Exploring the Structure of Germanic Folksong

Dissertation

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of Philosophy

in the Graduate School of The Ohio State University

By

Andrew Brinkman

Graduate Program in Music

The Ohio State University

2020

Dissertation Committee:

David Huron, Advisor

Daniel Shanahan, Co-Advisor

Ryan Skinner

1

Copyrighted by

Andrew Brinkman

2020

2

Abstract

The history of musical development in cultural groups across Europe is vast and varied.

We know, for instance, that groups of people use music as a way of both creating social cohesion and differentiation or exclusion (Hagen & Bryant, 2003). The relationship between cultural group identity and musical characteristics is one that has gathered some attention in European folk music scholarship over the years. However, the question of how musical features act as a reflection of unique cultural groups needs further exploration. Additionally, the Essen Folksong Collection, one of the only large databases of Western folk music, has seen extensive overuse over the past few decades, thus eliciting further inquiry into the representativeness of the Collection itself. In this dissertation, I attempt to provide an in-depth analysis of the Essen Folksong Collection as a representation of 19th-century German folksong by exploring its origins, how scholars have used the Collection, and what the Collection still has left to tell us about the nature of Western folksongs. In particular, I discuss the somewhat unclear background of the

Essen Collection and how Helmuth Schaffrath and his colleagues in Essen, Germany first came upon the materials that make up the Essen Collection. Afterwards, I provide a small meta-analysis of the research that has cited the Essen Collection, pointing out three general topic trends. Finally, I provide an exploration of the relationship between basic statistical properties of the Essen Collection and their geographical spread across the

iii

German region. By taking this long look at the history and usage of the Essen Collection, we can begin to better understand how empirical research in folksong studies has been shaped by the past and how best to approach it in the future.

iv

Acknowledgments

Without the help of my advisors and faculty support of the Ohio State University. Great thanks is given to all of these people for their dedication to making my dissertation a reality and to God who has stood by me every step of the way.

v

Vita

2014 ...... B.A. Music Education, Midwestern State University

2016 ...... M.M, Music Theory, Louisiana State University

2016 to present...... Graduate Teaching Assistant, Department of

Music, The Ohio State University

Publications

Brinkman, A., & Huron, D. (2018). The leading sixth scale degree: A test of Day-

O'Connell's theory. Journal of New Music Research, 47(2), 166-175.

Fields of Study

Major Field: Music

vi

Table of Contents

Abstract ...... iii Acknowledgments...... v Vita ...... vi List of Figures ...... viii Chapter 1. Introduction ...... 1 Outline of Chapters ...... 11 Conclusion ...... 14 Chapter 2. The Background and Structure of the Essen Folksong Collection ...... 17 The Origins of the Essen Collection ...... 21 The Structure of the Essen Collection ...... 34 The Issue of Capta Versus Data ...... 50 Statistical Properties of the Essen Collection ...... 55 Chapter 3. Scholarly Attention to the Essen Collection ...... 70 Proscriptive Studies with the Essen Collection ...... 78 Computational Modeling Studies ...... 84 Descriptive Studies ...... 96 Pitfalls and Mentions ...... 99 Chapter 4: Geographical Distributions of Statistical Properties ...... 104 Conclusion ...... 145 Bibliography ...... 149

vii

List of Figures

Figure 1. Total counts of folksongs in the Essen Collection ...... 38

Figure 2: Tallies of scale degrees across six subsections of the Essen Collection as percentages. Each colored portion represents the percentage of tallies of a given scale degree found across all folksongs in each subsection of the Essen Collection ...... 57

Figure 3: Tallies of scale degrees across six subsections of the Essen Collection ...... 58

Figure 4: Tallies of diatonic intervals across six subsections of the Essen Collection. .... 62

Figure 5: Tallies of diatonic intervals across six subsections of the Essen Collection ..... 63

Figure 6: Tallies of durations in the Essen Collection. Specifically, breves, whole notes, half notes, quarter notes, eighth notes, eighth-note triplets, and sixteenth notes are included ...... 66

Figure 7: Tallies of durations in the Essen Collection. Specifically, breves, whole notes, half notes, quarter notes, eighth notes, eighth-note triplets, and sixteenth notes are included ...... 67

Figure 8: Bins represent instances of perfect fifth intervals across folksongs in the Essen Collection where lower instances are more green and higher instances are more red. Transparent black points represent tallies of quarter notes taken from individual folksongs and appear at the point indicated by their latitudinal and longitudinal coordinates. Larger circles represent higher instances of quarter notes while darker shades represent multiple different coordinates in a similar location...... 109

Figure 9: A reproduction of Figure 8 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of quarter notes from the overall mean of quarter notes across all folksongs...... 111

Figure 10: Presence of scale degree 1, represented by a binned heatmap, and quarter-note tallies, represented by shaded circles...... 114

viii

Figure 11: A reproduction of Figure 10 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of quarter notes from the overall mean of quarter notes across all folksongs...... 116

Figure 12: Presence of scale degree 1, represented by a binned heatmap, and eighth-note tallies, represented by shaded circles...... 118

Figure 13: A reproduction of Figure 12 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of eighth notes from the overall mean of eighth notes across all folksongs...... 120

Figure 14: Presence of scale degree 1, represented by a binned heatmap, and sixteenth- note tallies, represented by shaded circles...... 122

Figure 15: A reproduction of Figure 14 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of sixteenth notes from the overall mean of sixteenth notes across all folksongs...... 124

Figure 16: Presence of scale degree 5, represented by a binned heatmap, and quarter note tallies, represented by shaded circles...... 127

Figure 17: A reproduction of Figure 16 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of quarter notes from the overall mean of quarter notes across all folksongs...... 129

Figure 18: Presence of scale degree 5, represented by a binned heatmap, and eighth-note tallies, represented by shaded circles...... 131

Figure 19: Presence of scale degree 6, represented by a binned heatmap, and eighth note tallies, represented by shaded circles...... 132

Figure 20: Instances of perfect fourth intervals, represented by a binned heatmap, and tallies of scale degree 5, represented by shaded circles...... 134

Figure 21: A reproduction of Figure 20 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of scale degree 5 from the overall mean instances of scale degree 5 across all folksongs...... 136

ix

Figure 22: Instances of perfect fifth intervals, represented by a binned heatmap, and tallies of scale degree 5, represented by shaded circles...... 138

Figure 23: Instances of scale degree 1, represented by a binned heatmap, and tallies of major second intervals, represented by shaded circles...... 140

Figure 24: A reproduction of Figure 23 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of scale degree 1 from the overall mean instances of scale degree 1 across all folksongs...... 142

x

Chapter 1. Introduction

For quite some time, music scholars have observed that music tends to reflect unique cultural groups (Hagen & Bryant, 2003). In all the many diverse ways that differences between cultures might manifest themselves, music plays at a least a partially significant role. Music can strongly highlight, or draw attention to, the uniqueness of the cultural group from which it arises regardless of whether such a practice is interpreted as good or bad. For instance, popular American music often reflects the party-seeking, fun-loving culture of American youth through its upbeat tempo and rhythms, positive lyrics, and catchy melodies (Shuker, 2012). Similarly, it has been said that music from the Czech

Republic tends to emphasize the tenacity of the Czech people to overcome adversity and develop independence (Beckerman, 1986). The above descriptions, however, provide only one perspective on how music both interacts with and is affected by its cultural point of origin.

Apart from the above claim that music tends to reflect unique cultural groups, it is also commonly thought that musical features can be used to characterize a culture

(Bohlman, 2013; Ling, 1997). In the Czech example used above, while scholars certainly stress that musical features alone do not define the entirety of Czech music, they also stress that Czech music does include some relatively specific musical characteristics (e.g. use of raised scale degree 4 and syncopated rhythms). Studies of European folk music

1 more broadly have often looked at how regional folk musics vary based on differences in unique musical features (Juhász & Sipos, 2010; Savage et al., 2015). Again, these scholars often draw attention to uniqueness in musical features as characteristics rather than as definitions of a cultural group’s music. Yet, notice that the two claims (that music reflects a cultural group and that musical features characterize a cultural group) are distinctly different claims.

The initial claim presents the idea that music is a reflection of culture. This claim also suggests that music is able to paint a “sonic picture” of the cultural group in question, meaning that a listener might be able to learn quite a lot about a cultural group simply from listening to their music. Conversely, the second claim advances the idea that culture itself is characterized by its music. Said another way, the second claim strongly implies that features present in a cultural group’s music might directly affect how we understand that group in a broader context. Of these two claims, the second certainly raises some thorny ethical issues and invites further examination.

As a possible proponent of the second claim, Hagen and Bryant (Hagen & Bryant,

2003), have summarized research showing how music simultaneously defines who/what belongs to a culture and who/what does not belong to a culture. In field observations conducted over centuries, music has been used as a tool to both unite and segregate different cultural groups. Hagen and Bryant’s research found that tribal groups often distinguish themselves from one another by the musics that each employs. For example, one group might have a very specific and complicated dance that follows alongside heavily syncopated drum rhythms whereas another group uses those same syncopated

2 drum rhythms but in tandem with multiple different drums, resulting in a heterophonic drum “choir”. In this example, while each group might make use of relatively similar musical features (namely, syncopated rhythms), each also has their own unique version

(one with elaborative dance, the other with heterophonic textures). Here, the uniqueness found between the musics of these two groups is important for differentiating between tribal memberships. Hagen and Bryant make the claim that, if a member of the first

“dancing” tribe were to try and infiltrate the second “heterophonic” tribe, that that individual would have an exceptionally difficult time blending in as the musical practices might require him/her to understand, enjoy, and possibly even take part in, a musical practice that is wholly foreign. In this case, members of the either tribe would then be able to differentiate themselves from the intruder, thus preventing infiltration.

Of course, while the above example is merely a hypothetical scenario, the danger implied by Hagen and Bryant’s assertions, and more broadly in the second claim of musical features as defining characteristics of a culture, is the use of musical characteristics as a source of social exclusion. Hagen and Bryant do strongly suggest that musical features can be used to exclude individuals from a specific cultural group.

However, in describing culture-specific musical features, it is preferable to regard these features more as reflections of a cultural group rather than definitions of a cultural group.

Tackling the ethical issues of social exclusion in cultural groups through musical characteristics is a problem that is much too large and unwieldy to address in this dissertation.

3

One issue in describing a cultural group through its musical features is defining the group itself. Related to that idea is also what would be an appropriate group size.

There can be different sizes of cultural groups and people can belong to more than one cultural group simultaneously. For example, a cultural group might be as small as a few friends who share a love of two or three popular bands. Alternatively, a cultural group might be as big as “Western culture”, which embraces hundreds of millions of people spanning multiple nations. Both of these example cultural groups highlight the fact that a person can belong to multiple different cultural groups simultaneously. Of course, we can refine the idea of simultaneous cultural group inclusion a bit further.

Historically, a particularly salient cultural group has been evident in the concept of a nation. Today, we are more likely to acknowledge that nations rarely form homogenous cultures. This is especially pertinent in the ever-rising trend of Western globalization in recent decades. However, in the nineteenth century there was a much greater tendency to conceive of a nation in terms of homogenous cultural features or traits

(Bell, 2009). One need not look any further than the political landscape of Europe during the nineteenth century to see an example of this, where nationalist sentiments often superseded small-group identity. In fact, the notion of nationhood and its relationship to music is reflected in the way in which folk music resources were assembled in the nineteenth century.

Often thought of as the “music of the people”, folk music has served as a contrast to more “classical” or “art” music (Wiora, 1949). With the rise of nationalist sentiments during the nineteenth century, folk music saw an immense surge in popularity.

4

Composers frequently incorporated whole folksong melodies, or composed in a manner in which folk music idioms were emulated. Political groups were using folk musics as a tool, much in the way described by Hagen and Bryant, to differentiate between peoples of relatively similar geographical location. In order to distinguish themselves from neighboring rivals, national leaders were explicitly interested in collecting and preserving folksongs believed to be belonging to their own cultural/national group. Therefore, it became apparent that the conception of what a cultural group is defined as was tacitly reflected in the materials included in a particular folksong collection.

The history of the emergence of the modern German state in the nineteenth century highlights many of the above issues of national/cultural definition and folksong collection. For example, from the early years of the nineteenth century until 1866, the geographical region corresponding to modern-day Germany was primarily made up of small, independent states in the south with the large nation of Prussia occupying regions in the north. At the midpoint of the century, Prussia united many of the northern

Germanic states into what was known as the North German Confederacy while the southern states continued to exist as independent entities. During this time of independence, these states (including Prussia) were explicitly interested in developing nationalist sentiments through collecting national folksongs. Individual German states like Bavaria and Württemberg were collecting Germanic folksongs, not Bavarian or

Württembergian folksongs, suggesting that collectors were focused on a larger cultural identity. Similarly, scholars from the North German Confederacy were also collecting folksongs that emphasized a pan-Germanic outlook, much like scholars from the

5 independent southern states. The pan-Germanic influences on collecting folk material coincided with efforts across the Germanic region to create the modern state of Germany.

In light of the above example, it would appear that Europe (and more specifically,

European nationalism) would be a good place to begin an investigation of the idea of cultural groups and their reflections in musical characteristics. Phillip Bohlman

(Bohlman, 1988, 2013) has reviewed much of the existing research done on Central

European folk music within the past few decades. More specifically, Bohlman has brought together much research that has been done on the folk music of Germany and condensed it into a more concise form. Additionally, scholars like Walter Wiora (Wiora,

1964), have done exceptional work on defining Germanic folk music and on European folksong studies as a whole. The work of these many scholars has contributed to our current understand of Germanic folksong. For instance, Bohlman and others have spent a considerable amount of time and resources investigating Germanic folksong genres, lyrical forms, and performance practices (Laufhutte, 1991; Meier et al., 1935; Rohrich,

1973). Additionally, Wiora and others have investigated topics of Germanic folksong’s melodic and formal structures (Steinbeck, 1976; Wiora, 1977). In empirical studies, even more work has been conducted on 19th-century Germanic folksong (especially concerning the Essen Folksong Collection) for various other purposes and in order to achieve many different research goals. The bulk of this literature will be explored in

Chapter 3. Thanks to the work of these scholars, we know quite a bit about various aspects of Germanic folksong from the nineteenth century and, by extension, we have

6 also learned more broadly about Germanic culture in general through the exploration of these folksongs.

Much time could be spent on an exhaustive review of what has already been done by the above scholars. I could investigate the relationship between Germanic folksong structure and the structure of Western popular music (Bausinger, 1973), how the practice of yodeling has affected genre distinctions across Southern Germany (Schneider, 1982), the differences between religious and labor-related folksongs (Schroubek, 1973), and even how Germanic folksong melodies relate to the structure of Germanic prosody

(Laade, 1988). However, relatively little of the research presented by these many scholars has focused on the sonic-musical content of Germanic folk music, and still further, even less scholarship has been dedicated to thorough analysis of primary source materials for

Germanic folk music. For the purposes of this dissertation, I define sonic-musical content as including such features as the pitch and rhythm content, phrase and formal structures, and manner of performance, including vocal and instrumental timbres. Much of this information is lost to us, such as the specific performance practices including customary or traditional vocal timbre. Consequently, the principle focus of this dissertation will be on what sonic-musical content is available for analysis and how that content might be associated with cultural groups related to modern-day Germany.

Part of the reason why there seems to be less of a focus on the sonic-musical content is because analyzing music is highly labor intensive. Normally, music scholars apply analysis to individual works in the hopes that it will illuminate the work in some way. In more ambitious situations, scholars might analyze a large number of works by a

7 single composer as a way of illuminating that composer’s style. Leonard Meyer’s work on style in music describes this practice of analysis in great detail (Meyer, 1989).

However, when we are interested in cultural groups as opposed to individual musicians, we necessarily have to analyze even more music in order to be able to draw valid generalizations. Consequently, if we want to focus on the sonic-musical content for a cultural group, then we need to do lots of analysis. We must examine and test many different works and features in order to say anything substantial about the sonic-musical content of a specific cultural group. While many brave scholars have attempted this kind of research by hand or with relatively little computational assistance (Lomax, 1976) the kind of labor-intensiveness necessary for a study of sonic-musical content within an entire cultural group suggests the use of computational methods. Through computational methods, it becomes relatively simple to analyze a large body of works within a reasonable amount of time. Therefore, if one were interested in looking at the music of a cultural group as large as a nation (the 19th-century Germanic cultural group for instance), such use of computational analysis would be exceptionally helpful.

Of course, conducting computational analyses presumes the existence of appropriate databases. Analysis can only proceed when there is music to analyze.

Similarly, without a sufficient amount of music to analyze, it is difficult to make any claims regarding the use of sonic-musical features in nation-sized cultural groups. In this dissertation, I discuss and make use of an existing database, the Essen Folksong

Collection (Schaffrath, 1995), in order to address many of the issues presented in the paragraphs above. As one of the largest databases of Western folk music, the Essen

8

Collection has been used extensively by scholars interested in a variety of topics within the past few decades. However, relatively few of those scholars have sufficiently addressed the question of what exactly the folksongs in the Essen Collection represents.

Over the course of the following chapters, I present an in-depth analysis on the history and usage of the Essen Collection. In so doing, I also address the many topics presented in the paragraphs above regarding the idea of a nation-sized cultural group, the multitudinous cultural groups that exist within the nation, and how musical features can reflect cultural groups especially when it comes to collecting and archiving folksong materials.

