Lyric Mining: Word, Rhyme & Concept Co-occurrence Analysis

Computational creativity is one area of NLP which requires extensive analysis of large datasets. Laalalaa [1] framework for Lyric analysis and generation proposed a lyric analysis subsystem that required statistical analysis of Tamil lyrics. In this paper, we propose a data analysis model for words, rhymes and their usage in Tamil lyrics. The proposed analysis model extracts the root words from lyrics using a morphological analyzer [2] to compute the word frequency across the lyric dataset. The words in their unanalyzed form are used for computing the frequent rhyme, alliteration and endrhyme pairs using adapted apriori algorithm. Frequent co-occurring concepts in lyrics are also computed using Agaraadhi, an on-line Tamil dictionary. Presenting the results, this paper concludes by discussing the need of such an analysis to compute freshness, pleasantness of a lyric and using these statistics for Lyric Generation.