SrikantaVerse — Sarat Chandra Chattopadhyay Bengali Literary Corpus

Bengali literary prose by Sarat Chandra Chattopadhyay (1876-1938).

Sixteen novels and novellas, including all four parts of Srikanta (1917, 1918, 1927, 1933), Palli-Samaj, Charitrahin, Debdas, Parinita, Biraj Bou, Pather Dabi, Panditmashai, Chandranath, Baikunther Uil, Arakshaniya, Baradidi, and the unfinished Shesher Parichay. Novels are split into their printed chapters.

Eleven short stories, complete with the collection each was first gathered into: Bindur Chhele, Ramer Sumati and Path-Nirdesh; Mejdidi, Darpachurna and Andhare Alo; and Kashinath, Alo O Chhaya, Mandir, Bojha and Anupamar Prem. Each story is one whole record.

Most of the corpus comes from volume one of the three-volume শরৎ রচনাবলী, whose modern typesetting reads considerably better than the mid-century single-book scans; four works absent from it are taken from their own editions.

Every record carries its work, chapter or story title, the collection it was gathered from, the year of first book publication, and the page-level OCR confidence scores that produced the text, so consumers can filter to the quality level their use case needs.

At a glance

records	204
works	27 (16 long-form, 11 short stories)
characters	4,010,627
words	669,239
tokens (est.)	1,251,388
language	bn (Beng)
author	শরৎচন্দ্র চট্টোপাধ্যায়
first published	1913–1939
editions scanned	1947–1959
mean OCR confidence	79.4 / 100
licence	cc0-1.0

Each record is one chapter of a novel or one complete short story, so records carry whole narrative arcs rather than fragments.

Text is uncorrected OCR. Rather than silently shipping the errors, every record carries the OCR confidence scores that produced it, so you can filter to whatever quality level your use case needs — see Filtering by quality below.

Contents

Long-form works

work	title (en)	first published	edition	chapters	chars	tokens	mean conf
চরিত্রহীন	Charitrahin	1917	1958	31	639,903	209,242	74.3
পথের দাবী	Pather Dabi	1926	1958	1	540,324	167,182	75.9
শ্রীকান্ত — তৃতীয় পর্ব	Srikanta — Part Three	1927	—	14	257,525	76,570	82.1
শ্রীকান্ত — চতুর্থ পর্ব	Srikanta — Part Four	1933	—	9	251,664	77,690	78.2
শেষের পরিচয়	Shesher Parichay	1939	1959	15	247,282	80,240	79.5
শ্রীকান্ত — দ্বিতীয় পর্ব	Srikanta — Part Two	1918	—	13	236,822	71,728	79.6
শ্রীকান্ত — প্রথম পর্ব	Srikanta — Part One	1917	—	7	232,989	70,658	80.4
পল্লী-সমাজ	Palli-Samaj	1916	—	13	197,474	59,081	82.3
বিরাজ বৌ	Biraj Bou	1914	—	13	154,858	47,153	82.3
পণ্ডিতমশাই	Panditmashai	1914	—	11	153,130	45,836	82.6
দেবদাস	Debdas	1917	1947	14	149,639	49,800	74.5
চন্দ্রনাথ	Chandranath	1916	—	17	113,346	36,018	80.1
বৈকুণ্ঠের উইল	Baikunther Uil	1916	—	9	96,154	29,264	84.1
অরক্ষণীয়া	Arakshaniya	1916	—	7	94,966	31,011	74.0
পরিণীতা	Parinita	1914	—	9	83,942	26,276	78.4
বড়দিদি	Baradidi	1913	—	10	71,413	20,917	87.0

Short stories

One record each; collection_bn / collection_en name the volume they were printed in.

work	title (en)	first published	edition	records	chars	tokens	mean conf
বিন্দুর ছেলে	Bindur Chhele	1914	—	1	79,160	24,325	84.8
পথ-নির্দেশ	Path-Nirdesh	1914	—	1	56,596	18,234	74.4
রামের সুমতি	Ramer Sumati	1914	—	1	55,487	17,310	81.4
কাশীনাথ	Kashinath	1917	—	1	54,636	16,584	80.7
দর্পচূর্ণ	Darpachurna	—	—	1	49,397	15,135	81.2
মেজদিদি	Mejdidi	—	—	1	47,245	15,348	72.9
অনুপমার প্রেম	Anupamar Prem	1917	—	1	39,557	12,021	84.0
বোঝা	Bojha	1917	—	1	30,144	9,271	81.6
আঁধারে আলো	Andhare Alo	—	—	1	29,035	8,856	84.8
আলো ও ছায়া	Alo O Chhaya	1917	—	1	24,169	7,919	74.8
মন্দির	Mandir	1917	—	1	23,770	7,719	71.3

Record size distribution

	chars
smallest	819
median	13,356
90th percentile	32,882
largest	540,324

Sources

Every record traces back to a scanned printed edition on the Internet Archive.

source PDF	Internet Archive item	pages	works	status
2015.266432.Sarat-Rachanabali_text.pdf	in.ernet.dli.2015.266432	960	23	ready
2015.456034.Charitrahin_text.pdf	in.ernet.dli.2015.456034	320	1	ready
2015.457393.Debdas_text.pdf	in.ernet.dli.2015.457393	110	1	ready
2015.456067.Pather-Dabi_text.pdf	in.ernet.dli.2015.456067	304	1	ready
2015.456115.Shesher-Parichay_text.pdf	in.ernet.dli.2015.456115	99	1	ready
2015.465831.Dena-Paona_text.pdf	in.ernet.dli.2015.465831	319	0	excluded

Excluded sources

Kept in the manifest rather than deleted, so the reason stays on the record:

• 2015.465831.Dena-Paona_text.pdf (excluded) — NOT BENGALI. The item is catalogued language: mni -- Manipuri (Meitei), which is written in the Bengali script, so a Bengali-codepoint check passes it cleanly. It is a 1982 Imphal translation (imprint: রঘুমণি শর্ম্মা … ইম্ফাল … ১৯৮২) of ষোড়শী, the play adapted from the novel দেনাপাওনা; body text reads মহাকপু, হায়রকই, খুঙ্গং. Excluded three times over: wrong language, wrong genre, and a 1982 translation still in copyright. No Bengali দেনাপাওনা exists on the Internet Archive, so that novel is simply absent from this corpus.