In order for scholars to conduct studies on topics involving Germanic folksongs, it is necessary that they have access to a database of “Germanic” folksongs. The Essen

Collection is one example of such a database that has been encoded for computational analysis. However, as some scholars have noted (Cenkerová et al., 2018; Lattner et al.,

2015), the Essen Collection has been used quite frequently in Music Information

Retrieval, computational modeling, and music cognition research. In each of these circles, it seems to be the case that the Essen Collection is the preferred database when questions related to folksongs or monophonic melodies are addressed. In nearly all cases involving citations of the Essen Collection, very few scholars discuss what the Essen

Collection actually represents, and even fewer discuss where the Collection comes from.

It is important to note that the Essen Collection represents a reflection of Helmut

Schaffrath’s conception of a cultural group. This means that it is dangerous for scholars to assume that the Essen Collection is an accurate representation of “19th-century”

9

“Germanic” “folksong”. This dissertation will attempt to clear up some of these issues regarding the origins, structure, and representativeness of the Essen Collection.

By taking into consideration the existing scholarship on Germanic folk music and scholarship covering the Essen Collection itself, it may be possible to reach a finer definition of what exactly the materials in the Collection represent. One scholar who has worked tirelessly in determining the relationships between geographical location, cultural differences, and Germanic folksong is Bret Aarden. In their work with the Essen

Collection, Aarden and Huron (Aarden & Huron, 2001), found that geography is correlated with musical features. In their work, they used geographical coordinates, added by Bret Aarden, within the folksongs of the Essen Collection in order to determine the sonic-musical feature landscape of Central Europe. More specifically, they found a strong distinction between folksongs in more west-of-central Europe and those in east-of-central

Europe (The distinction used here with Central Europe as focus is because, while the

Essen Collection does include folksongs outside of the modern-day German region, these folksongs do not come from areas much further away from Germany than Western

France to the west and central Balkan nations to the east). From a historical perspective, the collection of folksongs for the Essen Collection during the nineteenth century coincided with the strong north/south political divisions across the Germanic region during that time. Prussia (and later the North German Confederacy) split Germany into northern marshlands and the southern alpine region (of mostly independent nation-states).

Additionally, we should remember that, before the nineteenth century, the region of modern-day Germany was even more geo-politically divided as the Holy Roman Empire

10

(which dominated much of the Germanic region for centuries) heavily emphasized the relative independence of the many German nation-states.

Given the sonic-musical feature findings of Aarden and Huron it seems highly possible that the geographical information within the Essen Collection can tell us more about the feature content of the German region. Schaffrath included information for each folksong concerning where in Germany the original archivists and collectors came upon said folksong, and Aarden has provided more detailed coordinates for each folksong based on Schaffrath’s original encodings. By making use of this information, it is possible to create maps of the German region that show areas where specific sonic- musical features tend to concentrate and to speculate as to why this or that feature might be found in any given area.

Outline of Chapters

This dissertation is organized into four chapters. Chapters 2 and 3 provide an examination of the history and usage of the Essen Collection. These chapters serve as the primary focus of this paper, providing a broad review of the existing empirical literature on

Germanic folksong as it relates to the Essen Collection, addressing many of the questions, issues, and claims related to what it means for the Essen Collection to be representative of 19th-century Germanic folksong presented in this initial chapter, and starting a discussion on how best to approach this and other databases like it in future research tasks.

11

In Chapter 2, I explore what we currently know about the origins and structure of the Essen Collection. I delve into the history of the Essen Collection in the Essen

Associative Code (EsAC) and discuss some of the issues related to Schaffrath’s initial encodings and the primary source materials collected by the original 19th-century archivists. There seems to be some mild disagreements between the folksongs encoded in the Essen Collection and the primary sources that supposedly serve as the material from which Schaffrath pulled the Essen materials. Discussions with Ewa Dahlig-Turek, one of the original contributors to the Essen Collection and a close colleague to Schaffrath provide insight into the original decisions that were made concerning how to organize and present the Essen Collection materials. Additionally, I delve deeply into the archives of the Center for Computer-Assisted Research in the Humanities (CCARH) for different perspectives on the origins of the Essen Collection including alternative histories of the

Collection’s conception and potential sources for Schaffrath’s inspiration to put together the Collection. In general, this chapter is meant to be an introduction into the Essen

Collection and provides thoughts on how we can better understand the representativeness of the Collection. Along the way, we will find that the Essen Collection’s representation of 19th-century Germanic folksongs are rather troublesome and that there is a thorny question of whether the materials in the Collection are truly “19th-century”, “Germanic”, and/or “folksongs”. Finally, the last portion of this chapter provides some brief insights into the twelve subsections of the Essen Collection, a topic regarding the structure of the

Collection that has almost never been addressed in the scholarly literature. In order to understand exactly what the materials in the Essen Collection can tell us, it is important

12 that we consider the structure and organization of these materials within the context of their origins and the history of the Collection itself.

Following the origins and structure of the Essen Collection, a follow-up meta- analysis is reported in Chapter 3 that involves a thorough investigation of over 170 citations and mentions of the Essen Collection across the scholarly literature. In general, the literature addressing the Essen Collection can be pooled into three different categories: those involving proscriptive topics (typically music cognition), topics related to computational modeling, and those involving descriptive topics (typically musicological topics). This chapter details the literature in each of these categories, addressing both the insights and the potential pitfalls provided by these studies and situating them within the context of our understanding of the Essen Collection. More than anything else, this chapter is meant to provide an understanding of how the Essen

Collection has been used since its inception just a few decades ago and to show how scholars might best handle the Collection in future projects in order to avoid dangerous assumptions. Overall, this chapter shows that, while most scholars do not make any egregious misuses of the Collection, the bulk of the research suffers from a lack of understanding regarding exactly what constitutes the Essen Collection and its representativeness. As such, in order to obtain the best understanding of exactly what the

Essen Collection is and its best usage, Chapters 2 and 3 should be understood as different perspectives on the same topic: the representativeness and structure of the Essen

Collection as it relates to 19th-century Germanic folksong.

13

The final chapter in this dissertation provides a positive look on what the Essen

Collection can still offer researchers even after decades of potential overuse. In particular,

Chapter 4 provides an exploratory study into the relationship between basic statistical properties of the Essen Collection and geographical location. At its core, this chapter shows how simple musical features like scale degrees, intervals, and rhythmic values represented by note durations can be found in different concentration levels across the

German region. This chapter does not provide any statistical testing regarding the strength of these relationships, but does show interesting facets of the sonic-musical feature content of the Essen Collection and their geographical locations. I hope that this initial exploration will result in further research into questions of geographical location and sonic-musical feature content as a way of understanding the relationship between cultural groups. For now, the information provided in this chapter is meant merely to spark continued interest in the Essen Collection.

Conclusion

At this point, it is appropriate to remind the reader of the main goals of this dissertation.

First and foremost, this dissertation serves as an initial investigation into the sonic- musical features that characterize Germanic folksong. While much of this dissertation is dedicated to an examination of the Essen Folksong Collection, this is done primarily due to the importance of this database on our current understanding of Germanic folksong.

The studies discussed across these four chapters do not always have an explicit tie to the sonic-musical feature content of Germanic folksong. However, in order to explore the 14 structure of Germanic folksong more fully, it is important that we understand the tools that we work with. That is, to discuss Germanic folksong without addressing the Essen

Folksong Collection would be to dismiss a rather substantial amount of scholarly literature in folk music research. Of course, it should be understood that the perspective taken in this dissertation, one that has, at its focus, an analysis of the Essen Collection, is not the only perspective that can be taken regarding understanding the structure of

Germanic folksong. Instead of “casting my net too wide and too far”, I have elected to focus on one fairly extensive perspective on the topic. The conclusions that are drawn regarding the Essen Collection in this dissertation are not meant to dissuade scholars from making use of its materials, nor are they meant to suggest that empirical study of

Germanic folksong is the best way forward. I merely wish to draw attention to this one perspective on Germanic folksong research and to address as many aspects related to it as

I can.

It should be understood that this dissertation also not be construed as providing any definitive statement regarding the structural features of Germanic folksong.

Similarly, this dissertation also should not be regarded as offering a definition of

Germanic folksong. As discussed above, to attempt to define Germanic culture through an investigation of all possible musical features is both ethically troublesome and, quite frankly, impossible to conduct. However, in order to reach any conclusions regarding this topic, it is important that substantial evidence be provided. As a means of distancing myself from any dangerous claims concerning Germanic folksongs, this dissertation only provides some focused discussions and investigations of potentially important musical

15 features in Germanic folksong as they relate to the Essen Folksong Collection. The specific sonic-musical features of interest in this dissertation (as is especially the case in

Chapter 4) may very well turn out to be relatively unimportant to Germanic folksongs.

However, by examining the relationship between the Essen Collection and folksong research, we might begin to better understand the structure of Germanic folksong.

16

Chapter 2. The Background and Structure of the Essen Folksong Collection

Schaffrath and colleagues at the Hochschule für Musik in Essen, Germany started putting together the Essen Collection in 1982 while working on a new method for quickly and accurately encoding printed music into a computer-friendly format. This early encoding method was named the Essen Associative Code or “EsAC”. Publications of Schaffrath’s work with the Essen Collection began to appear quick rapidly in the empirical music journal Computing in Musicology between 1992 and 1994. These first public releases of the Essen Collection provided nearly 6,500 folksongs collected from Volkslied publications from the nineteenth century. It was divided into 25 different subsections, each supposedly named after the archivist who first collected the printed manuscripts.

Shortly after Schaffrath’s sudden death in 1994, the EsAC project, now a conglomeration of multiple databases including the Essen Collection itself and others passed to the leadership of Polish ethnomusicologist Ewa Dahlig-Turek who has since presided over the project until today.

Under Professor Dahlig-Turek’s leadership, and through the work of music scholar David Huron, the folksongs within the Essen Collection were transcribed from their original EsAC format to the new **kern format. Now existing in two different formats, the Essen Collection quickly became one of the most highly used large database of symbolically noted music across the field of computational and systematic

17 musicology, boasting an astounding 216 citations within the past twenty-five years. As it currently stands, the Essen Collection is one of the largest collections of symbolically notated folk music, only recently being rivaled by the Meertens Tune Collection (Van

Kranenburg et al., 2014). Perhaps one of the primary reasons for the overall success of the Essen Collection comes from the fact that it exists in two highly user-friendly formats for computational analysis. Another reason for its success might be due to its representativeness, being one of the only substantial databases of 19th-century Germanic.

Still another reason might simply be due to how recognizable the Essen Collection is.

After all, the Essen Collection has appeared as one of the key databases used in some of the most influential and widely known scholarly works in the fields of music cognition and empirical musicology (Huron, 1996; Patel, 2010; Marcus T. Pearce & Wiggins,

2006; Temperley, 2004). It could easily be the case that scholars have tended to use the

Essen Collection simply because other influential scholars in the field have also used the database.

In any case, the importance of the Essen Collection and its impact on the scholarly work of the past twenty-five years cannot be dismissed and warrants further examination. The Essen Collection has served as the primary training and test sample for a bulk of the work in melodic segmentation, modeling of meter and rhythm, melodic contour analysis, cross-cultural enculturation and structure studies, expectation and similarity studies, and rule-based/probabilistic framework studies to name just a few of the more popular research topics. For these scholars, the Essen Collection has helped define much of their academic career, serving as the primary point of focus across much

18 of the development and growth of their work during the 2000’s.1 For others, the Essen

Collection has helped them explore various relationships between different musical styles, genres, time periods, locations, and cultures as exemplified in papers on the relationship between European and non-European folksong, Germanic and non-Germanic folksong, traditional and Western Art music, music of the fourteenth century to the nineteenth century, and even modern popular music and popular “music of the people”.

Given the lack of available resources, the usefulness of the Collection itself, and the growing interest in these research topics, it is unsurprising that the Essen Collection has become so widely used, a fact that would probably make Schaffrath very proud (Dahlig-

Turek, personal communication, January 19, 2020). However, underneath the immensely successful surface of the Essen Collection lies some rather troublesome questions that even twenty-five years later have yet to be answered.

In light of recent developments in the way that we understand what it means to construct and analyze a “corpus”, perhaps it is time that we look once again at the Essen

Collection and its structure. This time however, we do not ask what the Essen Collection can tell us about music at-large, but instead we ask what the Essen Collection can tell us about itself. Scholars have taken much information from the Essen Collection, and very little of that information has been freely given to them. That is, if we are to understand the Essen Collection as a representative of some sort of larger musical population, we must ask ourselves whether we truly understand what that musical population is. Have we

1 These authors include David Temperley (2004; 2007), Marcus Pearce & Geraint Wiggins (2004; 2006), David Huron (1996; 2001), Tuomas Eerola & Petri Toiviainen (2004a; 2004b), Bret Aarden (2001; 2003), Rens Bod (2002a; 2002b) as some of the primary examples. For additional examples of scholars who have made extensive use of the Essen Collection throughout their academic career, please see Chapter 3. 19 assumed that the Essen Collection is a collection of German folksongs incorrectly? Or, have we incorrectly assumed that the Essen Collection is a collection of German folksongs? To take this rather simple example one further step, is it correct to assume that the Essen Collection is a collection of German folksongs? Without actively placing our own suppositions onto the definition of the Essen Collection, can we accurately and succinctly define exactly what the Essen Collection is?

In this chapter, I attempt to provide as clear a definition of the Essen Folksong

Collection, as Helmut Schaffrath envisioned it, as possible. In doing so, perhaps it will be possible for future scholars to make clear and unbiased decisions on whether or not to include the Essen Collection as their testing material without incorrectly assuming that the Essen Collection is what many have claimed it to be. Thus, we can make sure that future research is conducted using clear and transparent methodology, free from concerns regarding the appropriateness of their data. In addition to addressing concerns regarding the origins and background of the Essen Collection, I also provide a brief analysis of the structure of the Collection, drawing attention to important facets of the Collection that have gone unnoticed in much of the scholarly research and providing new perspectives on the way we should handle the Essen Collection in the future.

This chapter is divided into two sections. The first is a broad overview on the history of the Essen Collection and on the Essen Associative Code. In this section, I explore questions regarding the nature of folksong collection and archival, the processes involved in transcribing written notation into symbolic notation, issues of representation and information loss, and cultural appropriation and inclusion. While attempting to

20 provide a clear definition of the Essen Collection, I also draw attention to some of the more troublesome issues related to the above topics and the Essen Collection. In the second section, I discuss the properties of the Essen Collection that have gone largely unmentioned in scholarly research with the database. I point out some of the broader concerns that arise when scholars have not acknowledged these properties and how future research might address these concerns in the future. Throughout the course of this chapter, I draw primarily upon what little primary source material exists on the Essen

Collection and on an interview conducted with one of the primary contributors of the

EsAC project, Ewa Dahlig-Turek to provide the aforementioned background information.

This background information is supplemented with a meta-analysis of the scholarly work that has made use of the Essen Collection. Ideally, in providing the following background and analysis below, we might begin to better understand the role that the Essen

Collection plays in the future of corpus study and music cognition research.

The Origins of the Essen Collection

According to scholars at the Center for Computer Assisted Research in the Humanities, the origins of the Essen Collection can be traced back through the initial ideas behind the

Essen Associative Code (Packard Humanities Institute’s Center for Computer Assisted

Research in the Humanities at Stanford University, 2015). Initial clues as to the origins of the Essen Collection’s primary source material originate in the work of Wolfgang

Suppan, an ethnomusicologist specializing in the collection and research of German folksongs (Suppan, 1966). Suppan notes that collection efforts for Germanic folksong 21 materials could be traced to 18th-century Austrian emperor Joseph II, a member of the

Hapsburg dynasty. As an effort to maintain and preserve aspects of the “Hapsburg culture”, Joseph II helped to fund collection efforts for Germanic musics all across

Central Europe. Around the same time, Suppan notes that the German philosopher

Johann Gottfried von Herder first introduces the concept of the Volkslied, a specifically

German song produced and enjoyed by the lower classes of society, often in rural environments. Herder’s suggestion that music of the masses could be considered an additional cultural product boasting the strength of the Germanic peoples was an idea quickly supported by the quickly growing parties who wished to see a unified German nation. As notions of nationalism and unification grew moving into the nineteenth century, so too did efforts for the collection and preservation of folk music.

In his book on British “fakesongs”, Dave Harker (1985) notes the interesting and rather short history of folksongs in the British Isles. Harker points out that, during the 18th century, various members of high-class society became interested in the music of the

“lower-classes”. Specifically, Harker notes that clergymen, interested in the songs sung by parishioners while going about their daily lives, felt that these songs somehow represented the musical lives of the working-class individuals. Before this point in

Britain’s history, Harker remarks that there is little evidence that “music of the lower- classes” was ever recorded or archived. In fact, it is primarily due to the nationalist sentiment that begins to grow and develop over the next two centuries that the bourgeois become so interested in these “folksongs”. Harker’s writings echo much of the same sentiments expressed in German attention to volkslied in the nineteenth century,

22 suggesting that folksongs, as we understand them today, are primarily a product of class divisions in the 18th and 19th centuries. Harker’s work certainly calls into question the origins of the “folksong” movement. However, it is important to note that, regardless of the fact that large-scale collection of folksongs did not begin in most parts of Europe until the nineteenth century, “folksongs” or songs created and performed by the working-class have existed since antiquity, a point that is made clear by Harker’s work.

It should be understood that the horrors stemming from the result of German nationalism in the nineteenth and twentieth centuries are without excuse. It is also important to recognize that, for better or worse, proponents of a unified German state and culture were on the forefront of folksong collection and archival during this time. If the songs in the Essen Collection were collected around the nineteenth century, then we must keep in mind the historical context that allowed this archival work to occur. Considering the tenuous nature of the German nationalist movement, it is important that we recognize how our understanding of German culture, both in its historical context, and in modern terms, affects the representativeness of the Essen Collection and vice versa.

