Melodic expectation: The state of the field and some new results (June 18)

Abstract: It has often been said that expectation plays a central role in our experience and enjoyment of music. Many studies have explored the factors affecting musical expectation (especially in melodies) and how it relates to enjoyment. In this talk I survey some important findings and present some new results from my own work. I begin by discussing approaches to the modeling of melodic expectation. Two prominent approaches are the rule-based (or “Gestalt-based”) approach and the statistical (n-gram) approach. I discuss pros and cons of the two approaches, with regard to both experimental support (including two recent studies of my own) and cognitive plausibility. I then turn to research on expectation and enjoyment. A wide range of views have been put forth on this issue; it has been suggested that people enjoy highly expected events, moderately expected events, or unexpected events. A widely discussed idea in the last few years is that enjoyment may be affected not only by expectedness of events but also by our level of uncertainty about them, usually quantified with entropy. I discuss three recent studies that explore this idea; I raise doubts about their methodology and also about the general idea of entropy as a factor in musical enjoyment.  

Anticipatory Syncopation in Popular Music (June 19)

Abstract: It is well-known that popular music in the 20th century (and the 21st) features a much higher degree of syncopation (conflict between accents and meter) than music of previous eras. It is less well-known that syncopation in popular music is subject to strong constraints. In the vast majority of cases, syncopations are _anticipatory_: They are understood as belonging on the following strong beat. This is most evident in vocal music: treating weak-beat stressed syllables as anticipatory allows the usual alignment of meter and stress to be preserved. In this talk I will present a theoretical framework for the study of anticipatory syncopation, examine some of the specific forms it takes, and trace its evolution over the 20th century. While my main focus is on English-language popular music, I will also consider the role of anticipatory syncopation in Korean popular music. 

Between Pattern and Expression:  Gugak at the Intersection of Mathematics and Language 

Abstract: In the context of this conference on mathematics, music, and language, this presentation aims to reconsider how music is constituted in relation to mathematical structure and linguistic expression. Music has been conceptualized, on the one hand, as a system of patterns, proportions, and regularities, and, on the other hand, as a medium of expressive communication that conveys meaning and affect. Mathematical approaches have emphasized the structural dimensions of music—such as repetition, ratio, and formal organization—whereas linguistic approaches have focused on processes of meaning-making and expression. What is at stake, however, is not the distinction itself, but how musical pattern and expression interact and are flexibly formed within performance. Furthermore, gugak is characterized by modes of transmission and performance that extend beyond standardized systems of notation. This suggests that certain aspects of musical practice cannot be fully accounted for solely through fixed mathematical patterns or formalized linguistic models, thereby requiring a more flexible analytical approach.

Gugak, Technology, and AI: Ethnomusicological Perspectives on Digital Mediation and Computational Creativity in Korean Musical Tradition 

Abstract: This paper examines the intersection of ethnomusicology, technology, artificial intelligence, and gugak (Korean traditional music) by exploring how technological mediation transforms a largely oral, embodied, and improvisatory tradition when it is translated into computational form. It begins by tracing earlier research initiatives and institutional efforts that brought technology and gugak into dialogue prior to the rise of AI, with particular attention to the distinctive challenges gugak presents for computational modeling. Its flexible rhythm and melodic contour, microtonal inflections, timbral nuance, improvisatory ornamentation, and performer-specific interpretation all resist standard forms of digital sampling and quantification. Building on these earlier technological engagements, the paper then examines how research has increasingly shifted toward AI-driven approaches to analysis and creative production, highlighting the new issues and challenges that emerge in this process. It turns to emerging AI-based practices in both Korean traditional music and broader ethnomusicological contexts, including work on other musical traditions. By situating gugak alongside these developments, the paper highlights shared concerns in AI and ethnomusicology, including methodological challenges and culturally specific approaches to technological translation. 

Temporal patterns in the complexity of child-directed song lyrics reflect their functions 

Abstract: Content produced for young audiences is structured to present opportunities for learning and social interactions. This research examines multi-scale temporal changes in predictability in Child-directed songs. We developed a technique based on Kolmogorov complexity to quantify the rate of change of textual information content over time. This method was applied to a corpus of 922 English, Spanish, and French publicly available child and adult-directed texts. Child-directed song lyrics (CDSongs) showed overall lower complexity compared to Adult-directed songs (ADsongs), and lower complexity was associated with a higher number of YouTube views. CDSongs showed a relatively higher information rate at the beginning and end compared to ADSongs. CDSongs and ADSongs showed a non-uniform information rate, but these periodic oscillatory patterns were more predictable in CDSongs compared to ADSongs. These findings suggest that the optimal balance between predictability and expressivity in information content differs between child- and adult-directed content, but also changes over timescales to potentially support multiple children’s needs. 

When Speech Meets Rhythm: What Igbo Children’s Songs Reveal About Language Structure 

Abstract: In Igbo, adjacent vowels across word boundaries often coalesce in speech, sometimes appearing acoustically as a single vowel. Yet when these same sequences are sung in children’s songs, speakers consistently align them with more rhythmic space than a singleton vowel. This talk examines how musical text-setting reveals aspects of prosodic structure that are not always visible in speech alone. Using a well-known Igbo traditional children’s song, I analyze how compound names containing vowel sequences are mapped onto rhythmic positions in the melody. Listener judgments and patterns of musical alignment show that speakers prefer mappings that treat coalesced vowel sequences as occupying greater rhythmic weight than single vowels, even when vowel duration differences are minimal. The results suggest that musical rhythm reflects speakers’ underlying representations of timing structure and syllable weight. More broadly, the study demonstrates how musical practices can provide independent evidence for phonological structure in tonal languages. 

