| --- |
| language: |
| - en |
| - hi |
| - mr |
| license: apache-2.0 |
| pretty_name: IndianLegal-QA |
| tags: |
| - question-answering |
| - legal |
| - indian-law |
| - nlp |
| - multilingual |
| - indian-constitution |
| - ipc |
| - crpc |
| - customs |
| - tariff |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: question_answers.csv |
| --- |
| |
| # IndianLegal-QA |
|
|
| A question-and-answer dataset derived from Indian legal and government documents, |
| covering the Constitution of India, the Indian Penal Code, criminal and civil |
| procedural law, customs and tariff classifications, and numerous central and |
| state acts. The dataset is suitable for building, fine-tuning, and evaluating |
| retrieval and question-answering systems over Indian legal text. |
|
|
| This dataset is also hosted on GitHub at `Sakib-Dalal/IndianLegal-QA`. |
|
|
| ## Overview |
|
|
| Each record is a paired question and answer extracted from an Indian legal source |
| document. The questions ask factual questions about provisions, definitions, |
| classifications, and procedures found in the source material, and the answers |
| provide the corresponding textual response. Records use a simple two-field schema |
| (`question`, `answer`). |
|
|
| ## Statistics |
|
|
| | Property | Value | |
| | ------------------- | --------------------------------- | |
| | Number of pairs | 120,640 | |
| | Fields per record | `question`, `answer` | |
| | Languages | English, Hindi, Marathi | |
| | Multilingual pairs | ~23,630 (Devanagari script) | |
| | Formats | CSV, JSONL, plain text | |
| | License | Apache-2.0 | |
|
|
| ## Data format |
|
|
| The dataset is provided as `question_answers.csv` (comma-separated pairs, one per |
| row). The same content is also distributed on GitHub in `question_answers.jsonl` |
| and `question_answers.txt`. Each example has two string fields: |
|
|
| | Field | Description | |
| | ---------- | ---------------------------------- | |
| | `question` | The question about a legal provision | |
| | `answer` | The corresponding answer | |
|
|
| ## Usage |
|
|
| ```python |
| from datasets import load_dataset |
| |
| ds = load_dataset("Sakib-Dalal/IndianLegal-QA") |
| print(ds["train"]) |
| print(ds["train"][0]) |
| ``` |
|
|
| For local use, the CSV can be read with pandas: |
|
|
| ```python |
| import pandas as pd |
| |
| df = pd.read_csv("question_answers.csv") |
| ``` |
|
|
| ## Domain coverage |
|
|
| The pairs draw from a wide range of Indian legal sources. Approximate coverage by |
| topic: |
|
|
| | Domain | Indicative count | |
| | -------------------------- | ---------------- | |
| | Acts (central & state) | ~25,550 | |
| | Sections of enactments | ~13,916 | |
| | Schedules | ~2,155 | |
| | Articles of the Constitution| ~1,552 | |
| | Constitutional provisions | ~983 | |
| | Customs & tariff headings | ~550 | |
| | Indian Penal Code (IPC) | ~88 | |
| | Criminal Procedure Code | ~22 | |
|
|
| The dataset spans the Constitution of India, IPC and CrPC, customs and tariff |
| classification headings, civil and criminal procedure, and a broad set of central |
| and state acts. A number of records are provided in Hindi and Marathi, making the |
| dataset useful for multilingual legal NLP. |
|
|
| ## Example |
|
|
| ```text |
| Q: What is described as the flexibility of the Constitution? |
| A: Its flexibility lies in its amendments. |
| ``` |
|
|
| ## License |
|
|
| This dataset is released under the Apache License 2.0. |
|
|
| ## Citation |
|
|
| ```bibtex |
| @dataset{dalal2026indianlegalqa, |
| title = {IndianLegal-QA}, |
| author = {Dalal, Sakib}, |
| year = {2026}, |
| repo = {https://github.com/Sakib-Dalal/IndianLegal-QA}, |
| url = {https://huggingface.co/datasets/Sakib-Dalal/IndianLegal-QA}, |
| } |
| ``` |
|
|