Suppan notes that, from the first moments of folksong collection and classification, there has been a need to make the process much faster and more refined.

As the sheer overwhelming number of folksong material increased into and through the nineteenth century, it became even more pertinent that scholars be able to rapidly and efficiently organize collected materials into comprehensible assortments. This increase in need for archiving folksong materials eventually resulted in the printing of large and numerous books of transcribed folksongs. While the writing and publishing of these

23 books certainly helped in the preservation process, the intense labor involved in by-hand transcription and classification proved highly taxing for early folksong scholars.

Unfortunately, while classification and collection methods could be streamlined by developing techniques in information sciences, it would not be until the twentieth century that transcription methods would become less labor intensive. With the invention of typescript notation, the first prototype versions of the EsAC came into being through various publications of folksong materials in the mid-twentieth century (CCARH, 2015).

Suppan himself discusses some of the ways that folksong transcription might be refined as early developments in computers begin to find success in ethnomusicological endeavors (a task carried out in work like: Bronson, 1969; Lomax, 1976). From Suppan’s work in 1966 until 1982, various institutions in Europe attempted to develop their own methods for electronic transcription of folksong materials, each to varying degrees of success (CCARH, 2015). However, a successful and efficient method for electron transcription would not find mass approval until the work of Helmut Schaffrath and his colleagues at the Hochschule für Musik. Starting in 1982 and lasting until 1994, Helmut

Schaffrath directed efforts in Essen to the adaptation and refinement of the Essen

Associative Code to the encoding and collection of Germanic folksong materials. During this time, many important decisions were made regarding the structure of the EsAC format and the initial presentation of the Essen Folksong Collection. Unfortunately, much of our understanding of Schaffrath’s selection process and initial decisions regarding

EsAC and the Essen Folksong Collection have been lost since the scholar’s death.

24

What the form the Essen Folksong Collection was meant to originally take is a question that holds some controversy. As it stands today, the Germanic portion of the

Essen Collection contains over 6,000 monophonic melody lines. Broadly assumed, the

Essen Collection is a collection of German folksongs, but there are many issues with this assumption. According to the CCARH, there seems to be nine books of folksongs from which the original twenty-five subsections of the Essen Collection are derived (CCARH,

2015). Yet, it is clear when searching through the original source material presented through the CCARH web platform that they are not entirely certain that these nine books are the printed scores that form the encodings found in the Essen Collection. In some cases, it is entirely likely that the primary source suggested by CCARH is the same used by Schaffrath in 1982. Nine of the original twenty-five subsections are named after prominent nineteenth century German folksong archivists who put together several books of Germanic music. Similarly, ten of the twenty-five subsections have short names that seem to match well with the titles of two nineteenth century books on Germanic music.

Conversely, several of the subsections either do not have a primary source attributed to them. Some suggested primary sources also seem highly unlikely to be the actual sources used by Schaffrath (as what appears to be the case with the “ballad” and “allerkbd” subsection). Because several of the subsections in the Essen Collection include geographical information related to where the folksong was recorded, it seems likely that the subsections labelled by the name of the primary source archivist are encodings of the very same primary sources suggested by CCARH. While this possibility is certainly

25 comforting, it nonetheless does little to ease uncertainty regarding the origins of the data encoded in the Essen Collection.

As an aside, while the issue of scholarly interaction with the Essen Collection will be addressed later in this chapter, it is important to note that many have perpetuated the idea that the folksongs found in the Essen Collection are derived from printed nineteenth century Volkslied publications. It is important to clear up some of the misconceptions that surround this statement. First, if we are to assume that the subsections named after actual archivists from the nineteenth century are indeed derived from the work of those same scholars, then we must address the fact that several of these books do not contain only folksong materials of the nineteenth century. For instance, the subsections pulled from the work of archivist Franz Magnus Böhme include folksongs from the eighteenth century as well.2 Few, if any, of the Essen Collection’s primary source material suggest that the folksongs included in the collection are somehow directly related to the nineteenth century. Given that archivists during this period of intense German nationalism were intentionally collecting all the folksong material that they could get their hands on, it makes sense that they would not necessarily be interested in only 19th- century materials. Of course, it is entirely possible that the majority of the folksongs in the Essen Collection are indeed products of the nineteenth century. However, with little information regarding the exact content of the original source material, it is rather

2 Examples of this can be seen in several places. Firstly, the full name of the Böhme’s collection is Volksthümliche Lieder der Deutschen im 18. und 19. Jahrhundert which strongly suggests that songs in this collection come from different time periods. A brief search of the source provides many dates attributed to songs before the year 1800. Some examples include: Das Deutsche Mädchen (1790), Reiterlied aus “Wallenstiens Lager” (1797), Erneuter Schwur (1724), Der Burgunder (1745). 26 dangerous to assume that the Essen Collection is a representative sample of 19th-century folksong.

In addition to these issues regarding identification of primary source material, there is also the fact that many songs mentioned in the primary sources do not appear in the Essen Collection itself, and vice versa. Therefore, it is difficult to say with certainty that a primary source suggested by the CCARH was the same source encoded by

Schaffrath in the original EsAC form of the Essen Collection. The largest subsection of the Essen Collection, supposedly derived from Deutscher Liederhort: Auswahl der vorzüglicheren deutschen Volkslieder, nach Wort und Weise aus der Vorzeit und

Gegenwart (Erk, 1893, 1894), provides an opportunity to examine this problem in more detail. Titles of various songs in Erk’s collection cannot be found in the “erk” subsection of the Essen Collection itself. Similarly, searching for various songs in the “erk” subsection returns no results in Erk’s original sources. It is possible that names of songs have been changed during the encoding process, and it is also possible that certain songs from Erk’s collection were simply left out of the Essen Collection. With so many different possible explanations for the issues apparent between the primary sources and the Essen Collection encodings, the problem of identifying the true primary source materials of the Essen Collection is a difficult one to address.

Alongside questions of primary source attribution is also the issue of how exactly to classify the songs in the Essen Collection, in terms of both genre and culture. In the

Erk collection mentioned above several different genres of folksong can be found including ballads, hymns, children’s songs, and many others. Each of the nine books that

27 supposedly make up the Essen Collection seems to cover widely different musical genres and styles. In the Erk example discussed, the same situation regarding genre occurs in primary source material for each of the twenty-five subsections of the Essen Collection

(including those found under the various “balladXX” labels). As part of his detailed work in encoding metadata for the Essen Collection, Schaffrath worked to include genre classifications for each song found in the Collection. However, given that genre identification is rather unclear for many songs in the source materials, one cannot help but wonder how exactly Schaffrath came to classify these songs. Given that the exact origin of the primary source material is suspect, we are left to assume that whatever materials Schaffrath originally encoded did include the metadata information that

Schaffrath transcribed. A similar situation arises in the case of geographic location. For each of the songs in the Essen Collection, Schaffrath included metadata information regarding the location of origin dictated in the primary source material with varying levels of specificity. Many of the folksongs included in the sources provided by the

CCARH do not appear to include any specific geographical information of note. It therefore, becomes rather difficult to determine exactly where Schaffrath came upon this metadata information for the Essen Collection encodings.

One could make an argument regarding the primary source material and genre/geographical information that perhaps the information Schaffrath originally recorded comes from early editions that are no longer available or from similar sources by the same authors. In this case, it seems likely that the disagreement between the sources suggested by the CCARH and the actual metadata information encoded in the

28

Essen Collection is a result of differing materials. However, one issue with the primary sources encoded in the Essen Collection that cannot be eased by the explanation given above pertains to the German identity of the folksong materials. The problem with the assumption that the Collection consists of German folksongs from the nineteenthcentury lies in how we define both the time and the place. Germany as a unified nation did not come into existence until 1871, and what would be considered German culture would not be unified until after the World Wars. Therefore, the problem with defining German folksong becomes one based heavily on geographical location, time period, and cultural identification. Given that certainty regarding location and time cannot be established, we must then rely on cultural identification in order to define what folksongs are considered

“German”. We can make an initial and rather simplistic assumption regarding the

German identification of folksongs in the Essen Collection based on the fact that the original archivists considered themselves to be German individuals and that they titled their books often with the words “German folksong” somewhere in them. Just because the archivist elected to label the folksongs German does not actual make them German folksongs. In the nineteenth century, what regions and cultures are allowed to be considered a part of “German” culture and which are forced to be separated from the category of “German” culture (and this question is immensely troubled when we begin to consider earlier time periods)? Conversely, given the power of the nationalist movement in German-speaking lands during the nineteenth century, which regions and cultures are forced to be considered German? Perhaps just as troubling is the fact that the nineteenth century brought with it the height of the Austro-Hungarian empire with the Austrian

29 region being an immensely powerful German-speaking region while the Hungarian region was forced to adopt many Austrian cultural practices despite having their own separate identity. Similar situations occur across many parts of Central Europe with the

Bohemian and Prussian regions also being considered parts of the German culture with varying levels of willingness. Regarding the folksongs of the Essen Collection, how are we to understand what it means for a folksong in the collection to be given the label

“German” given our knowledge of the tenuous circumstances of the nineteenth century?

If we assume that Schaffrath’s geographical location metadata are correct for the entries in the Essen Collection then we find that the problems of cultural identity have proliferated throughout the entire database. Songs coming from Czechia, Switzerland,

Denmark, Poland, Hungary, Italy, Austria, France, and the Low Countries can be found throughout the Collection. At one point or another in history, various parts of these nations have found themselves under German rule or have been identified with German culture in some way. However, just because a region is dominated by a specific cultural power (whether willingly as was the case during various points in this history of nations like Bohemia and Hungary, or otherwise) does not necessarily mean that the region in question is a member of the dominant power’s cultural community. To take this a step further, even if the region willingly and actively chooses to identify with German culture

(as was the case in many parts of 19th-century Prussia) it does not mean that the people living in those regions are in agreement with this decision or that they are willing to part completely with their separate cultural identities.

30

Historical anthropologists have noted for quite some time that cultures, and the regions in which they are located, change frequently over time. A cultural group or region that was once a willing and active participant and contributor to the sense of a Germanic culture may not feel the same way today. In discussing the nature of German cultural identity, I only wish to draw attention to the troublesome nature of labelling folksongs in the Essen Collection as “German” folksongs. I also want to point out the impossibility for modern scholars to be able to retroactively determine which of the folksongs actually belong to a different culture or identity. Therefore, with these concerns in mind we might ask how these issues might affect the content of the Essen Collection.

While archivists were collecting materials for the primary sources that would come to make up the Essen Collection, one of the biggest issues with the malleable definition of “German folksong” arises concerning which music gets to be collected.

Given the wide geographical range of songs included in the Collection, it seems possible that the original archivists were collecting folksongs from regions all sharing the use of the German language. Therefore, folksongs found in the Bohemian and Hungarian regions, if there existed versions with German translation, would eventually find themselves placed in the Essen Collection. Similarly, folksongs found in Austria and

Switzerland would also be included in the Essen Collection due to their strong relationship with the German cultures to their north regardless of cultural superpower- like quality of the Austro-Hungarian Empire and the immense cultural diversity of the

Swiss region. Conversely, songs that very well may have had Germanic origin but now only exist in another language would be quickly skipped over during the initial collection

31 efforts. Given the fluidity of cultural identity, it seems entirely likely that this would have happened to a number of folksongs that, while no longer to be found with German text, could easily fit the requirement of being “German”. Many of the assumed primary sources for the Essen Collection present only folksongs that have lyrical information.

Needless to say, this form of presentation strongly suggests that the original archivists did not collect any instrumental folksong material.3

Looking back to our discussion of genre, we might then also wonder whether certain genres of folksong performance were either preferred or ignored over others during the collection process. For instance, there seems to be a large representation of children’s songs in the Essen Collection, appearing across several subsections, and many potential hymn songs, suggesting a preference for music of a spiritual quality. However, beyond these two socially related cultural settings, there do not appear to be many folksongs coming from other social contexts. For instance, it is not entirely clear which, if any, songs in the Essen Collection are classified as drinking songs, work songs, or songs used in other forms of social gathering. There also appears to be large number of songs based on musical genres such as the ballad. While ballads can appear in folksong setting, in the nineteenth century, ballads were also popular song forms for art music traditions. Given that social context is not typically given for these songs, one wonders whether these ballads were originally meant to be published musical materials written by career composers or actual recorded folksongs found in various places across the cultural

3 Of course, it remains to be seen whether there exists instrumental folksongs that do not have any sort of sung text attributed to them. 32 community. If the former happens to be the most accurate description of the origin for these ballads, then we are suddenly confronted with the problem of having somewhere between 680-1,000 songs in the Essen Collection that are not technically “folk” songs.

Lastly, it is important to recognize that since the first presentation of the Essen

Collection in 1995, scholars have continued to work with EsAC in order to provide more and more databases of folksong materials. Professor Dahlig-Turek has overseen much of the archival work that has been done with EsAC project in the years since Schaffrath’s death. In that time, one substantial database and several small collections of folksongs have been added to the project covering Eastern, Southern, and Western Europe and even various parts of China. A variety of other national folksongs have been added to the collection, some from the Americas, a small portion from Africa, and even a fairly large portion from modern-day nations outside of Germany in Europe. The addition of folksong materials outside of Central Europe as well as the controversy surrounding the origins of the Germanic folksongs themselves suggest that perhaps we redefine what the

Essen Collection actually means. In professor Dahlig-Turek’s words, “[The] Essen

Collection is an informal term used for the oldest encodings…These new repositories do

NOT belong to the Essen Collection.” (emphasis in the original) “In [Dahlig-Turek’s] opinion, there is not much use of speaking about [the] Essen Collection, but rather the

EsAC Collections”(Dahlig-Turek, personal communication, January 19, 2020). As the current overseer of the EsAC project, perhaps we begin to adopt professor Dahlig-

Turek’s manner of thinking about the Essen Collection ourselves. The Essen Collection is a collection of musical data. What that data actually tells us is up for debate. In addition,

33 we might consider how much of role our “taking” of the data in the Essen Collection has affected the way we think about this database. How much of what we use the Essen

Collection for is a forcefully taking of data (or even better, capta) that may or may not actually be there? How much of the data collected by scholars using the Essen Collection comes as an artifact of what the Essen Collection can tell us about the tumultuous population of musical information that it represents?

The Structure of the Essen Collection

While the origins of the Essen Collection is certainly interesting, we might begin to ask ourselves whether the actual content of the Essen Collection also shares in the intrigue.

Of the roughly 6,000 songs in the Collection, only 5,370 are nominally German (Huron,

1995). The rest of the songs in the collection come from different areas in Europe. Recall that it is entirely possible that these songs have no relationship with Germanic culture. In

David Huron’s 1995 encoding of the Essen Collection into the **kern format, the folksongs with non-Germanic origins have been placed into an entirely different subsection regardless of which primary source material the folksong was pulled from. For example, a folksong that is said to have originated in Denmark but was found in the primary source material that makes up the “erk” subsection was placed into a separate subsection labelled “danmark”. The result of Huron’s manual splitting of folksongs from

Germanic regions and non-Germanic region is an extra level of division above that of the subsection level. We might label this upper-level subsidiary division as the Essen

Collections constituent parts. Of the 6,220 folksongs in the Essen Collection, 850 34 folksongs, supposedly related to Germanic culture (at least as Schaffrath understood them), were found in regions outside of modern Germany. This leaves 5,370 folksongs left in the Germanic-origin part of the Essen Collection.

Before continuing further, it is important to point out that Schaffrath’s original layout of the Essen Collection did not include this “part”-level division of folksongs into those of Germanic-region and non-Germanic-region origin. In Schaffrath’s original layout, many of the subsections of the Essen Collection mixed in folksongs from

Germanic and non-Germanic regions. Again, the idea behind this was probably that the non-Germanic-region folksongs were still authentically “German” in quality. Perhaps as a way of dealing with the issue of “German quality” in the non-Germanic songs of the

Essen Collection, Huron made the executive decision to add another level of division into the Essen Collection’s layout (Huron, personal communication, March 3, 2020).

Regardless of whether this needed to be done, the result seems to have caused some confusion with many scholars who have used the Essen Collection data. Out of the roughly 106 papers that have made use of the Essen Collection as an experimental test or training sample, 25 mention in passing that the Essen Collection contains some German

6,000 folksongs while 12 mention that the Essen Collection contains some German 5,000 folksongs.4 Depending on the way one understands the context of the Essen Collection, either of these interpretations could be correct. However, none of the papers discussed

4 The remaining 69 studies make no mention as to the total size of the Essen Collection. Often these studies make use of a specific subset of the Collection therefore eliminating the need to report on the total overall size. 35 above make any mention of the division between the Germanic-origin folksongs and the non-Germanic-origin folksongs.

As an example, imagine you are a scholar interested in using the Essen Collection for your next big research project. You have two different formats of the Essen

Collection to choose from, based solely on your own preferences. If, for some reason, you choose to use the EsAC formatted version of the Essen Collection, you would find that there are 6,220 German folksongs in the database and this would be the number that you report in your paper, all the while having obvious indication that only 5,370 of those folksongs are actually from the German region. Conversely, if you chose to use the

**kern-formatted Essen Collection then you would likely believe that only 5,370 of the folksongs in the database are “authentic German” folksongs without any indication that the other 850 folksongs are supposedly “German” folksongs as well. Herein lies the problem. Without a strong definition of “German folksong” in the Essen Collection, it becomes impossible to know for sure whether the total sum of the folksongs in the Essen

Collection equals 6,220 or 5,370. In keeping with the original assumptions of the two primary encoders, perhaps it is best for future scholars to align themselves with whatever interpretation of “German attribution” is held by encoder of the Essen Collection’s format that is used for the scholar’s current research. That is, if one chooses to use Huron’s

**kern format, then perhaps it is best to think of the Essen Collection as having 5,370

German folksongs and if one chooses to use Schaffrath’s EsAC format, then they should think of the Essen Collection as containing 6,220 folksongs. In either case, it behooves future scholars to take the issue presented by the two different formats of the Essen

36

Collection very seriously should they choose to use the database in their own future research.

