Coreference Resolution for Better Information Retrieval from Indian Classical Music Forums

Information retrieval from music related text is an integral part of Music Information Retrieval(MIR) augmenting content based MIR. Discussion forums on music are rich sources of information gathered from a wider audience. There are a few music forums related to Indian classical music having notable information pertaining to entities including artiste, music concepts including raga, location etc. The forum posts generally contain anaphoric references to the main topic of discussion or to an intermediate mention. Coreference resolution assists resolving anaphoric references thus improving the yield of relation extraction from the posts. In this paper we explore features for supervised approach to coreference resolution felicitous to discourses of aforementioned nature. Along with prevalent features for coreference resolution, we experimented with grammatical features obtained from dependency parsing. Results with Naive Bayes and SVM classifiers are compared along with analysis of different relevant features. The dependency role of the mentions, specifically the mentions to be checked for coreference and other mentions in the vicinity, are found to be relevant for coreference resolution, especially in short discourse of text.