Schema

column	type	meaning
text	string	the training field: NFC Bengali prose, paragraphs split by \n\n
id	string	stable record id, <dataset>/<work>/<unit>
work_bn / work_en / work_id	string	the work this record belongs to
collection_bn / collection_en	string	set for stories in a collection, else null
unit_type	string	chapter or story
unit_index	int32	position within the work
n_chars / n_words / n_tokens_est	int32	size; n_tokens_est is null if no tokenizer was configured
first_published / edition_year	int32	year of the work vs. of the scanned printing
ia_identifier / source_url / source_pdf	string	provenance
ocr_engine / ocr_params / ocr_source	string	how the text was produced
conf_mean	float32	mean OCR word confidence (0-100), null if unknown
conf_p10	float32	10th-percentile page confidence — catches local damage a mean hides
low_conf_word_frac	float32	fraction of words scored below 70
pages	list<struct>	per source page: page_index, printed_page, char_start, char_end, conf_mean, low_conf_word_frac

conf_* columns are null, never 0.0, when confidence is genuinely unknown — a fabricated zero would read as worst-possible quality and silently drop good text.

Filtering by quality

from datasets import load_dataset

ds = load_dataset("protibimbo/SrikantaVerse", split="train")

# 1. drop whole records that are mostly noise
good = ds.filter(lambda r: (r["conf_mean"] or 0) >= 75.0)

# 2. or excise only the bad pages, keeping the rest of the chapter
def drop_bad_pages(rec, floor=70.0):
    keep = [rec["text"][p["char_start"]:p["char_end"]]
            for p in rec["pages"] if (p["conf_mean"] or 0) >= floor]
    return {"text": "".join(keep)}

clean = ds.map(drop_bad_pages)

The pages spans are contiguous and tile text exactly, so slicing and rejoining them never loses or duplicates characters.

Intended use

Continued pretraining and LoRA/PEFT style tuning on next-token prediction — the records are raw prose, not instruction pairs. Text is not pre-chunked to a token window: packing to 2048/4096 with an EOS at document boundaries is a training-time decision, and pre-chunking would sever sentences and take that choice away. n_tokens_est is provided so you can plan packing without tokenizing first.

There is a single train split. For style evaluation, hold out a work yourself (filtering on work_id) so the held-out text is genuinely unseen.

# hold out one novel for evaluation
train = ds.filter(lambda r: r["work_id"] != "aranyak")
eval_ = ds.filter(lambda r: r["work_id"] == "aranyak")

How the text was produced

1. Extract — text and per-word confidence are read from the Internet Archive hOCR derivative of each scan. The PDF's own invisible text layer carries the same characters but no confidence, so hOCR is preferred wherever an item exists.
2. Normalise — Unicode NFC. Zero-width joiners are preserved: in Bengali they carry the reph / ya-phala distinction, so stripping them would change words.
3. Strip page furniture — running headers and page numbers are removed. The same running title is OCR'd differently on nearly every page, so headers are detected by fuzzy-clustering candidate lines rather than by exact match.
4. Reflow — printed line breaks are rejoined into paragraphs, including across page boundaries. Paragraphs are separated by \n\n.
5. Segment — split into works by verified page ranges, then into chapters by printed headings.

Limitations

• OCR is uncorrected (tesseract 5.0.0-alpha-20201231-10-g1236). Mean record confidence is 79.4/100, and 46 of 204 records fall below 75. Expect wrong conjuncts, stray Latin characters and mangled punctuation. The text is good enough to learn style and syntax from, and not a reliable edition to quote from.
• Front matter, editorial prefaces and critical appendices were excluded by page range, but header and page-number removal is heuristic and a few survive mid-text.
• Paragraph boundaries are reconstructed from printed line breaks and are approximate; dialogue segments more reliably than continuous narration.
• Chapter splits come from printed headings. Books whose headings did not survive OCR appear as a single long record rather than being split arbitrarily.
• Scans are of mid-century reprints, not first editions, so spelling reflects the printing house's conventions of the day.
• Single author and single genre: this is literary prose from one writer, and is not a general-purpose Bengali corpus.

Provenance and rights

Scans and OCR come from the Internet Archive; see ia_identifier and source_url on each record, and the Sources table above.

Cataloguing note: Internet Archive/DLI author fields for this collection are not always reliable. Attribution here was checked against each scan's own title page, not taken from the metadata record.

Public domain in India: the author died in 1938, so copyright expired in 1998 under life + 60.

US status is clear for most of the corpus and open for two works. Works first published before 1929 are in the US public domain; works first published in 1929 or later may have had US copyright restored under the URAA. Filtering first_published >= 1929 isolates exactly those two — Srikanta Part Four (1933) and Shesher Parichay (1939) — leaving the rest unaffected. Confirm before relying on either.

first_published is the year of first publication in book form, transcribed from the গ্রন্থ-পরিচিতি printed in the শরৎ রচনাবলী. Serial publication in periodicals is generally one to two years earlier, so a work may have appeared before the year given here. It is null for Mejdidi, Darpachurna and Andhare Alo, whose entry records only serial dates, and the single date not taken from that bibliography is Shesher Parichay's, which is its posthumous book publication.