Regarding the subsections themselves, each format version of the Essen

Collection is structured in a slightly different way. Many of the subsections within the original EsAC format pull from the same source material and seem to be divided into different portions of that source material. For instance, the subsections that supposedly all come from the same book of German folksong ballads, Lieder, Balladen und Romanzen für das Piano-Forte (Zelter, 1812), are each labelled “balled” following by some number.

The resulting ballad subsections of the EsAC Essen Collection are read as “ballad10”,

“ballad20”, “ballad30”, “ballad40”, “ballad50”, “ballad60”, “ballad70”, and “ballad80”.

It should be noted that the ballad subsections are the most numerous of the split subsections in the EsAC Essen Collection. It is unclear how Schaffrath divided these different subsections nor is it clear how the divisions relate to the original source material. For instance, the largest overall combined subsections in the Essen Collection are pulled from the “erk” subsections (containing 1,705 folksongs). However, there are only four different divisions of the “erk” subsection in the EsAC format. Conversely, the

“ballad” divisions of the EsAC Essen Collection only contain a total of 687 folksongs, less than half as much as the “erk”, but with a total of eight unique divisions. The

“zuccal” subsection of the Essen Collection contains 616 folksongs but exists in the

EsAC version as only one subsection with no divisions. A graph of the overall counts of folksongs per subsection is provided below in Figure 1 below. While this issue of divisions of subsections is not necessarily that troublesome, the fact that we do not know

37 why or how Schaffrath divided up the subsections in the Essen Collection is rather troublesome and only adds to the mystery surrounding the origins of the Essen

Collection.

Figure 1. Total counts of folksongs in the Essen Collection

While transcribing the EsAC version of the Essen Collection into the widely accessible **kern format, Huron elected to make some executive decisions regarding the definition and distribution of the original twenty-five subsections. Perhaps the most 38 striking decision made was to reduce the original twenty-five subsections into just twelve. According to Huron himself, the reasoning for this was to reduce the number of divisions in the original EsAC format. Huron mentioned that the divisions encompassing the original twenty-five subsections were a result of space limitations on 1.44MB floppy disks, the primary mode of distribution for the Essen Collection during the mid-90’s.

Because the **kern format would be hosted via the CCARH online platform which was not limited by size constraints, it seemed reasonable to reduce the number of divisions between subsections (Huron, personal communication, March 3, 2020). In the **kern version of the Essen Collection, only one of the original twenty-five subsections is divided into two different divisions, the “altdeu1” and “altdeu2”. The suggested primary source material for these two subsections comes from a book of German folksongs supposedly ranging from the twelfth to the seventeeth centuries and collected by 19th- century German folksong archivist Franz Magnus Böhme (1895).5 This primary source provides 625 of the total 5,370 German folksongs in the Essen Collection. Given that there is no longer a need to preserve the original twenty-five subsections based on size constraints, it is not clear why the division still exists between these two subsections today. However, assuming that the suggested primary source for the “altdeu” subsections is correct, it seems to be the case that “altdeu1” covers all of the folksong of the original source from number 1 to number 362 where then “altdue2” covers the folksongs from number 363 until number 660, the final folksong in the book. In the primary source itself,

5 Interestingly enough, Böhme served as either the editor or original collector for several of the primary sources suggest by the CCARH. His name is attributed to the “boehme”, the “erk”, the “kinder”, and the “altdeu” subsections. 39 there does not appear to be any significant reason to suggest that dividing up the folksongs at the three hundred sixty-third folksong would be a good place to do so.

Instead, it is probable that Schaffrath (and by extension, Huron) decided to divide the book at this point simply because the area around the three hundred sixty-third folksong happens to be about the middle of the book. In personal correspondences with Huron, he expressed that he was not certain why it was that these two subsections were divided and elected to keep it this way in his **kern encodings. Regardless of the how the decision to split the book was made, it is not entirely clear why Huron kept the original divisions of the “altdeu” subsections while simultaneously combining Schaffrath’s various other subsection divisions.

It should be noted that neither Schaffrath’s nor Huron’s organization of the subsections in the Essen Collection should be considered somehow “better” or more

“preferable” than the other. One might argue that Huron’s organization is easier to understand and more succinct. However, Huron’s organization assumes that Schaffrath’s decision to divide the subsections was based only on an arbitrary issue of floppy disk space availability and that the different divisions are not so different from one another to warrant existing as their own separate subsection. While this is certainly possible and perhaps even likely, it is nonetheless dangerous to assume this is the case. At best, we can only make guesses as to the reasoning behind his decision and scholars who use the

Essen Collection, in either format, would do well to keep these issues in mind.

The question of information loss discussed in the issues above is one that pervades much of the Essen Collection and any such corpus encoded from musical data. 40

It is inherently the case that, during the process of encoding and/or transcription, information present in the source material will likely be lost. We can see this loss of information also in the Essen Collection itself. Assuming the suggested primary source material is correct, we might begin to ask ourselves what kinds of information has been lost between the original printed materials and Schaffrath’s EsAC transcriptions.

Furthermore, one interested in the relationship between performances and collectors of performance materials (in this case, the folksongs themselves) might ask the same question with respect to the traditional oral presentation of the folksongs and their initial transcription by the folksong archivists who collected these materials. While it would certainly be of interest to address the latter topic, I instead defer to the work of ethnomusicologists and cultural anthropologists who have struggled with the issue of oral and written traditions with much more rigor.6 Additionally, attempting to address the loss of information that exists between the oral tradition and the original transcriptions is immensely confounded, in the case of the Essen Collection, by our lack of knowledge regarding the exact origins of the source materials. Considering these issues, I will instead focus on the relationship between the printed primary source material and that of the EsAC encodings of Schaffrath’s collection.

6 The amount of writing that covers this topic is vast. For the purposes of this dissertation, I draw attention to two particulars scholars who have struggled with this topic. Blacking (1974) and Nettl (2010) have both discussed how information is lost whenever scholars decide to transcribe music derived from an oral tradition. Written music is not currently able to represent all of the intricate aspects of music and often only presents guidelines for how a song should be performed. This is especially true the more that a music differs from the Western tradition where performers in the oral tradition could be using a variety of musical techniques that are not conducive to recording in Western notation. 41

Recall that we truly do not know exactly why Schaffrath chose the supposed primary source materials as the materials to be encoded in the Essen Collection. The

CCARH has suggested that the primary source materials were chosen due to their use in

Wolfram Steinbeck’s dissertation on German folksongs (1982). However, in discussions with professor Dahlig-Turek, it is not entirely clear that this is the case. Dahlig-Turek claims, “[she] NEVER heard from Prof. Schaffrath about ANY influence from

Steinbeck’s dissertation” (emphasis in the original) (Dahlig-Turek, personal communication, January 19, 2020). Of course, Dahlig-Turek does not assert with certainty that Schaffrath was not influenced by Steinbeck’s work. She only mentions that there did not seem to be any real relationship between Schaffrath’s and Steinbeck’s work.

The scholars at the CCARH mention that Schaffrath’s Essen Collection includes much of the same primary source material as that found in Steinbeck’s own collection of folksong materials (1982), and while it is entirely possible that both scholars would have the same primary sources in their collections, that does not necessarily mean that Schaffrath pulled his collection materials from Steinbeck’s work. A few of the primary sources suggested by the CCARH have been cited in several highly regarded scholarly works within the past century, easily suggesting that Schaffrath could have encountered these collections completely on his own.7 Regardless of whether Steinbeck’s work had an impact on

Schaffrath’s collection, this does not change the fact that all evidence points to the Essen

Collection being essentially a glorified convenience sample. Schaffrath was immensely

7 This has been particularly apparent in ethnomusicological work in recent years. See Morgenstern (2018), Davis (2019), Tayler (2017), and Robb (2016) as some examples. 42 interested in two specific topics related to folksong research: refining and improving the way we collect folksong data, and the study of Southeast and East Asian music. Given that collections of Southeast and East Asian music were probably not widely accessible during Schaffrath’s work in the 1970’s and 1980’s, it is highly unlikely that he wanted to

(let alone be able to) attempt to encode such materials with the Essen Associative Code.

As of today, it is unknown exactly why Schaffrath chose the specific primary sources represented in the Essen Collection to encode in the EsAC format. This fact combined with Schaffrath’s lack of expertise in European folk music strongly suggest that the primary sources of the Essen Collection were chosen mainly because of convenience. If this is the case, then we might suggest that the Essen Collection is essentially a large convenience sample put together for the purposes of presenting the abilities and possible uses of the Essen Associative Code. This view of the origins of the Essen Collection is echoed by the words of professor Dahlig-Turek who stresses that the Essen Collection’s purpose is not presenting a large database of German folksongs, but instead to highlight the Essen Associative Code. She writes that, “I may only suppose that the choice for [the primary sources of the Essen Collection] was (at least partly) determined by IPR issues

(public domain)… For both of us, Schaffrath and myself, the real topic was (and for me still is) EsAC, not the Essen Collection…” While Dahlig-Turek does not explicitly claim that the Essen Collection was a convenience sample, it is easy to read from her own statements regarding the origins and how scholars should understand the Collection that

Schaffrath’s sampling decisions were born out of convenience rather than specific purpose.

43

In addition to the loss of information that has arisen in regard to the reasons behind Schaffrath’s sampling process, other losses appear simply as an artifact of the

EsAC encoding process. Many, if not all, of the folksongs presented in the suggested primary sources all include lyric information in addition to the printed scores.

Unfortunately, due in large part to the initial limitations of the EsAC format, lyric information was chosen not to be included in the original transcriptions. In the years since the first versions of the EsAC format, scholars have attempted to add lyric information into the encodings. In Dahlig-Turek’s words, “The display was rather ugly, so we experimented with Music-TEX for nice results. Since for musicological studies this was of limited use, we stopped working on this.” For better or worse, the original lyric information continues to be a lost artifact in encodings of the Essen Collection, distinctly absent from both the EsAC and **kern formats.

In addition to this loss of lyric information is a loss of two more rather substantial bits of musical information: key areas and phrases. For most of the suggested primary source material, neither key information nor phrase information is given for any of the songs in any book. Instead, most folksongs follow the same pattern where a key signature is given in addition to a meter designation (in many cases, the time signature is absent) followed by the melody itself. While it is certainly nice that the original archivists were able to collect information related to key signatures, the actual key areas, with specific model designations, were suggested by the encoders of the Essen Collection themselves.

The biggest issue with encoders determining modality based on key signatures alone is the possibility that major and minor modes might be confused. That is, it is entirely

44 possible that a folksong with three flats in its key signature may be incorrectly labelled as

C minor when it is, in fact, Eb major. Unfortunately, there does not appear to be any real solution to this problem as there does not exist any specific statements regarding modality for the original folksong sources. We must instead rely on the music analysis skills of the original encoders as being adequate enough to provide reliable interpretations of mode based on key signatures alone. Additionally, meter designations were often extrapolated by the encoders based on the ways that the original archivists inserted barlines into the scores themselves. While this method of determining meter is not inherently incorrect, it is important to recognize that encoders may have been biased in their decisions to encode specific meters. Admittedly, the issue of meter designation does not come up very often in the primary source materials, at least as far as the author is aware. However, that does not change the fact that a small portion of the Essen

Collection may be affected by this problem. Regardless, the issue of key designation is a problem across the entire collection.

Perhaps more troublesome than the question of key and meter is the issue of phrase segmentation. In Chapter 3, I mention that the bulk of the work on melodic expectation that has been done within the past twenty years relies on the phrasing information present in the Essen Collection. This reliance on phrasing information is immensely troubled by the fact that none of the original folksong materials in the Essen

Collection actually contain any phrase-related information. There are no phrase markings, no formal designations, nor even many special barlines or repeat signs that might indicate phrase boundaries. Instead, the original encoders of the Essen Collection took it upon

45 themselves to insert phrase boundaries in places where they felt were most likely points of phrase segmentation. There were many possible reasons for encoders to include such information and we can only speculate as to which reason is most accurate. Much of that information has been lost since 1995. However, given that very little, if any, of the widely available electronic databases of music included phrasing information at that time,

Schaffrath and his colleagues may have felt that including such information would have been especially useful for future scholars. Of course, if this is the case, then Schaffrath was correct. In the years since the Essen Collections original publication, many scholars have used the phrase boundaries provided in the Essen Collection to conduct a wide variety of studies. The work of these scholars has often helped define entire subdisciplines in music cognition and empirical musicology (the subdisciplines of melodic expectancy and modelling phrase segmentation are prime examples).

Unfortunately, the echoes of encoding bias still ring true through the Essen Collection even today.

Perhaps the biggest issue in relying on the original encoders to interpret and insert phrase boundaries is related to the notion of the “jump-phrase” phenomenon. Bod

(2002d), in his extensive work attempting to develop computational models that can accurately segment musical data found that nearly 15% of all phrases in the Essen

Collection exhibit a phenomenon he termed “jump-phrase” or “jump-boundary”.

According to Bod, the “jump-phrase” is a phrase boundary that has been precariously placed on either side of a large melodic interval rather than directly inside it. That is, throughout his work Bod notes that many musical phrases tend to begin with the

46 consequent note of a large melodic interval and end with the antecedent note of such an interval. The problem with the 15% of folksongs in the Essen Collection with “jump- phrases” is that a phrase whose beginning or ending is found either before or after a large melodic leap rather than directly inside the leap is highly unlikely to exist, at least according to all computational modelling of melodic segmentation that currently exists.

Therefore, it is entirely possible that, at least according to Bod, 15% of phrase designations in the Essen Collection are incorrect. Given that there exists somewhere around 35,808 explicitly marked phrases in the Essen Collection (Eerola & Bregman,

2007), Bod’s claim suggests that over 5,000 of those phrases are incorrectly placed.

Assuming that Bod is correct, then this is a problem. To make matters worse, the phrasing issues that Bod addresses are only the most widely discussed issues. There exist many more potential issues regarding phrasing in the Essen Collection that have yet to be addressed (a topic that is addressed in Chapter 2).

Before diving into the structure of the various subsections in the Essen Collection, it is also important to point out some of the smaller issues of lost information. Folksongs in the primary source materials occasionally include other kinds of musical information that is not encoded in the Essen Collection. This includes dynamics, articulations, expressions, and ornamentations. Such information is certainly rare throughout the primary source material. However, a few of the suggested primary sources seem to include rather substantial transcriptions of folksongs, even including harmonic and/or accompanimental information (Zelter, 1812; Vogel, 1845). Due to the focus on (and possibly limitations because of) the monophonic quality of melodies in the Essen

47

Collection, this information has also been lost. Of course, with developments in the EsAC format since 1995, it seems entirely possible that we could have a more modern version of the Essen Collection that includes polyphonic material in the future. In the meantime, the Essen Collection is focused on providing pitch, rhythm, and phrasing information

(albeit rather hazardously) in its current form.

Shortly after the publication of the Essen Collection in the EsAC format, Huron took up the task of encoding the EsAC materials into the **kern format in order to increase user-accessibility to the Essen materials. In any situation where musical information is transferred across different modes of presentation, it is impossible not to lose part of the original information. Huron attempted to preserve the original EsAC information in the Essen Collection during his conversion to **kern, and for the most part, it seems as though Huron accomplished this task rather successfully. In addition to preserving the pitch, rhythm, and phrasing information, Huron also included metadata records for each folksong matching the original entries provided by Schaffrath and his colleagues. This metadata information includes geographical location, supposed year of origin (information provided by the primary source material), genre information, and even the original encoder’s notes. In addition, Huron also elected to include measure information based on the meter and duration information provided in the EsAC materials as a means of making the folksong transcriptions more easily readable for **kern parsing tools. If any information has actually been lost during the transcription from EsAC to

**kern, then all evidence suggests that it must arise only from potential human error.

6,249 folksongs (the total number of EsAC materials available in the **kern format

48 which include non-Germanic folksongs) is a lot of material to transcribe. Given that

Huron spent a considerable amount of time working on the transcriptions, it is entirely possible that transcription errors may have been made during this process. However, if it is the case that such errors have occurred, it is entirely possible that these errors should have been noticed by one of the many scholars who have worked with the Essen

Collection in both formats.

In order to be as clear and precise as possible with the information available, I refer to the subsections in their **kern format. This is primarily because it is not clear why Schaffrath divided the subsections into the initial twenty-five subsections. Some of what follows includes statistical properties of the subsections and because of this, it is perhaps best to address the subsections in a form that most accurately represents the suggested primary source material. In light of the choice by Huron to preserve the original division of the “altdeu” subsections, I will also keep the division for the purposes of the following statistical property analyses. This leaves us with only two subsections that have not yet been accounted for in this chapter: the “test” and “variant” subsections.

Little, if any, information is known regarding the origins and purpose behind these two subsections. Each are exceptionally small, with the “test” subsection including only twelve folksongs while the “variant” includes twenty-six. Contrary to what one might imagine based on the names of these subsections, there is very little to suggest that they are somehow test-related or that they include the only folksong variants in the Essen

Collection. However, regardless of the origin or purpose behind these subsections, they are fully encoded materials in the Essen Collection and have been preserved across the

49

EsAC-to-**kern transcription process and, as such, will be included in the following analyses.

