author = "Sileo, Damien  and
      Moens, Marie-francine",
    booktitle = "Proceedings of the 12th Joint Conference on Lexical and Computational Semantics (*SEM 2023)",
    month = jul,
    year = "2023",
    address = "Toronto, Canada",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.starsem-1.41",
    doi = "10.18653/v1/2023.starsem-1.41",
    pages = "469--476",
}