Learning to Freestyle: Hip Hop Challenge-Response Induction via Transduction Rule Segmentation

Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pages 102–112, Seattle, Washington, USA, 18-21 October 2013. c⃝2013 Association for Computational Linguistics Learning to Freestyle: Hip Hop Challenge-Response Induction via Transduction Rule Segmentation Dekai Wu Karteek Addanki Markus Saers Human Language Technology CenterMeriem Beloucif Department of Computer Science HKUST, Clear Water Bay, Hong Kong {dekai|vskaddanki|masaers|mbeloucif}@cs.ust.hk Abstract We present a novel model, Freestyle, that learns to improvise rhyming and fluent re- sponses upon being challenged with a line of hip hop lyrics, by combining both bottom- up token based rule induction and top-down rule segmentation strategies to learn a stochas- tic transduction grammar that simultaneously learns both phrasing and rhyming associations. In this attack on the woefully under-explored natural language genre of music lyrics, we exploit a strictly unsupervised transduction grammar induction approach. Our task is par- ticularly ambitious in that no use of any a pri- ori linguistic or phonetic information is al- lowed, even though the domain of hip hop lyrics is particularly noisy and unstructured. We evaluate the performance of the learned model against a model learned only using the more conventional bottom-up token based rule induction, and demonstrate the superi- ority of our combined token based and rule segmentation induction method toward gen- erating higher quality improvised responses, measured on fluency and rhyming criteria as judged by human evaluators. To highlight some of the inherent challenges in adapting other algorithms to this novel task, we also compare the quality of the responses generated by our model to those generated by an out-of- the-box phrase based SMT system. We tackle the challenge of selecting appropriate training data for our task via a dedicated rhyme scheme detection module, which is also acquired via unsupervised learning and report improved quality of the generated responses. Finally, we report results with Maghrebi French hip hop lyrics indicating that our model performs surprisingly well with no special adaptation to other languages. 1 Introduction The genre of lyrics in music has been severely under- studied from the perspective of computational lin- guistics despite being a form of language that has perhaps had the most impact across almost all human cultures. With the motivation of spurring further re- search in this genre, we apply stochastic transduc- tion grammar induction algorithms to address some of the modeling issues in song lyrics. An ideal start- ing point for this investigation is hip hop, a genre that emphasizes rapping, spoken or chanted rhyming lyrics against strong beats or simple melodies. Hip hop lyrics, in contrast to poetry and other genres of music, present a significant number of challenges for learning as it lacks well-defined structure in terms of rhyme scheme, meter, or overall meaning making it an interesting genre to bring to light some of the less studied modeling issues. The domain of hip hop lyrics is particularly un- structured when compared to classical poetry, a do- main on which statistical methods have been applied in the past. Hip hop lyrics are unstructured in the …

Content truncated to 500 words