The Issue of Capta Versus Data

Before moving on to the final analyses of this chapter, it is important that we address the role that musical information plays in our understanding of the Essen

Collection. In his work on communicology, Richard Lanigan discusses the topic of capta versus data through presenting the methodologies that phenomenologists and positivists use while carrying out research tasks (1994). Both kinds of scholars make use of the same two processes of observation and measurement. However, each scholars moves through the process of observation and measurement in different ways. According to Lanigan, the phenomenologist begins with acknowledging their personal experience with an event/thing (observation) and follows up with judgement of that event/thing through measurement derived from their personal consciousness of the original phenomenon.

Throughout the observation process, “S/he does not pretend to be in the situation (so- called “objectivity”) or to be omnipotent (so-called “authority)” (parentheticals in the original) (Lanigan, 1994). While recording and analyzing, the phenomenologist moves backwards, reflecting on what was experienced and analyzing his or her own personal interpretation of the phenomenon. In this process, the phenomenologist “takes” from the experience, describing what it was that was “taken” while reporting on the phenomenon itself. Lanigan calls this “taking” of the experience capta and suggests that this is the only guaranteed way that a researcher can be both accurate and abstraction in their depiction 50 of the results. The positivist process involves a simultaneous process of observation and measurement. He writes that, “Since all experiencers are assumed to be objectively the same (ironically, a very subjective assumption), all forms of consciousness

(“measurement”) will have the same experience (“observation”).” (Parentheticals in the original; Lanigan, 1994, p. 113). In this process, the researcher is “simply researching what is researched” (Lanigan, p. 113) without any regard for the impact that the researcher’s own personal experience has on the analysis of the outcome. Lanigan calls this “giving” of an assumed logic, unaffected by individual experience, to a phenomenon in order to predict whether an outcome is correct or not data.

We might better understand how scholars have understood the structure and origins of the Essen Collection by making use of Lanigan’s distinction between capta and data. As we will see in Chapter 3, the bulk of the research conducted using the Essen

Collection adopts a positivist perspective that interprets the Collection as data, a phenomenon that produces the same experience for each instance of the same observation in . Therefore, much of the research involving the Essen Collection merely “gives” some sort of observation to the Collection itself that is then recorded or “measured”. Because all researchers are purported to be objectively the same in the manner by which they observe phenomenon in the Essen Collection, the phenomenon measured through this observation should be the same. In the positivist approach, the only difference between scholarly studies involving the Essen Collection is in what phenomenon in the Collection scholars choose to observe. It does not matter so much how the phenomenon is observed,

51 because the process is purported to be objective and therefore unaffected by individual consciousness, the measurements taken from these observations should be the same.

If we are to assume that Lanigan is correct in stating that the phenomenologist approach is the more correct research methodology, then instead of “giving” to the Essen

Collection and reporting on the outcomes, we might ask ourselves what we “take” from our observations of the Essen Collection and how those “takings” or capta differ from one another. As an example, imagine a researcher is exposed to the songs in the Essen

Collection. Through their experience with the Collection they make the observation that there appears to be a lot of scale degree 4 across folksongs in the Collection. They then count all instances of scale degree 4 across folksongs in the Collection and find that the total is rather low. At the end of this project, they report that, while they felt that there was an abundance of scale degree 4 in the Essen Collection, their raw counts of all instances of scale degree 4 produced a low number. From this report they reach the conclusion that there must be another explanation for their original experience with the

Essen Collection (perhaps there are more instances of scale degree 4 in the Collection then there is in musics they are more familiar with or perhaps there is a higher percentage of scale degree 4 per each individual song rather than overall). In this example, the researcher takes from his or her own personal observation an experience (capta) that is then followed up by an analysis.

Another perspective on the distinction between capta and data comes from

Christopher Chippindale (2000). While discussing the nature of archaeological information, Chippindale mentions that data, information that is given from observations

52 of artifacts and the like, does not accurately describe how archaeological information is received. Instead, he mentions that it would be better to use the word capta to describe this process as researchers are “capturing” information from the material regarding some specific phenomenon. Notice that this is a slightly different definition of the two terms compared to Lanigan’s. However, the two scholars do share a similar point in that there are issues regarding how information is used for research purposes. The bulk of the scholarly research making use of the Essen Collection would be categorized as capta, according to Chippindale, as scholars are merely “capturing” information from the songs in the Collection in order to better understand their own research topic. Another way to understand this is that the musical information in the Essen Collection is not necessarily

“given”. Musical information in the Essen is just that, merely information. The scholars who choose to interact with the Essen Collection choose which information to pay attention to, observe, and analyze. The information that is “captured” by these scholars may or may not be completely relevant for their research topic, or they might be unintentionally ignoring relevant information provided for elsewhere in the Collection. I might therefore argue that many of the scholars using the Essen Collection are acting as if the information is capta rather than data.

Where Lanigan and Chippindale differ is concerning how exactly the terms capta and data are defined. For Lanigan, information that is taken by way of personal experience with observation and measurement should be considered capta. For

Chippendale, the term capta is used to describe the fact that scholars are “capturing” information that is relevant for their research. For the most part, the general process in

53 which scholars choose to take certain information over others is the same between

Lanigan’s and Chippindale’s definition. However, Lanigan is approaching the topic more from a methodological standpoint in which he describes how scholars should be working with information where Chippindale is describing how scholars actually are interacting with information. For both Chippindale and Lanigan, it seems rather difficult, if not impossible, to interact with information is such a way as to accurately define it as data, that information which is readily given to the researcher. Each and every scholar comes to a research topic with their own agendas and biases and these biases define how they interact with the information. Both Chippindale and Lanigan agree that, while this process may not be the most objective way to handle scholarly research, it is the most accurate in describing how research is actually conducted.

The issue of capta and data, or the issue of subjectivity and objectivity has been a popular topic for digital humanities scholars in recent years (Eyers, 2013; Gardiner &

Musto, 2015; Golumbia, 2009; Kirschenbaum, 2016). Scholars interested in using computer-assisted research to aid in the study of humanities topics often run into issues regarding subjectivity. The authors above would have us recognize that the individual human experience drives much of our understanding of humanities, and therefore music, research. This is relevant to the Essen Collection because scholars should be careful in what they say and how they talk about the Collection. As the discussion in this Chapter has already addressed, there are a considerable number of potential issues with how scholars have understood the origins and structure of the Essen Collection. The Essen

Collection is capta after all, in both Lanigan’s and Chippendale’s definition of the term.

54

We, as scholars, take from the Essen Collection whatever we need in order to conduct our research. We take from the Essen Collection measurements based on various observations in order to test claims that are inherently subjective. If we are to continue to conduct research in the best manner possible, it is important that we recognize and address these issues whenever possible.

Statistical Properties of the Essen Collection

Much work has been done uncovering basic statistical properties of the Essen Collection

(a topic discussed in Chapter 3). However, this work has not been done on the individual subsections of the Essen Collection. Given that both Schaffrath and Dahlig-Turek find that it is the individual collections provided in the EsAC materials that are much more important than some notion of a generalized “Essen Collection” as scholars have come to know, perhaps we should begin to explore the properties of these individual subsections.

After all, nearly every one of these subsections pull from a different source that might be interested in preserving different folksongs based on a variety of factors. In order to fully understand the nature and content of the Essen Collection, it is important that we examine what kinds of musical properties are exemplified by the collection’s constituent parts. It may be the case, for example, that our understanding of the Essen Collection’s scale- degree and pitch distribution is heavily biased by the “erk” subsection, which represents nearly 32% of the folksongs encoded in the collection. Likewise, we might find that the interval content of the “fink” subsection contains some highly interesting properties that are obscured by the more “normal” interval content of the other eleven subsections. In 55 order to clear up any potential issues that may be present in the statistical properties of the Essen Collection’s subsections, I provide a brief analysis of these properties across the Collection’s twelve subsections. More specifically, I examine the overall scale- degree, interval, and duration content across the twelve subsections and discuss how these properties influence the way we consider both the subsections alone and the Essen

Collection as a whole.

The overall scale degree distribution of the twelve subsections of the Essen

Collection is provided below in Figures 2 and 3. In order to be as clear as possible, the twelve subsections have been divided into two parts with six subsections each.

56

Figure 2: Tallies of scale degrees across six subsections of the Essen Collection as percentages. Each colored portion represents the percentage of tallies of a given scale degree found across all folksongs in each subsection of the Essen Collection

57

Figure 3: Tallies of scale degrees across six subsections of the Essen Collection

As can be seen in the above figures, the scale degree distribution between the twelve subsections is fairly normal. In nearly all twelve subsections scale degree 5 appears more frequently than any other scale degree. The two exceptions are the “altdeu1” and “test” subsections. Given the small number of folksongs in the “test” subsection, it is not entirely surprising that it provides rather strange findings. Of more interest is the fact that

58 the “altdeu1” subsection differs slightly from the other subsections in its increased number of scale degree 1. Based on scale degree distributions of Aarden (2003) the most prevalent scale degrees across the entire Collection are 5, 3, and 1, respectively. Nearly all of the subsections provided above reflect this finding with the exception of the

“altdeu1”, “test”, “fink”, “variant”, “altdeu2”, and “ballad” subsections. Across these six subsections, the general trend in difference from Aarden’s findings are that scale degree 1 appears more often than scale degree 3. Again, these differences are rather small but are apparent, nonetheless.

On the other end of the spectrum, as in Aarden (2003), scale degrees 6 and 7 appear least often across the subsections. The only exceptions to this are the “kinder” and

“variant” subsections. Again, the rather small number of folksongs in the “variant” subsection may have an effect on this result. Interestingly, while scale degree 7 appears least frequently in the “kinder” subsection overall, just as in the other subsections, the

“kinder” subsection also has the lowest proportion of scale degree 7 overall when compared to the other subsections. Conversely, the “kinder” subsection also has more scale degree 4 than all other subsections (though still with a rather small proportion overall). Overall, scale degree 2 appears less frequently than scale degrees 1, 3, and 5 but more frequently than all other scale degrees. This reflects similar findings across the entire Collection as in Aarden (2003). The only exceptions to this are the “altdeu2”,

“test”, and “variant” subsections.

As a whole, the scale degree distribution for the various subsections of the Essen

Collection closely align to the overall distribution findings suggested by Aarden (2003). 59

The subsections that most closely match up to Aarden’s distributions are the “allerkbd”,

“boehme”, “dva”, “erk”, and “zuccal” subsections. Interestingly enough, the “erk”,

“fink”, and “zuccal” subsections also contain some of the largest number of folksongs across the Collection (the only other subsection to rival these three is the “ballad” subsection which contains more instances of scale degree 1 than the others). It should be noted that, when compared to scale degree distributions in Western art music (C. L.

Krumhansl & Kessler, 1982) these distributions are all slightly different. According to distributions in Western art music, we should expect to see and overabundance of scale degree 1, followed by 5, 3, 4, 2, and 7, in a major key context. The five subsections most closely reflective of Aarden’s findings all contain significantly more instances of scale degree 5 and 3 over scale degree 1. In addition, these same subsections also contain significantly more instances of scale degrees 2 over scale degree 4, another stark difference from the Western art music tradition. In fact, none of the subsections has a scale distribution that exactly resembles that found in Western art music. The closest candidates would either be the “altdue2” or the “kinder” subsections, but each still slightly differ from the Western art music distributions.

In general, while the various subsections of the Essen Collection differ only slightly in scale degree distribution from that of the overall distribution, it is important to keep in mind these subtle differences when conducting research using the Collection.

Most importantly, scholars should be wary of the representativeness of the “test” and

“variant” subsections. Both of these subsections differ wildly in scale degree distribution compared to the overall distributions and to the distributions of the other ten subsections.

60

Additionally, both of these subsections contain considerably less folksongs than the other subsections. Scholars interested in paying attention to the slight differences between subsections should also focus on the middle portions of the scale degree distributions as these portions reflect the most difference between subsections. That is, distributions of scale degrees 1, 3, 2, and 4 differ more across the subsections than do scale degrees 5, 6, and 7. Lastly, I would like to draw special attention to the “kinder” subsection, which supposedly represents 19th-century German children’s songs, and the “ballad” subsection, which is the least likely to contain vocal folksong melodies. As might be expected, the

“kinder” subsection contains the largest proportion of scale degree 1 and the smallest proportions of 7 and 4. This finding suggests that the “kinder” subsection has a slightly more pentatonic distribution than the other subsections. Conversely, even though the representativeness of the “ballad” subsection is questionable, its scale degree distribution is very similar to the overall distribution in the Essen Collection. Given that the primary source suggested for the “ballad” subsection contains scores of printed musical material, including piano, voice, and occasionally other instruments, rather than individual melodic lines with lyrical information, one might imagine that the scale degree distribution in the

“ballad” subsection would be more like the Western art music distribution. However, it is clearly the case that the “ballad” subsection is more closely related to the overall scale degree distribution of the Essen Collection.

The overall interval distribution of the twelve subsections of the Essen Collection is provided below in Figures 3 and 4. Again, each figure contains six subsections in order to enhance readability

61

Figure 4: Tallies of diatonic intervals across six subsections of the Essen Collection.

62

Figure 5: Tallies of diatonic intervals across six subsections of the Essen Collection

As far as this author is aware, while other scholars have created individual tallies of various intervals in the Essen Collection, only one paper covers the overall interval distribution of the entire Essen Collection (Marcus T. Pearce & Eerola, 2017). Their findings suggest that the most prevalent intervals across the Collection are the unison

63

(P1), major second (M2) and minor second (m2). For the most part, the interval distributions of the twelve subsections reflect this. The major second accounts for the largest proportion of intervals in nearly every one of the twelve subsections, followed by the unison and the minor second. Also like Pearce & Eerola’s findings, the largest intervals, the minor seventh (m7) and major seventh (M7) both constitutes the smallest interval proportions in each of the twelve subsections. Overall, the differences between the interval distributions of the twelve subsections are rather subtle.

The “kinder” and “variant” subsections both have the highest proportion of unison intervals (again, suggesting a simplistic pitch structure for the “kinder” subsection).

Conversely, the “test” and “zuccal” subsections contain the highest proportions of major and minor seconds. While it may be difficult to tell from the Figures, both the “kinder” and “dva” subsections contain the highest proportions of minor thirds. Overall, all of the twelve subsections contain proportions of perfect fourths and fifths that reflect the total distribution of the Essen Collection as suggested by Pearce & Eerola, where perfect fourths occur significantly more that perfect fifths. Lastly, the “altdeu” subsections both seem to contain the lowest proportions of major and minor sevenths.

Again, like the scale degree distributions, the interval distributions across the twelve subsections tend to be only slightly different from one another. Compared to the overall interval distribution of the Essen Collection, only a few subsections stand out.

These are the “altdeu” subsections due to their nearly complete lack of sevenths, the

“kinder” subsection for its overabundance of unisons, and the “zuccal” subsection for its overabundance of minor seconds. Other than these subsections, the slight differences in 64 interval distributions across the twelve subsections of the Essen Collection, while certainly should not be dismissed completely, might be considered negligible.

The overall duration distribution of the twelve subsections of the Essen Collection is provided below in Figures 6 and 7.

65

Figure 6: Tallies of durations in the Essen Collection. Specifically, breves, whole notes, half notes, quarter notes, eighth notes, eighth-note triplets, and sixteenth notes are included

66

Figure 7: Tallies of durations in the Essen Collection. Specifically, breves, whole notes, half notes, quarter notes, eighth notes, eighth-note triplets, and sixteenth notes are included

These final distributions of durations across the twelve subsections provide some final interesting points regarding the structure of the Essen Collection. First, notice that both of the “altdeu” subsections contain an abnormally large number of half- and whole 67 notes compared to the other subsections that are overwhelmingly filled with eighth- and quarter notes. It is entirely unclear why this is the case. Additionally, these two subsections also contain the only proportions of breves across the entire Essen Collection.

Given that there are no sixteenth notes in these subsections, it may be the case that rhythmic durations for these two subsections have simply been doubled in value. The only other subsection that deviates from the normal distributions is the “ballad” subsection, which contains slightly more half notes than sixteenth notes.

Overall, the primary rhythmic distribution across the nine of the twelve subsections seems to consist mainly of eighth notes, then quarter notes, sixteenth notes, half notes, and finally a very tiny proportion of eighth-note triplets. While no other scholars have looked at the rhythmic distributions across the entire Essen Collection, it is important to notice that the overall duration distributions for these nine subsections seem rather normal. Additionally, if we understand that the “altdeu” subsections have had the durational values in their folksongs merely doubled, then the variance in these subsections can be easily explained.

The pitch, interval, and duration distributions across the twelve subsections of the

Essen Collection do not wildly differ from one another. For the most part (especially for the durations), the distributions between the twelve subsections do tend to agree with one another. However, there are slight difference between subsections in the case of proportions of specific scale degrees and intervals that should not be ignored. Scholars should be careful when taking samples of the Essen Collection that they understand exactly what kinds of information they are more likely to get if they sample more of one 68 subsection over another. It could be the case that a study interested in faster rhythmic values might be troubled by using the “altdeu” subsections as the primary dataset.

Similarly, a study focusing on individual scale degree distributions should be careful to note that the “kinder” subsection seems to reflect more pentatonic pitch structures than the other subsections. Of course, overall, one might argue that the differences between these twelve subsections are rather insignificant. While it is true that, for the most part, the distributions in these twelve subsections do differ only slightly from one another, it is important to keep in mind that such differences may be what helps to make the folksongs in the Essen Collection unique. It would be unfortunate to dismiss these differences between the subsections merely on the basis that they are subtle. Instead, it would be good to conduct further studies into how and why the subsections do differ from one another and perhaps uncover how these differences have affected scholarship involving the Essen Collection.

