Indonesian–Madurese Bible-Domain Parallel Corpus with Religious-Cultural Entity Spans

Description

This dataset contains 30,006 aligned Indonesian–Madurese Bible-domain text pairs assembled from author-supplied files representing 66 distinct books. It also contains 63,556 automatically generated religious-cultural entity candidate spans across 19,148 text pairs. Each span has a label, character offsets, a sentence identifier, and source provenance. The five labels are REL-TERM, REL-TITLE, REL-PER, REL-LOC, and SOC-REL. The candidate spans have not been documented as individually validated by human annotators. Seventy-two additional exact duplicate bilingual pairs are retained and identified. This release does not provide a verified experimental train/validation/test split or reproduce the model scores reported in the associated manuscript. It is related to the Bahasa Madura Corpus Dataset (https://doi.org/10.17632/cgtg4bhrtf.1). See README.md and SOURCE_AND_LICENSE.md for provenance, limitations, and reuse conditions.

Steps to reproduce

1. Download all files in this dataset into the same folder. 2. Run python3 verify_dataset.py using Python 3. 3. The script checks SHA-256 file hashes, unique identifiers, annotation-to-corpus links, character offsets, and the counts reported in VALIDATION_REPORT.json. 4. Read DATA_DICTIONARY.md for field definitions and annotation_guidelines.tsv for the five label categories. This procedure verifies the released files and their structural integrity. It does not validate the semantic accuracy of candidate annotations or reproduce the model experiments reported in the manuscript.

Institutions

• Institut Teknologi dan Bisnis Asia Malang

  East Java, Kota Malang

Categories

Licence