Automatic Tamil lyric generation based on ontological interpretation for semantics

Abstract

This system proposes an N-gram based approach to automatic Tamil lyric generation, by the ontological semantic interpretation of the input scene. The approach is based on identifying the semantics conveyed in the scenario, thereby making the system understand the situation and generate lyrics accordingly. The heart of the system includes the ontological interpretation of the scenario, and the selection of the appropriate tri-grams for generating the lyrics. To fulfill this, we have designed a new ontology with weighted edges, where the edges correspond to a set of sentences, which indicate a relationship, and are represented as a tri-gram. Once the appropriate tri-grams are selected, the root words from these tri-grams are sent to the morphological generator, to form words in their packed form. These words are then assembled to form the final lyrics. Parameters of poetry like rhyme, alliteration, simile, vocative words, etc., are also taken care of by the system. Using this approach, we achieved an average accuracy of 77.3% with respect to the exact semantic details being conveyed in the generated lyrics.