Somtharu181coder/Legal_domain_ocr_extracted_Nepali_sft_dataset

Nepali Legal SFT Dataset — Software Development & Operation Committee Order, 2083

Dataset Summary

This dataset contains 31 single-turn instruction/response pairs in Nepali (Devanagari script), derived from a single Government of Nepal legal instrument:

सफ्टवेयर विकास तथा सञ्चालन समिति (गठन) आदेश, २०८३ (Software Development and Operation Committee (Formation) Order, 2083)

The order was issued by the Government of Nepal under Section 3 of the विकास समिति ऐन, २०१३ (Development Committee Act, 2013), establishing a national Software Development and Operation Committee and an executing Center under it, responsible for standardizing how public agencies commission, build, maintain, and secure software systems.

Each row pairs a natural-language Nepali question or instruction with a grounded, synthesized answer drawn from the order's provisions (definitions, committee formation and powers, meeting procedure, center functions, the software-request/procurement workflow, payment terms, data ownership, infrastructure, quality/security testing, executive director appointment and duties, staffing, expert selection, code of conduct, the committee's fund, audit, delegation of authority, reporting obligations, rule-making power, and the three annexed forms/schedules).

Source & Provenance

• Original document: OCR-extracted text from a scanned copy of the official order (16 pages, 504 raw OCR fragments with position/confidence metadata), as published by the Nepal Law Commission (lawcommission.gov.np), the official repository of Nepal Government law.
• Processing: The raw OCR fragments (out-of-order text blocks per page) were reordered and semantically merged into coherent section-level text, then used to author instruction/response pairs covering the document's substantive content.
• Content vs. dataset creation: The underlying legal text is a Government of Nepal public instrument. The conversations field (questions and answers) in this dataset is synthetically generated from that source text — it is not scraped dialogue and was not produced by human crowdworkers. See generation_type: synthetic on every row.

Supported Tasks

• Instruction-following / conversational QA in Nepali
• Domain adaptation for Nepali legal and government-administration language
• Low-resource-language (Nepali / npi) instruction-tuning data augmentation

Languages

• Nepali (ne / npi), Devanagari script (Deva). All questions and answers are written entirely in Nepali.

Dataset Structure

Data Instance

{
  "id": "sg_cf7c0925012d4ac184a93805e46c16c6",
  "conversations": [
    { "from": "human", "value": "सफ्टवेयर विकास तथा सञ्चालन समिति (गठन) आदेश, २०८३ को संक्षिप्त नाम र यो कहिलेदेखि लागू हुने भन्ने बारेमा बताउनुहोस्।" },
    { "from": "gpt", "value": "यस आदेशको संक्षिप्त नाम \"सफ्टवेयर विकास तथा सञ्चालन समिति (गठन) आदेश, २०८३\" रहेको छ। ..." }
  ],
  "source": "lawcommission.gov.np:sfotware-vikash-samiti-gathan-aadesh-2083:train",
  "source_name": "np_legal_software_dev_committee_order_2083",
  "source_repo": "lawcommission.gov.np",
  "source_config": "sfotware-vikash-tatha-sanchalan-samiti-gathan-aadesh-2083",
  "source_split": "train",
  "source_revision": "2083-05-03",
  "source_row_id": "1788921746646_pdf_ocr_extracted.json:p1:1",
  "language": "ne",
  "language_code": "npi",
  "script": "Deva",
  "license": "cc-by-4.0",
  "license_tier": "permissive",
  "task_type": "instruction-following",
  "generation_type": "synthetic",
  "condition": "model-generated,derived-from-legal-document",
  "url": "https://www.lawcommission.gov.np",
  "metadata_json": "{\"annotation_type\": \"synthetic-generation-from-ocr\", \"original_language\": \"Nepali\", \"document_title\": \"...\", \"document_type\": \"government-formation-order\", \"issuing_authority\": \"Government of Nepal, Office of the Prime Minister and Council of Ministers\", \"ocr_source_file\": \"1788921746646_pdf_ocr_extracted.json\", \"page_reference\": \"p1\"}"
}

