Entity and Verb Semantic Role Labelling for Tamil Biomedicine

摘要

The primary task of Semantic Role Labelling (SRL) is to indicate exactly what semantic relations hold among a predicate and its associated participants. This type of role labelling yields a first level semantic representation of the text in question. Since the field of computation in Tamil Biomedicine is rather unexplored. SRL is introduced to label the named entities with specific roles in the given domain. In contrast to many state-of-the-art SRL systems, we devise a new approach to define roles to predicate terms along with its constituent terms. In order to achieve this, a MEM based classifier model is built using the features obtained from parsed input sentences. The parsing is done on a syntactic level and a dependency parse tree is built. The classifier model is further strengthened by verb frame training, as their probability give an extra edge to determine verb roles. The MEM model is compared with linear classifiers such as SVM and Linear Regression classifier and is found to perform better than the others.