An automated system to identify pertinent legal objectives, actors and objects from SEBI regulatory documents.
Shravya Kanchi, Pulkit Parikh, Kamal Karlapalem, Sandeep Dash, Vrinda Lakhotia
Lawyers and the banking community require a tools to quickly sort, understand and traverse through innumerable number of regulation documents and SEBI Acts. Systems that make interpretations, find similarities/dissimilarities between any two rules(or regulations), connect cases to regulations described and understand parallels between SEBI regulations with other regulatory authorities(ex: SEC) need robust named entity recognisers specific to identify SEBI terminology. Though a lot of existing NERs output flat entities(ie. non-overlapping mentions) from the text, overlapping entities reveal greater semantics and relationships about the text. We proposed a scheme of named entities crucial for tasks related to penalty estimation and regulation estimation. Our schema of entities when applied on the SEBI regulations found that that about 13% of SEBI entities overlap with each other. Existing literature on nested named entity recognition shows that question answering techniques combined with BERT produce the state-of-the-art results. Our proposal is to combine our specialised SEBI-BERT and proposed tags to identify nested named entities on SEBI text. Further, we plan to extend this work by populating an ontology specific to SEBI. We also plan to semantically compare the legal cases with to retrieve cross referenced(or related) regulations. Also, we proposed to identify and annotate deontic(levels of duty/obligation) semantics of SEBI rules as an input to higher level systems like summarization systems, question answering models.
Data: HERE
Code: HERE
DEMO: HERE
etc.: Information Slides