Omarrran/BharatSetu-22Indic-1M

BharatSetu 22-Indic 1M

BharatSetu 22-Indic 1M is a private wide parallel translation dataset containing 984,764 English source rows translated into 22 Indic target languages.

The Hugging Face Dataset Viewer is configured to read the Parquet shards under data/train-*.parquet. Large preservation artifacts are also included under jsonl/, full/, and parquet/.

Dataset summary

• Rows: 984,764
• Source language: English (en)
• Target languages: 22 Indic languages
• Shape: one row per English sentence, with one column for each target language
• Primary viewer files: data/train-*.parquet
• Preserved artifacts: JSONL, XLSX, single-file Parquet, generation manifest, metadata
• Generation model family: sarvamai/sarvam-translate

Languages

• Assamese (as)
• Bengali (bn)
• Bodo (brx)
• Dogri (doi)
• Gujarati (gu)
• Hindi (hi)
• Kannada (kn)
• Kashmiri (ks)
• Konkani (kok)
• Maithili (mai)
• Malayalam (ml)
• Manipuri (mni)
• Marathi (mr)
• Nepali (ne)
• Odia (or)
• Punjabi (pa)
• Sanskrit (sa)
• Santali (sat)
• Sindhi (sd)
• Tamil (ta)
• Telugu (te)
• Urdu (ur)

Schema

Each row has:

• id: integer row id aligned to the original source corpus
• english: source English text
• source: provenance/source field from the input CSV
• Assamese: generated Assamese translation
• Bengali: generated Bengali translation
• Bodo: generated Bodo translation
• Dogri: generated Dogri translation
• Gujarati: generated Gujarati translation
• Hindi: generated Hindi translation
• Kannada: generated Kannada translation
• Kashmiri: generated Kashmiri translation
• Konkani: generated Konkani translation
• Maithili: generated Maithili translation
• Malayalam: generated Malayalam translation
• Manipuri: generated Manipuri translation
• Marathi: generated Marathi translation
• Nepali: generated Nepali translation
• Odia: generated Odia translation
• Punjabi: generated Punjabi translation
• Sanskrit: generated Sanskrit translation
• Santali: generated Santali translation
• Sindhi: generated Sindhi translation
• Tamil: generated Tamil translation
• Telugu: generated Telugu translation
• Urdu: generated Urdu translation

Files

• data/train-*.parquet — Hugging Face Dataset Viewer / load_dataset source.
• jsonl/english_to_22_indic_all_languages.jsonl — full wide JSONL export.
• full/english_to_22_indic_all_languages.xlsx — full wide Excel workbook.
• parquet/english_to_22_indic_all_languages.parquet — single-file Parquet export.
• metadata.json — row counts, schema, language list, file sizes/hashes.
• generation/merge_manifest.json — generation/merge manifest from the packaging job.

Loading examples

from datasets import load_dataset

ds = load_dataset("Omarrran/BharatSetu-22Indic-1M", split="train")
print(ds)
print(ds.column_names)

Select one target language

ks = ds.select_columns(["id", "source", "english", "Kashmiri"])
print(ks[0])

Select two target languages

hi_ur = ds.select_columns(["id", "english", "Hindi", "Urdu"])

Convert wide rows to long format

target_languages = ['Assamese', 'Bengali', 'Bodo', 'Dogri', 'Gujarati', 'Hindi', 'Kannada', 'Kashmiri', 'Konkani', 'Maithili', 'Malayalam', 'Manipuri', 'Marathi', 'Nepali', 'Odia', 'Punjabi', 'Sanskrit', 'Santali', 'Sindhi', 'Tamil', 'Telugu', 'Urdu']

for row in ds:
    for lang in target_languages:
        yield {
            "id": row["id"],
            "source_language": "English",
            "target_language": lang,
            "source_text": row["english"],
            "target_text": row[lang],
        }

Notes and limitations

• This is a machine-generated translation corpus, not a manually human-verified benchmark.
• Quality can vary by language, domain, and sentence style.
• The source column is preserved from the input English corpus for provenance.
• Use the Parquet shards in data/ for normal Dataset Viewer and load_dataset use.

Provenance