Data Fields

Field	Type	Description
id	string	Unique row identifier, sg_<uuid4hex>
conversations	list of {from, value}	Single-turn exchange: one human turn, one gpt turn, both in Nepali
source	string	Composite source identifier (repo:config:split)
source_name	string	Short internal dataset/document name
source_repo	string	Publishing body / repository (lawcommission.gov.np)
source_config	string	Slugified document name
source_split	string	Data split (train)
source_revision	string	Publication date of the source order (Bikram Sambat, YYYY-MM-DD)
source_row_id	string	Traceability key: <ocr_source_file>:<page_reference>:<row_index>
language	string	ISO 639-1 language (ne)
language_code	string	ISO 639-3 language code (npi)
script	string	ISO 15924 script code (Deva)
license	string	cc-by-4.0
license_tier	string	permissive
task_type	string	instruction-following
generation_type	string	synthetic — conversations were authored from the source document, not human-crowdsourced
condition	string	Provenance tags: model-generated,derived-from-legal-document
url	string	Publisher URL
metadata_json	string (JSON)	Extra provenance: document title, type, issuing authority, OCR source filename, page reference

Data Splits

Split	Examples
train	31

Dataset Creation

Curation Rationale

Nepali is a low-resource language for instruction-tuning data, and government/legal registers are especially underrepresented. This dataset packages a single, complete, freely-reusable Nepali legal instrument into clean instruction/response pairs suitable for SFT.

Source Data

• Initial data: A scanned PDF of the official order, OCR-processed into 504 text/table/header/footer fragments with page number, bounding box, and OCR confidence per fragment.
• Processing steps:
  1. Fragments were grouped by page and reordered by vertical/horizontal position to reconstruct reading order.
  2. Adjacent fragments were semantically merged into coherent clauses and sections (दफा १ – दफा २९, अनुसूची १–३).
  3. Instruction/response pairs were authored in Nepali against the reconstructed text, one or more per statutory section or schedule, plus one whole-document overview pair.

Annotations

All conversations content is synthetically generated (generation_type: synthetic) directly from the source legal text — no human annotators/crowdworkers were used to write the questions or answers.

Personal and Sensitive Information

None. The source is a public government administrative order; it contains no personal data about private individuals.

Considerations for Using the Data

• Not legal advice. This is a legal text-comprehension dataset for language modeling purposes, not a substitute for professional legal counsel or the official gazette text.
• Recency: The source order was approved by Nepal's Council of Ministers in Bhadra 2083 (~August 2026). It is a recent instrument and may be amended; this dataset reflects the order's text as OCR-captured at time of processing and does not track subsequent amendments.
• OCR noise: The source PDF's OCR had variable confidence (particularly on numerals/table cells). Reconstructed section text was manually disambiguated where fragments were garbled; residual transcription risk exists for the two staffing/expert-roster schedule tables (Schedules 2 and 3), which had the noisiest OCR.
• Scale: With 31 rows, this dataset is intended as a small, high-precision seed set — for production SFT it should be mixed with broader Nepali instruction data rather than used alone.

Licensing Information

Distributed under CC BY 4.0 (license_tier: permissive). Users must provide appropriate attribution to the Nepal Law Commission (lawcommission.gov.np) as the publisher of the underlying legal text. Government legal instruments are generally treated as public administrative record in Nepal; downstream users redistributing this dataset should independently confirm applicable government-works licensing terms for their jurisdiction of use.

Citation

@misc{np_software_committee_order_2083_sft,
  title  = {Nepali Legal SFT Dataset: Software Development \& Operation Committee (Formation) Order, 2083},
  note   = {Derived from OCR of a Nepal Law Commission (lawcommission.gov.np) publication},
  year   = {2026},
  license = {CC-BY-4.0}
}