Time delay embeddings to characterize the timbre of musical instruments using Topological Data Analysis

Abstract: Timbre allows us to distinguish between sounds even when they share the same pitch and loudness, playing an important role in music, instrument recognition, and speech. Traditional approaches, such as frequency analysis or machine learning, often overlook subtle characteristics of sound. Topological Data Analysis (TDA) can capture complex patterns, but its application to timbre has been limited, partly because it is unclear how to represent sound effectively for TDA. In this study, we investigate how different time delay embeddings affect TDA results. Using both synthetic and real audio signals, we identify time delays that enhance the detection of harmonic structures. Our findings show that specific delays, related to fractions of the fundamental period, allow TDA to reveal key harmonic features and distinguish between integer and non-integer harmonics. The method is effective for synthetic and real musical instrument sounds and opens the way for future works, which could extend it to more complex sounds using higher-dimensional embeddings and additional persistence statistics. 

Core systems of music perception

Abstract: Like vocalizations found across the animal kingdom, music serves a basic communicative function in our species: it transmits information from the minds of people producing it to the minds of people hearing it. In this talk I will present evidence that this communicative function is a core property of human cognition (review: Mehr 2025, Trends in Cognitive Sciences). First, I will show that the communicative properties of music are widespread, as it produces reliable psychological responses in listeners across cultures and across the lifespan. These findings suggest that human musicality is not an incidental outcome of cultural evolution, but instead forms a constituent part of our psychology, in a fashion comparable to other domains, like social cognition, number, and so on. Then, I will present ideas on the perceptual mechanisms that underlie music's communicative functions, focusing on the automatic hierarchical representation of pitch and rhythm, including new data from infant perception studies as well as citizen-science approaches developed in my lab. High-level music processing abilities appear to be universal, early-developing, and uniquely human, and can be considered specialized components of human audition.

Symbolic Structure in Music AI: From Multimodal Translation to Modeling Gugak

Abstract: This talk presents recent studies on symbolic structure in music AI across a broad spectrum of musical representations. I begin with U-MuST, a unified multimodal translation framework that connects score images, symbolic music, MIDI, and performance audio through a shared token-based architecture, made possible by large-scale score video data collected from YouTube. I then introduce gugak-oriented research in two directions: regenerating Korean court music ensembles through direct Jeongganbo encoding, and learning discrete symbolic tokens from continuous pitch contours with autoencoders for new forms of musical analysis. The Jeongganbo work demonstrates how culturally specific notation can support the generative modeling of 15th-century court repertory. In contrast, the autoencoder-based learning of discrete tokens moves beyond pre-existing notation systems such as staff notation and explores how musically meaningful symbolic representations can be derived directly from data with minimal prior assumptions. Overall, the talk explores both how symbolic information supports music AI and how music AI can in turn produce new symbolic forms for representing musical knowledge.

Human-AI Collaborative Composition in Sanjo: Exploring New Creative Possibilities for Traditional Music

Abstract: Sanjo is a prominent genre of Korean traditional music, recognized for its technical complexity as a solo instrumental form rooted in oral tradition. Sanjo is regarded as one of the most challenging genres to both perform and compose. This study investigates the potential of Human-AI collaboration in the compositional process of Sanjo, moving beyond machine generation to a co-creative framework. This study implements a collaborative workflow where an algorithm suggests melodic variations based on the topological analysis of existing melodies, while the human composer refines the 'Sigimsae' and constructs the structural narrative using artistic imagination. The research focuses on the interplay between algorithmic suggestions and human artistic intuition. The presentation will further discuss the challenges and possibilities of this co-creative approach, providing insights into the sustainable evolution of traditional arts in the digital age.

Edge-Prominence in Korean Song across Time and Genres

Abstract: This study investigates whether the alignment between lyrics and melody in Korean music reflects systematic phonological mapping. Our previous analysis of children’s songs showed that strong metrical beats align predominantly with word boundaries (initial and final syllables), suggesting a demarcative function of musical prominence in a non-stress language. The present project extends this scope to a diachronic corpus of Korean songs spanning multiple historical periods. We focus on whether the historical loss of vowel length and subsequent changes in word-level prosody have shifted alignment principles over time, or if patterns vary primarily by genre. To quantify the strength and distribution of these alignment patterns, we incorporate information-theoretic measures. By exploring the interaction between evolving linguistic structures and musical meter, we seek to determine the stability of edge-prominence across different eras and styles of Korean song.

Topological Optimization for Korean Music Generation and Entropy-Based Analysis

Abstract: Deep learning–based approaches to music generation typically require large-scale datasets for effective training, as learning the underlying patterns and implicit grammar embedded in musical data. However, in many cases, particularly for certain genres of Korean traditional music, the available datasets are limited and insufficient for standard data-driven approaches. To address this issue, we proposed a framework based on structural learning via topological data analysis (TDA). In our work, we extract topological features that capture the intrinsic structure of musical sequences and feed these features into the learning process. This talk will present how such topological representations can be embedded into machine learning models for music generation. While stability threorem ensures robustness of topological features under perturbations, they do not directly guarantee that minimizing losses of these features leads to isometry between the original and generated music. But, under certain conditions, topological constraints can enhance structural similarity between them. Furthermore, we use entropy-based analysis to quantify how well the generated music preserves the structural characteristics of the original compositions. We will illustrate these ideas through examples from Korean traditional music.