It is important to note just how few of the plethora of studies covering the Essen

Collection deal with its subsections (let alone even reference them). Of the roughly 171 studies that cite the Essen Collection, only seven refer at all to the subsections in the

Essen Collection. All of these seven only mention the fact that they chose to use data provided by a specific subsection. These studies make no mention whatsoever of the other subsections nor do they address why exactly their specific subsection was chosen.

They do not address any of the statistical properties inherent in their subsection nor do they attempt to examine whether their subsection is in any way different from the other subsections of the Essen Collection. Overall, scholars have assumed that the subsections

69 of the Essen Collection have no impact on the structure of the collection and it might even be suggested that many of these scholars probably have no idea why the subsections of the Essen Collection even exist (assuming they even recognize the presence of the subsections at all). Additionally, of the 171 studies citing the Essen Collection, only 22 make reference to the fact that the Essen Collection is a distinctly “German” folksong collection. This means that the majority of the studies that cite the Essen Collection refer to it as a collection of “European” folksongs, a designation that is highly troublesome. Of those that recognize the “German” quality of the Essen Collection, two address the time period from which these folksongs originate while ten discuss the fact that the Essen

Collection includes Germanic folksongs geographically situated outside of the German region. In sum, many of the studies covering the Essen Collection are making dangerous assumptions regarding the structure and content of their primary data materials. While a more specific address of these studies and their interaction with the Essen Collection is provided in Chapter 3, it is important that we acknowledge the shaky foundation upon which many of these scholars are building their studies. By not recognizing the independence of the twelve subsections, the tenuous nature of the “German” quality of the Essen Collection at-large, or by taking for granted that the Essen Collection is somehow representative of 19th-century German folksong, it could easily be the case that conclusions drawn from these studies are muddied by issues of data representation.

70

Chapter 3. Scholarly Attention to the Essen Collection

Since its release in 1995, the Essen Folksong Collection has seen over 200 citations across all types of scholarly publications. Researchers in the fields of music cognition

(Patel, 2010; Margulis, 2007), empirical musicology (Huron 1996), computer science

(Khoo et al., 2013, 2012), ethnomusicology (Eerola, 2004), and music theory

(Ammirante & Russo, 2015) have often made use of the Essen Collection as a primary source of readily-available musical data. Depending on their topic of interest, scholars have elected to use various formats of the Essen Collection with a fairly equal division between use of the EsAC format (Bod, 2002b, 2002d), the **kern format ((Huron, 2002), and some instances of transcribed MIDI formats (Hoos et al., 2001). In all of these cases, the Essen Collection has often been used to answer fairly substantial questions about musical topics. For instance, the Essen Collection has served, for the past 15 years, as the primary convenience sample for music information retrieval research (Lattner, Chacon, &

Grachten, 2015), appearing in several papers pulled from conference proceedings and events revolving around the International Society for Music Information Retrieval.

There are many potential reasons for why the Essen Collection has been so popular as a dataset for music research. Perhaps the best reason for this is simply that the

Essen Collection was one of the first computer-encoded databases of musical material entailing a rather large sample size. As such, it could be argued that many researchers

71 who have elected to make use of the Essen Collection have done so merely out of convenience. However, a cursory examination of the average years of publications that have cited the Essen Collection shows that a bulk of the scholarly work done with the

Essen Collection was published around 2005, ten years after the Essen Collections original release. This finding suggests that it took some time for scholars to find the

Essen Collection and to begin asking whether such a database would be relevant for research purposes. During that time, other databases of music information were made available in various formats outside of both EsAC and **kern. For whatever reason, scholars ultimately chose to focus on the Essen Collection as a primary source for musical information.

Since the early 2000’s, the general trend in research involving the Essen

Collection tends to focus on rather specific musical topics. For the most part, out of the over 170 studies that cite the Essen Collection, nearly all of them have, at their core, a focus on corpus studies or music cognition. That is, even those papers that might have a slanted view towards topics in ethnomusicology or music theory are, at their core, steeped in questions of corpus study research or music cognition. It seems eerily coincidental that both corpus study research and music cognition have exploded in popularity within the past few decades alongside the publication and rise in use of the

Essen Collection itself. Perhaps one might argue that the Essen Collection has become a sort of “poster child” database for music cognition and corpus studies, acting as a prime example of the kinds of data that are most useful for research in these fields. After all, the near-overuse of the Essen Collection throughout music information retrieval research

72 seems to suggest that other databases and collections of musical data are lacking some important facet or detail. Given the ethnic and cultural emphasis of the Collection itself, it is not surprising that cognition research or corpus studies on ethnomusicological and historical musicological topics tend to favor using the Essen Collection as their primary data material. Additionally, it may also be the case that the Western-centric nature of the

Essen Collection might explain why corpus studies on topics in music theory have also tended to use the Essen Collection. In any case and for better or worse, the importance and popularity of the Essen Collection in music research cannot be overstated.

In this chapter, I present the bulk of the research that has been conducted using the Essen Folksong Collection as the primary data material. More specifically, I examine the many different ways that scholars have used the Essen Collection and draw attention to some of the larger trends in research involving the Collection. Along the way, I point out some of the pioneering and pinnacle work that has been done using the Essen

Collection as primary data material and discuss the implications that using such a collection brings with it to this research. The goal of this chapter is to present a sort of meta-analysis of the work that has been done with the Essen Collection and to discuss the merits and consequences that coincide with its use. For the most part, I do not wish to suggest that the research involving the Essen Collection is in any way preferable to research that uses other databases, nor do I wish to suggest that research with the Essen

Collection is somehow flawed. Instead, I merely plan to draw attention to the many different ways that scholars have made use of what easily has to be one of the most widely-used databases of musical information. In the past twenty-five years since the

73

Essen Collection’s original release, few, if any, scholars have attempted to address trends in research involving the Collection. I attempt to examine these trends as a way of better understanding the material that we, as a scholarly community, are dealing with.

Much of the research involving the Essen Collection can be rather easily split into three broad and overlapping categories. These are: research focusing on broad topics of music cognition and empirical musicology, research focusing on computational modeling and music information retrieval, and a much smaller category that primarily focuses on musicological, theoretical, or ethnomusicological questions. Perhaps these three divisions can be more easily understood by describing them with research labels. As such, we might describe these three categories of research as proscriptive experimental research, computational modeling research, and descriptive experimental research. We might define proscriptive experimental research as including all those studies involving the

Essen Collection that do not focus on the Collection itself as the most important facet of their research. Proscriptive experimental research with the Essen Collection tends to focus on higher-order questions in music and often chooses to use the Essen Collection merely because it is a “good” sample or because it is a substantial and convenient sample of musical information. The computational modeling research involving the Essen

Collection often also has higher-order questions in mind yet still chooses the Essen

Collection as a substantial convenience sample. Unlike proscriptive experimental research however, computational modelling research often uses the Essen Collection as material with which to train different types of models using machine-learning and artificial intelligence techniques. Finally, the small category of descriptive experimental

74 research genuinely focuses on what the material in the Essen Collection can tell us about itself. That is, research in descriptive uses of the Essen Collection often focuses on

Germanic culture, European folksong structure, Western monophonic tonality, or some other topic for which the Essen Collection serves as not a convenience sample but an adequate representative of the population in question. This kind of research often is interested in saying something substantial about the content of the Essen Collection itself, rather than being interested in what the Essen Collection can tell us about the world at- large. In this manner, we might describe the dichotomy between proscriptive research and descriptive research involving the Essen Collection as research focusing on the capta quality of the Essen Collection and research involving the data quality of the Essen

Collection (a more detailed investigation of the distinction between capta and data is carried out in Chapter 2).

In the following four sections, I provide a broad meta-analysis of the works surrounding the Essen Collection. Each section deals explicitly with one of the three categories of research as described above. The first section covers topics related to the proscriptive research in music cognition and computational musicology. As its focus, this section deals with the work of David Temperley (Temperley, 2000, 2003, 2004, 2007,

2008, 2009, 2014, 2019), Aarden (Aarden & Huron, 2001; Aarden, 2003), and Toiviainen

& Eerola (Eerola et al., 2001, 2002; Toiviainen & Eerola, 2001a, 2003, 2006b, 2006a,

2001b, 2002, 2005, 2004) as some of the more frequently cited papers that have made use of the Essen Collection. In addition, I briefly touch on some of the less commonly cited papers that deal with this topic. For the most part, this section will deal heavily with work

75 on melodic feature analysis, melody perception and expectation, and various other smaller topics in music cognition. Overall, the number of papers mentioned in this section do not amount to the number covered in section two, but are considerably more than those mentioned in section three. Perhaps this is rather unsurprising given that much of the work in music cognition necessitates the use of human participants. However, we shall see that the Essen Collection is often used as stimulus material in such studies.

Additionally, we will see that the Essen Collection has served as a substantial convenience sample for many topics in corpus study research as a representative of simple Western melodies.

Section two covers the bulk of the research that makes use of the Essen

Collection. This research comes primarily out of the field of computational modeling, where researchers are interested in creating and testing models for human perception of various musical phenomena. The primary scholarly works that will be examined in this section includes the work of Pearce & Wiggins (2004; 2003, 2004; 2017; 2006, 2012),

Temperley (2008, 2009, 2014), Juhász (2009), and Bod (2001, 2002b, 2002a, 2005, 2008,

2002d, 2002c). Due to the substantial amount of research involving computational models and machine learning with the Essen Collection, it is impossible to cover every paper with the same level of detail. As such, this section focuses primarily on those papers that have been more frequently cited in the computational modelling research.

Topics that these papers have covered include melody perception, segmentation, and expectation, folksong classification, and musical complexity. Specifically, the topics of melody segmentation and expectation have received considerable amount of attention is

76 it relates to the Essen Collection. The primary explanation for this is the fact that the

Essen Collection is one of the only encoded databases of musical information that includes phrase segmentations, a topic that is explored more thoroughly in Chapter 2.

The smallest number of studies making use of the Essen Collection involve descriptive topics. While relatively infrequent, some of these studies have become immensely influential across all fields of music research. The heaviest contributions to the descriptive research include the work of David Huron (1995, 1995, 2002, 2013) and various other scholars (Aarden & Huron, 2001; Huron & Royal, 1996; Van Kranenburg et al., 2010; van Kranenburg et al., 2013; Van Kranenburg et al., 2007; Van Kranenburg

& Janssen, 2014; Von Hippel, 2000; Von Hippel & Huron, 2000). Given the nature of descriptive research, it could be argued that some of the studies mentioned in the previous two research categories might also belong to this category. After all, one can make the argument that any conclusions drawn from the use of the Essen Collection really only apply to the music in the Essen Collection itself. However, it is important to note the distinction between the work of scholars who pay close attention to the Essen

Collection and what it represents and works that focus on the convenience of the Essen

Collection.

Finally, the last section of this chapter involves a detailed look at the potential pitfalls and controversies that surround the frequent, and often nonchalant, use of the

Essen Collection. In addition to address the large proportion of studies that merely mention the Essen Collection, I discuss how a decent proportion of the scholarly work involving the Collection tends to make rather dangerous assumptions about the Essen

77

Collection and what the implications of those assumptions might be. I also point out the few scholars who have noted problems with scholarly attention to the Essen Collection and how these problems have manifested themselves in other scholarly works. The point of this section is not to imply that research involving the Essen Collection is somehow incorrect or troublesome. I only wish to point out the dangers in forcing available data to fit specific research topics without detailed attention payed to the origin of the data. In addition, I hope to show that future researchers interested in the Essen Collection should take seriously the construction and implications present in the use of the Collection before electing to include the Collection in their work. The Essen Collection is a wonderful tool for music research. However, we must make sure that we are treating the data in a careful and clear manner.

Proscriptive Studies with the Essen Collection

Many of the papers involving proscriptive research with the Essen Collection typically deal with the topic of melodic or meter expectation. This focus on expectation is often shared by the multitude of papers that involve computational modelling. However, the proscriptive research on expectation often approaches the topic from a distinctly different view. Perhaps the best example of this is Aarden’s dissertation (2003) involving melodic expectation. While not interested in creating models of human perception, Aarden is extremely interested in the ways that various rule-based music theory principles may explain or account for the way that humans perceive melodies. Aarden’s work is split into a meta-analysis on the pioneering work in musical expectation and a couple of different 78 cognitive studies meant to test the claims of those pioneering scholars. Aarden goes through the extensive work of Narmour (1990) and his Implication-Realization model,

Krumhansl, Schellenberg, and Kessler (Krumhansl & Kessler, 1982; Krumhansl &

Schellenberg, 1990) and their work with pitch perception and key-finding algorithms, and

Meyer (Meyer, 1989) on his theoretical principles of expectation. Aarden then goes on to test several aspects of expectation through cognitive experiments, addressing several topics like tonality, closure, duration, intervallic movement, and key-finding methodologies. As a whole, Aarden’s dissertation has easily become a rather important foundational work on expectation from the cognitive and theoretical perspective. In regards to the Essen Collection, Aarden’s work in drawing attention to its structure and in using the Collection as stimulus material is some of the first research with the Essen

Collection that approaches its use with a more cautious and detailed viewpoint. Aarden does not take for granted the work of previous scholars in regards to pitch distributions and key centers but instead focuses on what the Essen Collection might be able to tell us about these topics. However, Aarden is careful to address the fact that the pitch structure in the Essen Collection is not inherently representative of all Western music. Overall,

Aarden’s detailed and careful attention to the Essen Collection is part of the reason why his work has been so well received.

The various works of Temperley (2000, 2003, 2004, 2007, 2008, 2009, 2014,

2019) all have in common the mixed application of cognitive and computational approaches to the Essen Collection. Given Temperley’s frequent use of computational models, it is entirely possible that his work would also fit into the second section of this

79 chapter. However, what makes Temperley’s work unique from much of the other research on modelling is his attention to the content and structure of the Essen Collection and to the comparison of his models to cognitive research. In many of Temperley’s works, he goes out of his way to compare his models with the results of human participants’ experience with the same topic. Additionally, much of Temperley’s research relies heavily upon understanding some of the structural aspects of the Essen Collection like pitch distribution, duration and meter distribution, and intervallic content. Perhaps unlike Aarden, Temperley’s work has often inadvertedly stumbled upon interesting facets of the Essen Collection including specific aspects of pitch distribution (Temperley &

Marvin, 2008), interval/duration distribution and modality information (Temperley,

2008), and even contour and accent information (2014). While the finding of this information is certainly helpful to our understanding of the Essen Collection’s structure, often times this information is not the key findings in Temperley’s research. Instead, through comparison of computational models and cognitive experiments, Temperley often focuses on broader questions of expectancy, segmentation, and meter perception.

This attention to broader musicological questions is not in any way misguided of course.

However, it is important that we recognize what will become a common theme in the use of the Essen Collection that Temperley’s work exemplifies. That is, most, if not all, of our understanding of the content and structure of the Essen Collection comes as a side- note to the main goals of the works that make use of the Collection. Temperley’s work, while perhaps more enlightening to the Essen Collection’s structure than others, is still an example of this issue.

80

Another example of scholarly work that splits the boundary between computational modelling and cognitive studies is that of Toiviainen & Eerola (Eerola et al., 2001, 2002; Toiviainen & Eerola, 2001, 2003, 2006b, 2006a, 2001, 2002, 2005,

2004). This scholarly due often approaches issues related to our understanding of various musical cognitive processes through the combined application and comparison of modelling and cognitive experiments like Temperley. Where Toiviainen & Eerola differ is often in either their modelling approach or in their topic of interest. For example,

Toiviainen & Eerola present work on meter classification, folksong classification, folksong feature extraction, and different accent patterns. Some of the studies that these authors conduct tend to lend themselves much more to modelling practices (as in the case of the classification topics) while others heavily rely upon comparison to cognitive studies (as is often the case with Eerola’s separate work or work with other scholars:

Eerola, 1999, 2016, 2004; Eerola et al., 2006; Eerola & Bregman, 2007; Hannon et al.,

2004; Marcus T. Pearce & Eerola, 2017). Overall, many of Toiviainen & Eerola’s studies do not necessarily uncover exceptionally new information regarding the structure and content of the Essen Collection. Two of their most substantial discoveries include the overabundance of duple meters in the Essen Collection (especially 2/4 and 4/4; 2005) and the importance of specific kinds of accent patterns (2006). Instead, the bulk of Toiviainen

& Eerola’s work tends to provide additional evidence for the structural aspects found in the work of other scholars like Temperley, Aarden, and Huron.

In addition to the large proportion and contribution of the works discussed above, there are several other important works involving the Essen Collection that emphasize

81 cognitive perspectives. Before presenting some overviews of this research it is important to keep in mind some important trends. As with all proscriptive research, the papers below often to not emphasize the importance of the Essen Collection nor do they identify with the assumptions inherent in its structure and origins. Often, these scholars are more interested in their own research topics than in pointing out any specific facet of the Essen

Collection itself. In doing so, some of these scholars make dangerous claims regarding the Essen Collection often in a haphazard mannar. For example, Ammirante & Russo’s

(2015) work with instrumental ranges found that skips tend to occur more frequently in the lower ranges of a melody and used the Essen Collection as a primary sample of

Western vocal melodies. While the overall outcomes of this research are certainly interesting, it is highly dangerous to assume that the Essen Collection is comprised solely of vocal melodies given the tenuous nature of some of the Essen’s subsections (like the

“ballad” subsection which may include a large proportion of instrumental melodies).

Where relevant, I point out these kinds of assumptions as they appear in the literature review.

