Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Towards Estimating Personal Values in Song Lyrics

Andrew M. Demetriou, Jaehun Kim, Sandy Manolios, Cynthia C. S. Liem

TL;DR

The paper addresses estimating perceived personal values in song lyrics by combining a perspectivist human-annotation approach with automated scoring anchored to Schwartz's value theory. It introduces fuzzy stratified sampling to assemble a representative lyric corpus, collects multi-rater ground-truth annotations from a US population, and uses Robust Ranking Aggregation to produce ground-truth value rankings per song. Automated estimates are generated via multiple word-embedding and dictionary-based methods, with correlations to human-ground-truth rankings indicating meaningful signals though not yet a definitive benchmark. Overall, the work demonstrates feasibility for automated lyric-valued estimations and highlights practical implications for personalized music information retrieval, while outlining clear avenues for scaling and methodological improvement.

Abstract

Most music widely consumed in Western Countries contains song lyrics, with U.S. samples reporting almost all of their song libraries contain lyrics. In parallel, social science theory suggests that personal values - the abstract goals that guide our decisions and behaviors - play an important role in communication: we share what is important to us to coordinate efforts, solve problems and meet challenges. Thus, the values communicated in song lyrics may be similar or different to those of the listener, and by extension affect the listener's reaction to the song. This suggests that working towards automated estimation of values in lyrics may assist in downstream MIR tasks, in particular, personalization. However, as highly subjective text, song lyrics present a challenge in terms of sampling songs to be annotated, annotation methods, and in choosing a method for aggregation. In this project, we take a perspectivist approach, guided by social science theory, to gathering annotations, estimating their quality, and aggregating them. We then compare aggregated ratings to estimates based on pre-trained sentence/word embedding models by employing a validated value dictionary. We discuss conceptually 'fuzzy' solutions to sampling and annotation challenges, promising initial results in annotation quality and in automated estimations, and future directions.

Towards Estimating Personal Values in Song Lyrics

TL;DR

The paper addresses estimating perceived personal values in song lyrics by combining a perspectivist human-annotation approach with automated scoring anchored to Schwartz's value theory. It introduces fuzzy stratified sampling to assemble a representative lyric corpus, collects multi-rater ground-truth annotations from a US population, and uses Robust Ranking Aggregation to produce ground-truth value rankings per song. Automated estimates are generated via multiple word-embedding and dictionary-based methods, with correlations to human-ground-truth rankings indicating meaningful signals though not yet a definitive benchmark. Overall, the work demonstrates feasibility for automated lyric-valued estimations and highlights practical implications for personalized music information retrieval, while outlining clear avenues for scaling and methodological improvement.

Abstract

Most music widely consumed in Western Countries contains song lyrics, with U.S. samples reporting almost all of their song libraries contain lyrics. In parallel, social science theory suggests that personal values - the abstract goals that guide our decisions and behaviors - play an important role in communication: we share what is important to us to coordinate efforts, solve problems and meet challenges. Thus, the values communicated in song lyrics may be similar or different to those of the listener, and by extension affect the listener's reaction to the song. This suggests that working towards automated estimation of values in lyrics may assist in downstream MIR tasks, in particular, personalization. However, as highly subjective text, song lyrics present a challenge in terms of sampling songs to be annotated, annotation methods, and in choosing a method for aggregation. In this project, we take a perspectivist approach, guided by social science theory, to gathering annotations, estimating their quality, and aggregating them. We then compare aggregated ratings to estimates based on pre-trained sentence/word embedding models by employing a validated value dictionary. We discuss conceptually 'fuzzy' solutions to sampling and annotation challenges, promising initial results in annotation quality and in automated estimations, and future directions.
Paper Structure (11 sections, 8 figures)

This paper contains 11 sections, 8 figures.

Table of Contents

  1. Introduction
  2. Personal Values
  3. Primary Lyrics Data
  4. Fuzzy Stratified Sampling
  5. Ground-Truthing Procedure
  6. Reliability, Agreement and Initial Validation
  7. Automated Scoring
  8. Descriptive Analyses
  9. Limitations and Future Work
  10. Conclusion
  11. Ethics Statement

Figures (8)

  • Figure 1: Distribution of self-reported percentage of music library containing lyrics from two representative US samples, n=505 and n=600 respectively.
  • Figure 2: Visualization of the Schwartz 10-value inventory from schwartz1992universals used in this paper, such that more abstract values of Conservation, vs. Openness to Change, and Self-transcendence vs. Self-enhancement form 4 higher-order abstract values. Illustration adapted from maio2010mental.
  • Figure 3: MDS plots derived from the correlation plot reported in schwartz2001extending, and our participant responses as confidence-weighted means
  • Figure 4: Rank correlations between NLP systems / word counts and Robust Ranking Aggregation lists, by normalization scheme.
  • Figure 5: Rank correlations between word2vec scores Robust Ranking Aggregation lists, per genre grouping operationalized as Artist Tag Topic.
  • ...and 3 more figures