Home ➤ Publications ➤ kasige2025hybrid

Hybrid Approach for Information Retrieval in Sri Lankan Legal Domain

Niduni Kasige, Nishan Kavinda, Kisara Kodithuwakku, Nisansa de Silva
Proceedings of the Engineering Research Unit Symposium

Accessing a country’s legal domain is challenging, due to disparities in legal domain literacy among individuals and finding fragmented legal resources. This study addresses this predicament by developing an accessible information retrieval system tailored to users irrespective of their knowledge level in the legal domain. This solution employs a hybrid retrieval approach backed by keyword search retrieval with BM25 and semantic search retrieval with Legal-BERT, complemented by extractive summarization. This architecture enables users to identify the relevant documents based on their specific requirements. In addition to that, this research focusses on providing more context-aware responses, by Gemma-3 4B parameter model for domain-specific adaptation. Beyond the English language based implementation, this study aims to extend this framework’s capabilities to support legal queries in Sinhala language, which is a crucial yet unresolved problem in both legal and natural language processing domains. The proposed solution aims to bridge the gap between legal information accessibility and linguistic diversity, contributing toward a more inclusive and efficient legal information retrieval system.

Keywords: Natural Language Processing | Machine Learning / Deep Learning | Sinhala | Law | Information Retrieval | Semantic Similarity | 

@inproceedings{kasige2025hybrid,
  title={{Hybrid Approach for Information Retrieval in Sri Lankan Legal Domain}},
  author={Kasige, Niduni and Kavinda, Nishan and Kodithuwakku, Kisara and de Silva, Nisansa},
  booktitle={Proceedings of the Engineering Research Unit Symposium},
  year={2025},
  doi={10.31705/ERU.2025.33},
  abstract={Accessing a country’s legal domain is challenging, due to disparities in legal domain literacy among individuals and finding fragmented legal resources. This study addresses this predicament by developing an accessible information retrieval system tailored to users irrespective of their knowledge level in the legal domain. This solution employs a hybrid retrieval approach backed by keyword search retrieval with BM25 and semantic search retrieval with Legal-BERT, complemented by extractive summarization. This architecture enables users to identify the relevant documents based on their specific requirements. In addition to that, this research focusses on providing more context-aware responses, by Gemma-3 4B parameter model for domain-specific adaptation. Beyond the English language based implementation, this study aims to extend this framework’s capabilities to support legal queries in Sinhala language, which is a crucial yet unresolved problem in both legal and natural language processing domains. The proposed solution aims to bridge the gap between legal information accessibility and linguistic diversity, contributing toward a more inclusive and efficient legal information retrieval system.},
  pages={70-71},
  misc={https://dl.lib.uom.lk/server/api/core/bitstreams/2d222bc1-004e-40f3-a0a4-a15e8893998a/content,NLP:ML:SIN:LAW,https://goo.gl/iY6aTr}
}

Related Projects

SigmaLaw

Active

PI: Nisansa de Silva

This project delves into the synergy between Natural Language Processing (NLP) and the legal domain, presenting novel solutions to enduring challenges.