Some of the cognitive work involving the Essen Collection includes topics related to musical distinctiveness. This is most apparent in Bailes (2010) where listeners were exposed to various musical stimuli (including the Essen Collection) and were asked to rate how familiar they were with the tunes. Bernardes et al. (2016) use human participant’s ratings of how well automatically generated harmonic accompaniments match up with a given melody (often one from the Essen Collection) as a way of testing a new tool for harmony generation. Like the Aarden’s work, some scholars continue to test

82 the claims of theorists, musicologists, and the like using computational methods and corpus studies (with the Essen Collection) as in Brinkman & Huron (2018). Dean &

Pearce (2016) question the relevancy of information-theoretic models for melodic expectancy over rule-based ones in one of the few papers in cognitive studies that addresses the subsection issue of the Essen Collection. In the work of Hannon et al.

(2004), the question of how listeners determine meter based on temporal cues is addressed with the Essen Collection serving as the primary stimulus material.

Unfortunately, like some other studies, the scholars use the Essen Collection here based primarily on the assumption that the Collection is made up of simple melodies. Just because a melody might be derived from a monophonic folksong does not inherently mean that the melody is not complex. Work on sadness and silence has often used or mentioned the Essen Collection in some way. This includes Huron et al. (2010), Yim

(2014), and Margulis (2007) in which both studies mention the importance of the Essen

Collection as a useful database of musical information. Similarly, Schäfer et al.’s

((2015)) work on music and safety involves exposing listeners to uncomfortable metrical patterns based on alterations to melodies in the Essen Collection. One of the more substantial questions involving the Essen Collection’s structure revolves around the issue of phrase boundaries. The work of Jasmin et al. (2018) struggles with this in their paper on the relationship between perceptual strategies and abilities. Using phrase boundaries as correlates for sentence structures in language, the authors test listeners for their ability to rate how well different ending progressions “complete” a music passage. One potential issue in this paper lies in the assumption that phrase information in the Essen Collection

83 is accurate and valid. Additionally, Schaefer et al. (2004) attempt to tackle the issue of phrase boundaries by having listeners point out boundaries in unfamiliar melodies

(folksongs in the Essen Collection). Again, the problems surrounding the accuracy of the phrase boundaries in the Essen Collection continue. Li & Huron (2006) use computational models on the Essen Collection in order to determine probability information in melodic content. Most notably, during the course of their research, these scholars point out the vast disparity between major and minor folksongs in the Essen

Collection (with an overwhelming number of major melodies compared to minor melodies). Some of the cognitive work on the Essen Collection arises through meta- analyses as in the case of Pearce et al. (2010b). In these instances, it is hard to separate the purely cognitive studies from those focused purely on computational modally, a problem that the authors struggle with in this paper.

Computational Modeling Studies

The second and, by far, largest category of studies conducted using the Essen Collection deals with the topics of computational modeling and machine learning. Many of the studies in this category involve the creation, training, and testing of computational models of various aspects of human cognitive processes. Some of the more popular types of models trained and tested on the Essen Collection include ones that model musical perception (through melodic segmentation and expectation), classification (by folksong genre, dance style, and meter estimation), and musical feature analysis. There are many reasons why this category of research involving the Essen Collection is so large. One 84 possible explanation relies on the fact that the Essen Collection has long been a favored database for testing tools related to Music Information Retrieval (MIR) research. Due to the fact that the Essen is the only large database of musical data that includes pitch, rhythm, meter, mode, and phrase information, presented in simple and easy-to-work-with monophonic song structures and existing in multiple different formats, it is easy to see why scholars involved with MIR work would be interested in the Essen Collection.

Additionally, the Essen Collection has also served as the primary database for various

MIREX (Music Information Retrieval Evaluation eXchange) competitions over the past fifteen years, perhaps adding to the popularity of the database in computational modeling research.

Due to the nature of research involving computational modeling and machine learning, it is very common for papers on this topic to cover topics also related to proscriptive and/or descriptive research. For instance, much of Bod’s work involving computational modeling covers topics related to principles of musical grammar (2001,

2002a, 2002b). The work of Pearce & Wiggins tends to involve modeling of musical expectancy through studying musical context (2004, 2006, 2012). Additionally,

Temperley’s work with computational modeling focuses specifically on aspects of musical analysis and perception (2008, 2009, 2014). Some modeling work involves the emphasis on the role of meter (Eck, 2005; Gkiokas et al., 2015.; Lambert et al., 2014), while others focus on the differences between regional musics (Toiviainen & Eerola,

2001, 2005; Khoo et al., 2013; Ren, 2016; van der Weij et al., 2017). However, in all of these instances of interdisciplinary work between modeling and other musicological

85 topics, the focus is rarely on an investigation of the topic itself and almost always on the modeling techniques themselves. While the importance of the musicological topic cannot be dismissed in these studies, it is very common for these studies to focus on presenting new and better ways to model such topics. Often, these studies involve building off of models devised by earlier research as a way of showing continuing growth of modeling accuracy, reliability, and precision. It is for these reasons that we can think of the work with the Essen Collection involving computational modeling and machine learning as proscriptive research of the highest order.

As we will see, studies on computational modeling and machine learning involving the Essen Collection very rarely, if at all, discuss the Essen Collection itself with any detail. Many of these studies simply mention that the Essen Collection was used as the training/testing set and leave it at that. Some studies mention specifically how much of the Essen Collection was used, pointing out the exact number of folksongs included in their training/testing set or (in very rare cases) drawing attention to specific subsections of the Essen Collection. In all of the papers encountered while conducting this meta-analysis, only two scholars in this category of studies discuss the origins or background of the Essen Collection in a meaningful manner (2001, 2002b, 2002a, 2005,

2008, 2002d, 2002c; Khoo et al., 2013, 2012). A few scholars mention specific issues related to the use of the Essen Collection such as Cenkerová et al. (2018), and in one paper critiquing research with digitized collections, the authors mention the overuse of the Essen Collection in computational modeling research (Van Kranenburg & Janssen,

2014). The rest of the papers in this category of research interact with the Essen

86

Collection merely as if it were a necessary but perhaps unimportant tool for their research. This is highly troublesome for a variety of reasons, not the least of which being the tenuous origins of the Essen Collection and the sheer volume of assumptions that are made when a scholar elects to make use of the Collection. While it is certainly understandable that scholars in computational modeling and machine learning research have more than enough to worry about when conducting their research, it can be rather dangerous to make use of the Essen Collection in many of these contexts without pointing out potential pitfalls. As we move into a discussion of the more highly cited authors in this category of research involving the Essen Collection, it is important that we keep in mind the distant relationship between the scholars themselves and their use of the

Essen Collection. As we will see, some rather glaring issues have arisen in computational modeling research with the Essen Collection as a result of this distance.

The most widely known work involving the Essen Collection comes from Pearce

& Wiggins (2003, 2004, 2006, & 2012), Rohrmeier et al. (2011), Conklin et al. (2013;

2006; 2008) , and Bod (2001, 2002a, 2002b, 2002c, 2002d, 2008). Temperley’s work with computational models might also fit into this category as it heavily involves the use of modeling techniques. However, his focus on proscriptive focus on a variety of cognitive and theoretical topics places him more into the previous category of research involving the Essen Collection. In the following paragraphs, I briefly touch on the work of these various scholars and the contributions they have made to research involving computational modeling. The primary purpose for drawing attention to these authors is to discuss the ways that these scholars have dealt with the Essen Collection and what their

87 reliance on the Collection says about their contributions. As the point of focus for this chapter is on uses of the Essen Collection rather than methodologies and research techniques, I will avoid detailed discussion on computational modeling itself and instead focus on how these scholars have trained/tested their models on the Essen Collection.

The work on computational modeling done by Pearce & Wiggins covers a wide variety of topics. The authors frequently make use of the Essen Collection as a means of testing and training their various computational models. For the most part, their purpose for using the collection seems to be based on reasons of convenience. For example,

Pearce & Wiggins feel that the monophonic structure of the Essen’s folksongs, the incorporation of phrase boundaries, and the large overall sample size all make the Essen

Collection an ideal database for testing purposes. Their paper on modeling of melodic prediction (2003) involved the authors taking their own subset of the songs in the Essen

Collection that included 91 Alsatian songs, 119 Yugoslavian songs, 93 Swiss songs, 104

Austrian songs, 213 songs from the “kinder” subsection, and 237 songs from the Chinese collection. Of all their uses of the Essen collection, this subset happened to be used in several other studies after this initial one (Cherla, 2016; Cherla et al., 2013, 2015; Lattner et al., 2018), all of which also happen to be studies involving computational modeling. In general, Pearce & Wiggins are very clear about how and what parts of the Essen

Collection are used. Pearce has separately conducted work involving the Essen Collection

(Pearce, 2005), again taking care to be as specific as possible in mentioning which portions of the Essen Collection were used in his research. Both Pearce and Wiggins have published studies with other scholars, separately and together, again with a strong

88 emphasis on making clear how the Essen Collection was used (Dean & Pearce, 2016;

Hedges & Wiggins, 2016; Morrison et al., 2018; Müllensiefen et al., 2008; Müllensiefen

& Wiggins, 2011; Pearce et al., 2010b, 2010a, 2008; Pearce & Eerola, 2017; van der

Weij et al., 2017; Wiggins et al., 2009).

In general, the total studies published by Pearce and Wiggins, together, separately, and with other authors, constitutes the majority of studies published using the

Essen Collection. The primary commonality between all of these studies is the emphasis on either presentation, testing, or discussion of computational modelling. In all instances where relevant, the authors of each study meticulously detail what parts of the Essen are used and how those parts were used. For the most part, the work of these scholars has done well in accurately detailing uses of the Essen Collection. However, none of these studies discuss exactly why specific portions of the Essen Collection were chosen. For instance, in Pearce & Wiggins (2003), they mention that the “kinder” subsection of the

Essen Collection was used, but do not tell us why that specific subsection was chosen over any of the other subsections. Due to the large amount of random sampling in these studies, it is likely the case that most decisions to choose this or that portion of the Essen other another subsection were simply products of randomization. Of course, while ordinarily this random sampling would not be a problem, each of the subsections in the

Essen Collection are, at least, slightly different from one another and it is not entirely clear exactly what assumptions are being made when one makes the choice to use a specific subsection.

89

Rohrmeier et al. (2011) is another widely cited study with computational modeling involving the Essen Collection. However, like so many of these studies, use of the Essen Collection in this study is rather lax. It is not uncommon for scholars to merely take arbitrary measurements of some musical feature or another in the Essen Collection to use as part of the computational model itself. Similar uses of the Essen Collection in this manner include Eerola & Toiviainen (2002), Sadakata (2006), Mullensiefen &

Wiggins (2011), Elowsson & Friberg (2012), and Flossman (Flossmann, 2012). With the exception of Mullensiefen & Wiggins (2011) none of the above papers discuss where in the Essen Collection measurements necessary for a study were taken. As the Essen

Collection itself presents a diverse range of information, it is important that authors are as clear as possible when discussing how a given metric was determined using the Essen

Collection. Otherwise, these and future studies run the risk of not accounting for unknown variables present in the Essen Collection’s various subsections. This is especially true for situations in which measurements taken from the Essen Collection are used to build databases of “new” or “pseudo-“ music like that of Rohrmeier et al. (2011).

Conklin’s work with various other scholars, unlike those mentioned above, deals with computational modeling involving classification and pattern recognition tasks

(Conklin & Anagnostopolou, 2006; Conklin & Bergeron, 2008; Hillewaere et al., 2012;

Conklin, 2013). These scholars make use of the Essen Collection for two primary purposes: its large size as a sample of European folksong, and its incorporation of melodic phrase boundaries. In general, these scholars have dealt with the Essen

Collection very meticulously, specifically mentioning which parts of the Essen Collection

90 were used in their studies and reasons for why such parts were chosen. Additionally, some of the above studies specifically address the issue of overuse of the Essen

Collection. Hilllewaere et al. (2012) points out that the Music Information Retrieval

(MIR) community has a tendency to make use of the Essen Collection over other datasets for its various competitions and publications. Like other computational modeling studies, none of the above studies address the structure and content of Essen Collection, choosing to focus instead on what the Essen Collection can tell us. Like Pearce & Wiggins (2003),

Conklin & Anagnostopolou (2006) also makes use of a specific subset of the Essen

Collection involving Alsatian, Yugoslavian, and other regional folksongs though it is not clear how much the two papers are related. Unfortunately, one study (Conklin, 2013) makes some rather dangerous assumptions by electing to sample music from six countries sampled in the Essen Collection. Again, due to the issues of its origin, without naming the countries nor presenting the selection process results in some rather troublesome issues.

The last of the widely cited computational modeling studies involving the Essen

Collection comes from the work of Bod and other scholars (Bod 2001; 2002a; 2002b;

2002c; 2002d; Zaane et al., 2003; Schaefer et al., 2004; Bod 2005; 2008). In all of Bod’s solo studies, he goes to great lengths to discuss his sampling process and the EsAC format of the Essen materials. In each paper, Bod talks briefly touches on the origins of the Essen materials before discussing how one can read and interpret the EsAC format. In all of his solo papers, Bod makes use of most, if not all, of the Essen Collecction. As the bulk of his work deals with musical grammar and syntax, it is not surprising the his

91 models require a substantial amount of testing and training materials. Bod seems to be interested in the forces that guide the structure of melodic lines specifically from a cognitive and perceptual approach. However, the bulk of his work involves using computational models to segment melodic lines into musical phrases, using models derived from musical grammar. Of his two studies with multiple authors, only one deals with actual testing involving the Essen Collection (Shaefer et al., 2004). Interestingly, the authors choose to use only the “kinder” subsection of the Essen Collection, mentioning that, as this is supposedly a collection of children’s songs, it should have simplistic phrase structures more suited for their topic of interest. However, given that we know the phrase information in the Essen Collection was determined by Schaffrath’s adult colleagues while encoding the materials, it is dangerous to assume that the folksongs in the “kinder” subsection do indeed have simplistic musical phrasing.

Of the less-citied studies involving the Essen Collection, most do not address the content, origins, or structure of the database with any depth. The few that do tend to focus on one specific aspect of the Collection like Eck (2005) who focuses on the different meter signatures in the Essen Collection (but does not note that many meters in the Essen were derived by Schaffrath and his colleagues), and Morgan et al. (2019) who briefly touch on the existence of the different subsections in the Essen as a way of comparing multiple different models of musical grammar (though any detailed discussion beyond the mere existence of these subsections and their different sizes is sorely lacking). Several studies mention using one or more of the different subsections of the Essen Collection but do not address why those specific subsections were chosen (Juhász & Sipos, 2010; Khoo

92 et al., 2013; Frieler, 2015; Lattner et al., 2015; Müllensiefen et al., 2015; Harrison et al.,

2017; Kaliakatsos-Papakostas, 2019) One study (Spevak et al., 2002) should be noted for the fact that the authors actually discussed with Ewa Dahlig-Turek about the tenuous structure of phrase information in the Essen Collection. Thornton (2011) remarks on the issue of encoder bias in the phrase information of the Essen but does not extensively address it in his model. Honingh & Weyde (2008) use the phrase boundary information in the Essen Collection to determine overall phrase endings, noting that nearly all phrases in the databases begin with the same note as the final note of the previous phrase. In a similar manner as Hillewaere et al. (2012), Lattner et al. (2015) and Cenkerová et al.

(2018) also address the issue of overuse of the Essen Collection in the MIR community, stating that its popularity is the main reason why they also chose to use it. A small number of studies elect to use the Chinese portions of the Essen Collection (Khoo et al.,

2012; Khoo, 2013) Finally, a small portion of studies mention using portions of the Essen

Collection, tending to focus on describing the selection process (such as random sampling or choosing songs based on metrical content) rather than on the Collection itself

(Eerola & Bregman, 2007; Volk, 2012; Eerola, 2016; Pazdera, 2016; Alvarez & Gómez-

Martin, 2019; Arronte-Alvarez & Gomez-Martin, 2019)

The rest of the computational modeling studies involving the Essen Collection come with a fair number of issues. Most of these papers simply state that they are using a number of folksongs in the Essen Collection, not bothering to mention which subsection these folksongs came from, why they were chosen (other than for simplistic reasons like the Essen Collection being a sampling of folksongs, or of European music), or how they

93 were selected (Cherla et al., 2015; Juhász, 2009; A. Lambert et al., 2014; A. J. Lambert et al., 2014; Ronca, 2009; Sadakata et al., 2006; Tanji et al., 2008). In each of these cases, the authors draw attention to their sample size and how their sample was chosen, but the sizes themselves and the methods for sampling are different for each study. The remaining few studies have some rather troubling issues. Ammirante & Russo (2015) claim they use the Essen Collection because it is sampling of vocal melodies. While this may not be completely incorrect, the discussion of the origins of the Essen Collection presented in Chapter 2 suggest that a substantial portion of the folksongs in the Collection may be instrumental melodies. A few studies mention using the Essen Collection, but either do not mention how much of the Collection was used or do not provide any information regarding the sampling process (et al., 2015; 2016; Ren, 2016; McLeod &

Steedman, 2017). One paper cites using the Essen Collection but states that they are using all of the EsAC materials, including both the Germanic portion of the Essen Collection as well as the Chinese and all supplementary material (Chan & Hsiao, 2016). Lastly, Chai &

Vercoe (2001), an immensely influential study on the use of hidden markov models for computational modeling of musical features and in folksong classification makes use of the Germanic and Austrian portions of the Essen Collection. However, as far as I am aware, the authors do not cite Schaffrath or Huron (concerning the “**kern” format) as the primary originators of the Essen Collection.

