Hybrid approach for information retrieval in Sri Lankan legal domain

Abstract

Information retrieval in the legal domain has become a significant research area due to the specific nature of the language in the legal domain [1]. This complexity is further exacerbated in contexts such as Sri Lanka, where English literacy rates among individuals aged 15 and above are approximately 22%, posing a significant barrier to comprehending intricate legal documentation [2]. Current solutions present a trade-off: traditional keyword searches require technical vocabulary that lay users lack, whereas generic LLMs offer conversational ease but frequently 'hallucinate' or misrepresent local statutes [3]. To address this, we propose a hybrid system combining precise document retrieval with a domain-adapted LLM [4]. Our approach finetunes Gemma-3-4B-IT [5] on Sri Lankan legal data [6], combining exact statutory retrieval with simplified summaries for lay audiences.