Long Paper accepted at the AACL-IJCNLP 2026 (Findings)!
InLegalLaySum (Indian Legal Lay Summarization), a dataset comprising 9,570 judgements from the Supreme Court and several High Courts of India, paired with expert-written summaries in English.
The full dataset can be found under https://huggingface.co/datasets/trishitasaha/InLegalLaySum .
Folder Structure -
InLegalLaySum/
├── InLegalLaySum_Train
└── <id1>/
├── Judgement.txt # Original case judgement
├── Expert_Summary.txt # Summary of judgement written by law practitioner
└── Simplified_Summary.txt # Summary generated by Gemini-2.0-Flash
├── InLegalLaySum_Test
└── <id1>/
├── Judgement.txt # Original case judgement
└── Expert_Summary.txt # Summary of judgement written by law practitioner
InLegalLaySum_Train
It contains total 7470 samples.
InLegalLaySum_Test
It contains total 2100 samples.
| File Name | Description |
|---|---|
| GPT_Generate_Summary.py | Generates simplified summary from the expert-written summary |
| SFT.py | Supervised Fine Tuning |
| DPO_on_SFT.py | Direct Preference Optimization applied on the SFT model |
| Inference.py | Generates the final summary |
| File Name | Description |
|---|---|
| Generation of Lay Summary.txt | Prompt for generating simplified lay summary from expert-written summary |
| Inference.txt | Prompt for inference |
| LLM_As_Judge.txt | Prompt for LLM evaluating as a Judge |
| ZSI_Pipeline_Summarization.txt | Prompt for summarization in ZSI_Pipeline method |
| ZSI_Pipeline_Simplification.txt | Prompt for simplification in ZSI_Pipeline method |
| File Name | Description |
|---|---|
| Readability.py | Calculates readability scores - GFI, FKGL, FRE, DC, CLI |
| Rouge.py | Calculates quality scores - ROUGE-1,2,L |
| Bert.py | Calculates the Bert score |
| SARI.py | Calculates the simplification SARI score |
| LLM_As_Judge.py | Summary evaluation by an LLM performing as a Judge |
The InLegalLaySum dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. To view a copy of this license, visit CC BY-NC-SA 4.0.
If you use this dataset, please cite the following paper:
@inproceedings{saha-etal-2026,
title = "InLegalLaySum: A Novel Benchmark for Lay Summarization of Indian Legal Documents",
author = "Saha, Trishita and
Datta, Debtanu and
Tiwari, Shubham and
Kaushal, Mohit and
Rathore, Priyank and
Pandit, Mohammad Khalid and
Goyal, Pawan and
Ghosh, Saptarshi",
booktitle = "Proceedings of the 5th Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 15th International Joint Conference on Natural Language Processing",
month = nov,
year = "2026",
address = "Online only",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2022.aacl-main.77/",
abstract = "Legal documents are often lengthy and contain domain-specific terminology and complex legal reasoning, making them difficult for the general public to understand. This challenge is further
amplified in the Indian context, where a large portion of the population lacks proficiency in complex English. Although automatic legal summarization has been studied, most existing approaches generate summaries intended
for legal professionals and thus remain difficult for laypeople. To address this gap, we introduce InLegalLaySum, the first benchmark for lay summarization of Indian legal documents. InLegalLaySum comprises 9,570 judgements from the Supreme Court and several High Courts of India paired with expert-written summaries in English. We benchmark a wide range of generative models (ranging from 160M to 7B parameters) and perform comprehensive experiments with multiple training strategies, including supervised fine-tuning (SFT) and direct preference optimization (DPO). We observed that the hybrid SFT+DPO training achieved superior performance across readability metrics, which is further validated by domain experts."
}
For any inquiries, feedback, or collaboration opportunities, please contact to {trishitasaha[dot]kgp[at]gmail[dot]com}.