CSMD / README.md
davebulaval's picture
Update README.md
4d64b42 verified
metadata
license:
  - cc-by-4.0
multilinguality:
  - monolingual
  - aligned
task_categories:
  - text-classification
  - text2text-generation
source_datasets:
  - original
  - >-
    extended|other-turkcorpus,other-asset,other-questeval,other-simplicity_da,other-simp_da
language:
  - en
tags:
  - simplification-evaluation
  - meaning-evaluation
pretty_name: CSMD
size_categories:
  - 1K<n<10K
dataset_info:
  - config_name: meaning
    features:
      - name: original
        dtype: string
      - name: simplification
        dtype: string
      - name: label
        dtype: float64
    splits:
      - name: train
        num_bytes: 251558
        num_examples: 853
      - name: dev
        num_bytes: 27794
        num_examples: 95
      - name: test
        num_bytes: 117686
        num_examples: 407
    download_size: 397038
    dataset_size: 1355
  - config_name: meaning_with_data_augmentation
    features:
      - name: original
        dtype: string
      - name: simplification
        dtype: string
      - name: label
        dtype: float64
    splits:
      - name: train
        num_bytes: 1242540
        num_examples: 4267
      - name: dev
        num_bytes: 134726
        num_examples: 475
      - name: test
        num_bytes: 592052
        num_examples: 2033
    download_size: 1969318
    dataset_size: 6775
  - config_name: meaning_holdout_identical
    features:
      - name: original
        dtype: string
      - name: simplification
        dtype: string
      - name: label
        dtype: float64
    splits:
      - name: test
        num_bytes: 89866
        num_examples: 359
    download_size: 89866
    dataset_size: 359
  - config_name: meaning_holdout_unrelated
    features:
      - name: original
        dtype: string
      - name: simplification
        dtype: string
      - name: label
        dtype: float64
    splits:
      - name: test
        num_bytes: 247835
        num_examples: 359
    download_size: 247835
    dataset_size: 359
config_names:
  - meaning
  - meaning_with_data_augmentation
  - meaning_holdout_identical
  - meaning_holdout_unrelated
viewer: true
configs:
  - config_name: meaning
    data_files:
      - split: train
        path: train.tsv
      - split: dev
        path: dev.tsv
      - split: test
        path: test.tsv
  - config_name: meaning_with_data_augmentation
    data_files:
      - split: train
        path: train_da.tsv
      - split: dev
        path: dev_da.tsv
      - split: test
        path: test_da.tsv
  - config_name: meaning_holdout_identical
    data_files:
      - split: test
        path: identical.tsv
  - config_name: meaning_holdout_unrelated
    data_files:
      - split: test
        path: unrelated.tsv

Dataset Card for "Continuous Scale Meaning Dataset" (CSMD)

CSMD was created for MeaningBERT: Assessing Meaning Preservation Between Sentences.

It contains 1,355 English text simplification meaning preservation annotations. Meaning preservation measures how well the meaning of the output text corresponds to the meaning of the source (Saggion, 2017).

The annotations were taken from the following four datasets:

It contains a data augmentation subset of 1,355 identical sentence triplets and 1,355 unrelated sentence triplets (See the "Sanity Checks" section (3.3.) in our article).

It also contains two holdout subsets of 359 identical sentence triplets and 359 unrelated sentence triples (See the "MeaningBERT" section (3.4.) in our article).

Dataset Structure

Data Instances

  • Meaning configuration: an instance consists of 1,355 meaning preservation triplets (Document, simplification, label).
  • meaning_with_data_augmentation configuration: an instance consists of 1,355 meaning preservation triplets (Document, simplification, label) along with 1,355 data augmentation triplets (Document, Document, 100) and 1,355 data augmentation triplets (Document, Unrelated Document, 0) (See the sanity checks in our article).
  • meaning_holdout_identical configuration: an instance consists of 359 meaning holdout preservation identical triplets (Document, Document, 1) based on the ASSET Simplification dataset.
  • meaning_holdout_unrelated configuration: an instance consists of 359 meaning holdout preservation unrelated triplets (Document, Unrelated Document, 0) based on the ASSET Simplification dataset.

About the Data Augmentation

Unrelated Sentence

We have changed the data augmentation approach for the unrelated sentence. Instead of generating noisy sentences using an LLM, for each of the 1,355 sentences, we sample a sentence in the unlabeled sentence in ASSET (non included in the holdout nor the labelled sentence). We compute the Rouge1, Rouge2, RougeL and bleu scores to validate that the sentences are unrelated in terms of vocabulary. Namely, each metric score is below 0.20 or 20 for Bleu for all pairs. If a pair achieves a higher value, we select another sentence from ASSET to create a pair and reapply the test until a pair achieves a score below 0.20/20.

Commutative Property

Since meaning preservation is a commutative function, i.e., Meaning(Sent_a, Sent_b) = Meaning(Sent_b, Sent_a), we also include the commutative version of the triplet in the data augmentation version of the dataset for sentences that are not identical.

Data Fields

  • original: an original sentence from the source datasets.
  • simplification: a simplification of the original obtained by an automated system or a human.
  • label: a meaning preservation rating between 0 and 100.

Data Splits

The split statistics of CSMD are given below.

Train Dev Test Total
Meaning 853 95 407 1,355
Meaning With Data Augmentation 2,560 285 1,220 4,065
Meaning Holdout Identical NA NA 359 359
Meaning Holdout Unrelated NA NA 359 359

All the splits are randomly split using a 60-10-30 split with the seed 42.

Citation Information

@ARTICLE{10.3389/frai.2023.1223924,
AUTHOR={Beauchemin, David and Saggion, Horacio and Khoury, Richard},   
TITLE={{MeaningBERT: Assessing Meaning Preservation Between Sentences}},      
JOURNAL={Frontiers in Artificial Intelligence},      
VOLUME={6},           
YEAR={2023},      
URL={https://www.frontiersin.org/articles/10.3389/frai.2023.1223924},       
DOI={10.3389/frai.2023.1223924},        
ISSN={2624-8212},   
}