ro-paraphrase-bible / README.md
andyP's picture
Update README.md
4cd9770 verified
metadata
license: apache-2.0
annotations_creators:
  - machine-generated
language_creators:
  - found
task_categories:
  - sentence-similarity
language:
  - ro
multilinguality:
  - monolingual
tags:
  - sentence similarity
  - paraphrase
  - romanian
  - nlp
task_ids:
  - text-scoring
  - semantic-similarity-scoring
  - semantic-similarity-classification
pretty_name: Romanian Bible Paraphrase Corpus
size_categories:
  - 100K<n<1M
dataset_info:
  features:
    - name: text1
      dtype: string
    - name: text2
      dtype: string
    - name: score
      dtype: int8

Dataset Card for "Romanian Bible Paraphrase Corpus"

Table of Contents

Dataset Description

Dataset Summary

A paraphprase corpus created from 10 different Romanian language Bible versions. Since the Bible has all paragraphs uniquely numbered an alignment between two versions is straighforward. We compiled a complete combination of paragraph pairs from the following Bible versions:

  • Română Noul Testament Interconfesional 2009
  • Biblia în versuri 2014
  • Biblia Traducerea Fidela 2015
  • Biblia în Versiune Actualizată 2018
  • Ediția Dumitru Cornilescu revizuită 2022
  • Noua Traducere Românească
  • Noul Testament SBR 2023
  • Versiunea Biblia Romano-Catolică 2020
  • Biblia sau Sfânta Scriptură cu Trimiteri 1924, Dumitru Cornilescu
  • Traducere Literală Cornilescu 1931

In order to provide a similarity score between two paragraphs, we use FuzzyWuzzy library to compute the partial_token_sort_ratio (PTSoR) between the lemmatized versions of the two texts. The final similarity score between matching paragraphs was computed as follows:

score = 50 + ptsor // 2

where ptsor = partial_token_sort_ratio(lemmatize(text1), lemmatize(text2))

We also provide a set of non-matching texts, generated from a random pairing of paragraphs. The score for the non-matching paragraphs was computed using this formula:

score = int(ptsor / 2 +10)

The final dataset contains 904,815 similar records and 218,977 non matching records, totaling 1,123,927

Languages

Romanian

Dataset Structure

Data Instances

An example record looks as follows.

{
  'text1': 'Poporul a început să strige: „Este glasul lui Dumnezeu, nu al unui om!”',
  'text2': 'Norodul a strigat: „Glas de Dumnezeu, nu de om!”',
  'score': 82
}

Data Fields

  • text1, text2: compared Paragraphs
  • score: 0 - 100 matching score (integer) - records with score > 50 are matches

Data Statistics

Score Distribution:

Score distribution

Statistics for matching record scores:

count    904950.000000
mean         83.998228
std           7.083741
min          50.000000
25%          80.000000
50%          84.000000
75%          89.000000
max         100.000000

Statistics for non-matching record scores:

count    218977.000000
mean         29.400873
std           3.339921
min          10.000000
25%          27.000000
50%          29.000000
75%          32.000000
max          40.000000

Dataset Creation

Curation Rationale

Source Data

Romanian Language Bible translations

Initial Data Collection and Normalization

Who are the source language producers?

Annotations

Annotation process

Who are the annotators?

Personal and Sensitive Information

Considerations for Using the Data

Social Impact of Dataset

Discussion of Biases

Other Known Limitations

The dataset is biased towards some archaic language. Additionally, it involves religious figures and phrase constructions that are rarly used in the common language.

Additional Information

Dataset Curators

Licensing Information

This data is available and distributed under Apache-2.0 license

Citation Information

tbd

Contributions