Modeling Discourse Segments in Lyrics Using Repeated Patterns

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1959–1969, Osaka, Japan, December 11-17 2016. Modeling Discourse Segments in Lyrics Using Repeated Patterns Kento Watanabe1, Yuichiroh Matsubayashi1, Naho Orita1, Naoaki Okazaki1, Kentaro Inui1, Satoru Fukayama2, Tomoyasu Nakano2, Jordan B. L. Smith2, Masataka Goto2 1Graduate School of Information Sciences, Tohoku University 2National Institute of Advanced Industrial Science and Technology (AIST) {kento.w,y-matsu,naho,okazaki,inui}@ecei.tohoku.ac.jp, {s.fukayama,t.nakano,jordan.smith,m.goto}@aist.go.jp Abstract This study proposes a computational model of the discourse segments in lyrics to understand and to model the structure of lyrics. To test our hypothesis that discourse segmentations in lyrics strongly correlate with repeated patterns, we conduct the first large-scale corpus study on discourse segments in lyrics. Next, we propose the task to automatically identify segment boundaries in lyrics and train a logistic regression model for the task with the repeated pattern and textual features. The results of our empirical experiments illustrate the significance of capturing repeated patterns in predicting the boundaries of discourse segments in lyrics. 1 Introduction Lyrics are an important element of popular music. They provide an effective means to express the message and emotion of music. Similar to prose text, lyrics are a discourse: i.e., they are, typically, a sequence of related lines, rather than an unconnected stream of lines of arbitrary order. Thus, like texts, lyrics also have a discourse structure consisting discourse segments. Each discourse segment exhibits an individual topic in discourse, and the transition of topics over successive discourse segments constitutes a flow (Austin et al., 2010; Watanabe et al., 2014). Unlike prose text, lyrics have their own peculiar properties, such as frequent repetition of identical or similar phrases and extensive use of rhyme and refrain (Austin et al., 2010), as illustrated in Figure 1. Analogous to prose text, the lyrics in Figure 1 can be viewed as a sequence of discourse segments, wherein each segment is depicted by a colored box. Segments 1⃝ and 3⃝ appear repeatedly (e.g., segments 4⃝ and 8⃝ are identical to segment 1⃝), which is not typically observed in prose text. Our goal is to reveal the discourse structure of lyrics in popular music by quantitatively analyzing a large-scale lyrics corpus. The motivations behind the goal are as follows. First, there are hardly any studies that have focused on the data-oriented research of the discourse structure of lyrics; however, multiple studies have focused on the structure of music audio (Paulus et al., 2010, for a review). Second, a better understanding of the nature of lyrics structure combined with the existing theories of music audio could lead to a more comprehensive theory of the overall nature of popular music. Third, understanding the nature of lyrics structure will allow us to devise a variety of useful computer systems, such as systems that automatically generate lyrics, assist human lyrics writers, or evaluate the quality of lyrics. As a first but crucial step toward achieving this goal, this study explores the nature of the discourse structure of lyrics with a focus on the rbyof repeated …

Content truncated to 500 words