Semantic role labelling for information extraction of tamil biomedicine

Abstract

newline newlineTamil Siddha medicine an ancient medicinal system has yielded us newlinea wide range of untapped information about traditional medicines and their uses in a wide spectra of ailments The collection of such vast data provides some interesting opportunities for computational deployment In this work we propose to explore into the various Natural Language Processing techniques that can be implemented to this syntactically rich corpus The richness of the corpus can be understood only after a detailed analysis of its word structure and composition Hence we explore our collection of Siddha documents with medicinal information to identify aspects of it such as the type of documents statistical distribution of words characters syntactic features roles of noun phrases and verb frames sentence pattern and ultimately the information offered The impact of ambiguous words and non- Tamil words are also studied After the detailed analysis of the corpus in hand it is inferred that the domain information mostly concentrates on the central concepts Hence this work starts by identifying the Named Entities (NE) and categorizing them For any system of medicine the two broad categories around which data is presented is Medicine and Diseases That being the case an integrated Named Entity Recognition and Classification model is built to identify entities belonging to these classes with the help of syntax and frequency oriented features The model exhibits an accuracy as high as any state-of-the-art models newlinewith minimum computational complexities newline