Lyrics, Music, and Emotions

Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 590–599, Jeju Island, Korea, 12–14 July 2012. c⃝2012 Association for Computational Linguistics Lyrics, Music, and Emotions Rada Mihalcea University of North Texas rada@cs.unt.edu Carlo Strapparava FBK-irst strappa@fbk.eu Abstract In this paper, we explore the classification of emotions in songs, using the music and the lyrics representation of the songs. We intro- duce a novel corpus of music and lyrics, con- sisting of 100 songs annotated for emotions. We show that textual and musical features can both be successfully used for emotion recog- nition in songs. Moreover, through compar- ative experiments, we show that the joint use of lyrics and music brings significant improve- ments over each of the individual textual and musical classifiers, with error rate reductions of up to 31%. 1 Introduction Language and music are peculiar characteristics of human beings. The capability of producing and enjoying language and music appears in every hu- man society, regardless of the richness of its culture (Nettl, 2000). Importantly, language and music complement each other in many different ways. For instance, looking at music and language in terms of fea- tures, we can observe that music organizes pitch and rhythm in ways that language does not, and it lacks the specificity of language in terms of seman- tic meaning. On the other hand, language is built from categories that are absent in music (e.g., nouns and verbs), whereas music seems to have a deeper power over our emotions than does ordinary speech. Composers, musicians, and researchers in poetry and literature alike have been long fascinated by the combination of language and music, even since the time of the earliest written records of music encoun- tered in musical settings for poetry. Despite this in- terest, and despite the long history of the interaction between music and lyrics, there is only little work that explicitly focuses on the connection between music and lyrics. In this paper, we focus on the connection between the musical and linguistic representations in popu- lar songs, and their role in the expression of affect. We introduce a novel corpus of lyrics and music, an- notated for emotions at line level, and explore the automatic recognition of emotions using both tex- tual and musical features. Through comparative ex- periments, we show that emotion recognition can be performed using either textual or musical features, and that the joint use of lyrics and music can im- prove significantly over classifiers that use only one dimension at a time. We believe our results demon- strate the promise of using joint music-lyric models for song processing. 2 Related Work The literature on music analysis is noticeably large, and there are several studies concerning the music’s power over emotions (Juslin and Sloboda, 2001), thinking (Rauscher et al., 1993), or physical effort (Karageorghis and Priest, 2008). In particular, there has been significant research in music and psychology focusing on the idea of a parallel between affective cues …

Content truncated to 500 words