BanglaChhanda — pilot corpus

Bangla verse prepared for chhanda (ছন্দ) annotation: syllable structure, foot boundaries, and meter under অক্ষরবৃত্ত, মাত্রাবৃত্ত and স্বরবৃত্ত.

This is not gold data. Every label in pilot.jsonl was produced by a rule-based scanner and has not been checked by a human. The scanner itself flags 80% of these lines as needing review, and the annotation guideline it implements has not yet been reviewed by a prosody specialist. Treat this as a pre-annotation draft for annotators to correct — not as ground truth, and not as a benchmark.

What is here

File	Contents
data/poems.jsonl	153 poems, 7,009 lines, as fetched, with per-poem provenance
data/pilot.jsonl	316 lines from 19 poems, one pre-filled annotation record per line
docs/annotation-guide-v0.1.html	the annotation guideline these records follow

Sampling

Poem-sampled, stratified by closed-syllable density into three bands and by poet, taken round-robin, with a seeded shuffle so the selection reproduces.

Whole poems, because scoring a poem's metrical consistency needs its lines together. Stratified, because the closed-syllable rule is the only thing that separates the three meters — a closed syllable is worth 2 matra in matrabritta, 1 in svarabritta, and 1 or 2 in akkharbritta depending on whether it ends a word. A line of entirely open syllables reads identically under all three, so a random sample would fill up with lines that carry no signal about meter.

Two poets, both out of copyright: Rabindranath Tagore (d. 1941) and Jibanananda Das (d. 1954). The pilot is 191 lines of Jibanananda and 125 of Tagore, spanning closed-syllable density 0.20 to 0.35.

Kazi Nazrul Islam is deliberately absent. He died in 1976, so under life + 60 his work stays in copyright in Bangladesh and India until 2037. Wikisource carries his pre-1931 books because they are public domain in the United States; that is not enough for a dataset redistributed under CC BY-SA, so the fetcher refuses him by design rather than by oversight.

Jibanananda matters here beyond variety: much of his work is loose or free verse, where Tagore's কণিকা is rhymed payar throughout. If the rules only worked on regular metre, he is where that would show.

Record schema

{
  "poem_id": "কণিকা-অধিকার",
  "line_no": 1,
  "text": "অধিকার বেশি কার বনের উপর",
  "register": "unknown",
  "syllables": [
    {"i": 0, "span": [0, 1], "type": "open", "word_final": false,
     "conjunct": false, "flags": []}
  ],
  "feet": [[0, 4], [4, 7]],
  "meter": {"proposed": "akkharbritta", "pattern": "payar",
            "alignment": 1.0, "weak": false,
            "confirmed": null, "override_reason": null},
  "needs_review": true,
  "provenance": {"source_url": "...", "revision": 1941668, "retrieved": "2026-09-16"}
}

There is no matra field, on purpose. Matra is derived from syllable structure plus the meter's rules. Storing it would bake the textbook rules into the data and make it impossible to test whether those rules predict what annotators actually hear. Structure is annotated; matra is computed.

meter.confirmed is null throughout — the machine proposes, a human disposes.

Flags

The scanner marks what it is unsure about instead of guessing. Counts across the 303 pilot lines:

Flag	Lines	Meaning
conjunct_split	171	a written cluster divided across a syllable boundary
schwa_medial_uncertain	161	inherent vowel before another inherent vowel — variable, partly dialectal
phala_kept_in_onset	74	r-phala held in the onset: আক্রমণ is a-kro-mon
final_schwa_kept_after_conjunct	67	শব্দ stays shob-do rather than becoming shobd
y_phala_gemination_check	30	y-phala may geminate; a human decides

Per poet the review rate barely moves — 76% for Jibanananda, 74% for Tagore — so the rules are not simply falling over on modern verse.

Known limitations

• Final-schwa exceptions are not modelled. The rule deletes a word-final inherent vowel unless it follows a conjunct. That is wrong for a class of common words — ছোট is said cho-to, and the scanner gives one closed syllable. How often the rule misfires is a number the annotation should report, not something to patch by intuition.
• The foot patterns are a starting set, not a full inventory of Bangla line forms. Tripadi and the longer forms are absent.
• No prosodic authority is pinned yet. The matra rules follow the common textbook description; authorities differ on exactly the edge cases that decide inter-annotator agreement.
• Meter proposals rank, they do not classify. Under svarabritta every syllable is 1 matra, so any line whose syllable count divides by four fits exactly on the arithmetic alone. Each fit now carries meter.alignment: the share of its foot boundaries that land at word boundaries, which is where Bangla feet tend to break. That reorders the candidates sensibly and marks 15% of pilot lines weak (alignment below 0.5), but it still does not reject an unmetrical line — a deliberately prosaic line can score 0.67. Read meter.proposed as a starting guess for an annotator, never as a classification.
• Two poets. Nothing here establishes how the rules behave on Nazrul (excluded on copyright grounds until 2037), on prose poetry, or on song lyrics.

Citing this work

This dataset and code are released under CC BY-SA 4.0, which requires attribution. If you use the scanner, the guideline or any part of the corpus in research or in a derived resource, cite it:

@misc{morol2026banglachhanda,
  author    = {Morol, Md Kishor},
  orcid     = {0000-0002-4468-8260},
  title     = {{BanglaChhanda}: a rule-based scanner and pilot corpus for
               {Bangla} verse},
  year      = {2026},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.22798431},
  url       = {https://doi.org/10.5281/zenodo.22798431}
}

That is the concept DOI: it always resolves to the newest version, which is what you want when citing the work in general. To pin the exact release you used, cite the version DOI instead — v0.1.0 is 10.5281/zenodo.22798432.

Author ORCID: 0000-0002-4468-8260.

Please also credit Bengali Wikisource for the transcriptions, as the CC BY-SA terms of the source text require. The poems themselves are public domain.

If you correct or extend the annotations, CC BY-SA also requires you to release the result under the same licence.

Source, rights and attribution

Poems are by Rabindranath Tagore (died 1941), in the public domain in Bangladesh and India (life + 60) and in the United States.

Transcriptions come from Bengali Wikisource, and each poem record keeps its page URL, revision id and retrieval date, so any line traces back to the exact revision it came from. The dataset is released under CC BY-SA 4.0 in keeping with Wikisource's terms; the underlying poems are public domain.

Attribution is a licence condition, not a courtesy. Using this dataset without crediting the author and Bengali Wikisource does not comply with CC BY-SA 4.0. See the citation block above.

Code

The scanner, the fetcher and the pilot builder are at https://github.com/kishormorol/banglachhanda.