In sum, the studies making use of computational modeling techniques with the

Essen Collection cover a wide range of topics and deal with the Collection in a variety of ways. For the most part, the most widely cited studies in this category of research make

94 explicitly clear what the Essen Collection supposedly is, how they use it, and why it is important for their research. A large proportion of studies in this topic make use of the

Collection but do not delve deeply into questions of origin or its potential structural issues. Lastly, a small number of papers barely discuss the Collection at all outside of mentioning that they use it to test/train their model. As a side note, a number of studies in this category do not explicitly make use of the Essen Collection, but do mention its existence in some way or another. For the most part, these papers discuss how other scholars have used the Essen Collection and how the results of those scholars’ work can inform the present studies.

Of the three categories of research involving the Essen Collection, those in the computational modeling category have the most issues. While it is certainly nice that scholars are documenting how and why they make use of the Essen Collection, nearly all of the papers in this category assume that the Essen Collection is what others purport it to be. Namely, a database of “German”, “folksongs”. As noted in Chapter 2 this claim is rather contentious and the fact that no papers in this category address the Essen

Collection’s issue of representativeness is rather troublesome. I certainly do not think that these authors have intentionally ignored these issues, but I also believe that these issues can no longer be ignored if research on computational modeling techniques in music is to continue. As a few of the previous scholars have already mentioned, perhaps the best fix for this issue is to address the overuse of the Essen Collection, not just in the MIR community, but in computational modeling research as a whole. I fear that, if this issue is not addressed, the only other course of action is for scholars in this field to make

95 explicitly clear the assumptions regarding sampling that one makes when electing to use the Essen Collection in their own research.

Descriptive Studies

Constituting the smallest portion of studies citing the Essen Collection, research in the descriptive category involve investigating musical questions directly related to the content of the Essen Collection. Unlike proscriptive research which tends to focus on questions of perception and cognition, this category focuses on what kinds of musical phenomenon exist in the Essen Collection and how those phenomena might be related to other kinds of music. The most widely cited studies in the descriptive studies category primarily come from the same author, David Huron. To a lesser extent, some authors appear several times (Shanahan, van Kranenburg, von Hippel, and Yim). Given that there are only a handful of studies in this category, we will briefly touch on Huron’s research before diving into the rest of the materials.

Huron shows up as an author on nine descriptive studies (Huron, 1996; Huron &

Royal, 1996; von Hippel & Huron, 2000; Li & Huron, 2006; Huron et al., 2010;

Shanahan & Huron, 2011; 2012; Schafer, Huron, et al., 2015; Brinkman & Huron, 2018).

Each of these studies covers a different topic of interest including: scale degree 6 in

Western music, the melodic arch in Western folksongs, pitch and sadness, interval size and phrase position, skips at changes of pitch direction, and the sounds of stress and danger. Out of all the authors encountered while pouring through the literature citing the

Essen Collection, no other author provides as much detail regarding the Essen Collection 96 and its many potential issues as Huron does. Across these nine studies, Huron mentions several times that phrase marking information in the Essen Collection is heavily biased, that geographical information in the Collection is rather dubious, that there unequal split between major and minor mode folksongs in the Collection, that the question of

“German” quality in the Essen Collection is difficult to understand, and that the pitch information in the Essen Collection differs slightly between the EsAC and the **kern formats. Given that Huron is the creator of the **kern-encoded version of the Essen

Collection, it is unsurprising that he would both know so much about the Collection and be so straightforward in talking about its potential pitfalls. Huron and his colleague’s use of the Essen Collection changes based on the topic of study, ranging from full-scale corpus research using the entire Collection to small random samples of 50-100 folksongs.

Where relevant, Huron and his colleagues mention specifically which subsection of the

Essen they use for their study and exactly why that subsection was chosen.

However, in all of their openness in dialogue regarding the structure and use of the Essen Collection, there are some issues that Huron and his colleagues do not address.

For instance, when creating samples of the Essen Collection materials, Huron and his colleagues do not control for whether the samples end up being unequally distributed from the different subsections. When dealing with the issue of the “German” quality of the Essen Collection, the authors merely note that, due to geographical location of these folksongs, it is difficult to accurately label them as “German” folksongs, which, while admirable enough, does not adequately address the “German” quality issue as discussed in this dissertation. Admittedly, these are small quibbles with the way that Huron and his

97 colleagues have dealt with the Essen Collection compared to how other scholars have dealt with it, but it is important that we recognize how important it is to make issues regarding the Collection as clear as possible.

In addition to Huron and his colleagues’ work, a few other studies also address descriptive research involving the Essen Collection. Shanahan & Albrecht (2019) deal extensively with the Essen Collection, providing tallies and percentages of intervallic and pitch content. The authors focus specifically on comparisons of different musical databases treatments of motions to scale degree 1 in order to better understand the effect of oral transmission on folk music. Van Kranenburg et al. (2007; 2010) describes the

EsAC format of the Essen Collection in great detail, discussing the known origins of the

Collection through Schaffrath’s work and documenting the work of scholars who have frequently worked with the Essen Collection. Elowsson (2012) and Frieler (2014) both examine some statistical properties of the Essen Collection such as metric position of pitches and the relationship between intervals and durations. Von Hippel (2000) and Yim

(2014) both address specific issues involving the Essen Collection, touching on the topics of pitch proximity and emotional quality. Lastly, Prince et al. (2019), while carrying out a proscriptive perceptual study, find a few interesting properties of the Essen Collection including the fact that the Essen Collection seems to have an overabundance of 2- semitone intervals compared with other musical materials.

While the amount of descriptive research involving the Essen Collection is rather small, it should be noted that, for the most part, all of the studies discussed above are rather famous studies in the fields of cognitive and systematic musicology. Descriptive

98 studies of the Essen Collection are rare, but this work seems to deeply impact work in other fields in ways unlike proscriptive and computational modeling studies.

Additionally, it seems to be the case that these descriptive studies have influenced much of the later work involving corpus studies of musical materials. In fifty-seven studies citing the Essen Collection but not actually conducting any research with it, nearly half cite the research of one or more of the studies mentioned above in addition to the

Collection itself. Given this fact, it seems even more important that we continue to explore the Essen Collection, not only to better understand its content and structure, but also to make sure that we are using the Collection in a way that is conducive to accurate and bias-free research.

Pitfalls and Mentions

Of the 171 studies citing the Essen Collection, fifty-seven (33%) do not actually conduct any research using the Essen Collection. These fifty-seven studies either mention the Essen Collection in passing, refer to other works that directly involve the Essen

Collection, or state that the Essen Collection has been important, in some way or another, to a given researcher’s topic of interest. In the interests of being succinct, I will not go into detail for each and every paper mentioning the Essen Collection, instead, I will point out some of the more interesting aspects of these studies in regard to the Collection itself.

The majority of papers only mentioning the Essen Collection do so through passing reference to previous scholarship. Most of the time, these studies note some important facet of a past study and in the process point out that this past study used the

Essen Collection materials. Some of the papers that merely cite the Essen Collection do

99 so in reference to the fact that it is one of the only databases of its kind. These papers typically introduce new databases or new tools for MIR, modeling, or cognition research.

In these cases, the authors typically reference that their apparatus is similar to or can be used on the Essen Collection. Huron’s work with descriptive aspects of the Essen

Collection is cited heavily in these papers, as is Temperley’s mixture of cognitive and computational modeling research, and Eerola & Toiviainen’s perceptual experiments.

Unfortunately, a small number of these papers either note some issues with the

Essen Collection, or do not adequately (or correctly) handle research involving the Essen

Collection. Van Kranenburg & Janssen (2014, p. 118) note that “virtually all papers in the proceedings of the yearly conference on Music Information Retrieval (ISMIR)”

(parentheticals in the original), suggesting that the Essen materials were merely used for

“testing algorithms” and modeling musical features. They note that it is important for computational modeling research to explore new databases on which to conduct their studies, lest we run into even more issues of data overuse. Other papers make some rather contentious claims regarding the Essen Collection. Bruderer (2008) notes that the phrases in the Essen Collection were made by musicologists and ethnomusicologists. While one can certainly make the argument that Schaffrath and his colleagues were musically- trained, it is not clear that every person involved with the process of marking phrases in the Essen Collection was actually a musicologist or ethnomusicologist. McBride &

Tlusty (2019) claim that the Essen Collection contains some of the only examples of hexatonic scale systems in cultural musics. While not necessarily false, there is little evidence to back up this claim. One study remarks on the primary source material of the

100

Essen Collection, claiming that Wolfram Steinbeck used the Essen Collection in his own dissertation before Schaffrath published the EsAC format (Jan, 2018). This seems to be in keeping with claims made by the CCARH regarding the origins of the Collection but is the only study that cites the Essen Collection which makes this claim.

Both within and outside of these fifty-seven papers, some “pitfalls” regarding the use of the Essen Collection have arisen. In general, scholars make the assumption that that the Essen Collection is made up of “German” “folksongs”. While this is not necessarily false, based on the findings in Chapter 2 it is not entirely clear that this is the case. Because of the fact that 19th-century notions of “German” are different from modern-day German cultures, it is important that we keep in mind that we assume the

Essen Collection to be a collection of German folksongs and that, without being able to discuss the topic with the original archivists, it is difficult to determine whether the

“German” quality is accurate. Similarly, given that some of the primary source materials seem to include evidence for the encoding of actual compositions into the Essen

Collection, it is possible that not all of the songs in the Essen Collection are folksongs derived from the oral tradition. In a few studies, authors note that the songs in the Essen

Collection are all vocal melodies. Again, findings regarding the content of the primary sources for the Essen Collection suggest that this is not the case and that there exists at least some instrumental, or instrumentally-derived, folksong materials.

Quite a number of studies citing the Essen Collection delve into statistical properties of the Collection. Temperley, Eerola, Aarden, and others all discuss, at some point or another, topics of pitch, interval, and duration distribution in the Essen

101

Collection. Unfortunately, while the overall distributions tend to match up between studies, the specific findings rarely do. This can be explained by the fact that these authors are often using different sample sizes of the Essen Collection to make their claims. However, one wonders why claims regarding distribution information in the

Collection make use of different sample sizes. The simplest explanation is that scholars have been slightly confused by what exactly constitutes the Essen Collection. As Ewa

Dahlig-Turek put it, “[the] Essen Collection is an informal term used for the oldest encodings” (Dahlig-Turek, personal communication, January 19, 2020). This statement suggests that the Essen Collection should be understood as containing only the original subsections of Germanic folksongs and not any other materials. That is, folksongs from specific regions rather than the original source materials should be considered as supplementary material to the Essen Collection or as entirely separate collections. If scholars have misunderstood what parts of the Essen Collection to include in the investigation of statistical properties, then it is no surprise that measurements might slightly differ between studies.

More than any other potential problem with scholarly use of the Essen Collection lies in its overuse. As previous scholars encountered in this chapter have already mentioned, certain musical research communities tend to use the Essen Collection entirely too much. This is especially true of the computational modeling and MIR community which, outside of audio signal corpora used the Essen Collection almost exclusively from 2005-2015. While it does seem to be the case that scholars are moving away from the Essen Collection in recent years, it does not change the fact that the Essen

102

Collection is a popular database of musical information. The primary issue with data overuse of this kind is in the perpetuation of unknown biases and the lack of generality. If ten papers are written on one topic involving, for example, Western melodies and eight of those papers use the Essen Collection, then it is rather presumptuous to say that the findings of that research applies to all Western melodies. Rather, it is more likely that such research only applies to the sample of 19th-century Germanic folksongs present in the Essen Collection. If we are to make and test claims about music that go beyond the representative sample of the Essen Collection, then we must move away from the Essen

Collection as the only data source for this research.

The research pitfalls with the Essen Collection discussed above are not necessarily meant to dissuade scholars from using the Essen Collection in future research.

Instead, I would encourage scholars to continue to use the Collection materials as necessary and where relevant to their own research. The musical information present in the Essen Collection is a rich source of material that can tell us much about musical structure. However, it is important that we respect the origins and structure of the

Collection itself and that we handle the Collection materials with care. In the future, I would recommend that scholars make sure they understand what kinds of musical information they are dealing with when they elect to make use of the Collection. In addition, I stress that scholars recognize that some aspects of the Essen Collection are rather troublesome and should be brought up when conducting any relevant research involving the Collection.

103

Chapter 4: Geographical Distributions of Statistical Properties

Many other scholars have examined and recorded various statistical properties in the

Essen Collection. For the most part, these scholars have used these properties to inform other aspects of their study. For example, Aarden (2003) uses the scale degree distributions in the Essen Collection to address issues in cognitive impressions of key.

Huron et al. (2010) uses pitch distributions in the Essen Collection to inform the creation of sample melodies for a cognitive experiment. Von Hippel (2000) uses pitch and interval distribution information from the Collection to determine tessitura and mobility constraints on small pitch movements. What these studies all have in common is a specific attention towards how distribution information in the Essen Collection can help to better understand other research topics. However, as was the case in Chapter 2 and the various differences between distribution information in the twelve subsections, geographical information might also play a role in how distribution information is oriented in the Essen Collection.

In this chapter, I provide some analyses on the relationship between pitch, interval, and duration distribution and the geographical location of folksongs in the Essen

Collection. There is an abundance of geographical information in the folksongs of the

Collection and yet this information has rarely ever been used. While some studies are interested in geographical information as it relates to cultural and national distinctions

104

(von Hippel & Huron, 2000; Chai & Vercoe, 2001; Pearce & Wiggins, 2004; Conklin,

2013; Cherla et al., 2015), none of these studies make use of the coordinate information in the Essen Collection. That is, while information related to broader geographical distances like national boundaries might be of interest to scholars, overall, the smaller scale distances between different parts of the German region seem to be less popular. The coordinate information in the Essen Collection provides a substantial amount of information related to where, in the German region, a folksong was found. It is dangerous to assume that this geographical information is indicative of the actual place of origin for a given folksong, but the fact that the original archivists felt it necessary to mention where they came upon a given folksong provides some insight into that region’s music.

The following analyses provide some interesting insights into how the three statistical properties of pitch, interval, and duration are dispersed throughout the German region. In doing so, I hope to draw attention to the role that geographical location plays in the content of the Essen Collection while providing some insight into observations that have not been encountered before in the scholarly community surrounding the Collection.

Before continuing into the analyses, it is important to make a few notes regarding the geographical information. Please keep in mind that the coordinate information provided in the Essen Collection is only available in the **kern format and was put together by Bret Aarden (Aarden & Huron, 2001). Schaffrath made a point to encode, where possible, information related to the specific region or town from which a folksong was purported to originate, but did not transcribe that information into geographical coordinates. I do not wish to suggest that Aarden’s coordinate information is incorrect.

105

However, I would be remiss not to mention that, as with all data transcription, there is the possibility for loss of information in the form of clerical errors. Additionally, Aarden came upon an interest problem with Schaffrath’s geographical information. Namely, that only 1,928 of the 5,370 Germanic folksongs of the Essen Collection actually contain accurate coordinate information to the city or village level. All other folksongs in the

Essen Collection either do not have coordinate information, have coordinate information that was guessed by Aarden, or all come from the same point, a location matching nearly perfectly with the city of Essen, Germany. Of course, it is possible that the original archivists found a plethora of songs in Essen. However, it is much more likely that

Schaffrath’s location in Essen while putting together the Collection may have resulted in the large number of songs being attributed to this location. For the purposes of the following analyses, folksongs without accurate coordinate information were eliminated.

For the following analyses involving scale degree distributions, only major mode songs were included, and only diatonic scale degrees were tallied. All simple intervals were tallied for the songs included in the analysis and compound intervals were converted to simple equivalents. Finally, different kinds of durations were recorded by tallying instances of specific rhythmic values. Only a select number of different rhythmic values were tallied. These included quarter notes, whole notes, half notes, breves, eighth notes, sixteenth notes, and eighth-note triplets. These durations were chosen primarily due to the fact that they appear, to some degree, in each of the Collection’s subsections.

Other rhythmic values either did not appear in certain subsections of the Collection or

106 appeared too infrequently to record. Once the three different features were tallied and recorded, they were then mapped according to raw counts and geographical location.

In the figures that follow, each presents information regarding two different musical features. The colored portion of the figure represents the density of a given figure within a specific geographical location in the German region. Each bin represents the sum total counts of all instances of a feature within folksongs found in each coordinate inside of the bin’s area. That is, multiple different folksongs coming from different places that are all relatively close together will contribute their features to the same bin. The bins do not provide exact counts of a given feature but instead represent how the sum instances of a feature in a specific area compare to the sum instances in other parts of the German region. The point of including this binned heatmap is to show where different musical features tend to be found across the German region as represented by folksongs in the

Essen Collection.

In addition to the binned heatmap are individuals points representing a different feature. Each point represents a specific geographical coordinate and the size of the point corresponds with the sum total counts of the specific feature. Folksongs in the Essen

Collection can occasionally share the same exact geographical coordinate (as was the case for the numerous folksongs attributed to the city of Essen) and so there is a possibility that a given point may include sum instances of a feature pulled from multiple folksongs. If a folksong did not include any counts of a given feature, then that folksong was typically not included in the graph (with rare exceptions) in order to increase readability. It should be noted that all points on the figures below are meant to be

107 transparent. Thus, regions that include darker shades of black represent areas with folksongs from multiple different geographical coordinates.

The graphs below were not meant to relate to any post hoc-driven or ad hoc- driven empirical testing. These graphs are merely meant to be an exploratory investigation into some of the more interesting and rarely explored properties of the

Essen Collection. Any conclusions drawn from these graphs should be strongly scrutinized and are only meant to drive further interest in exploring the structure of

Germanic folksongs. Definitive claims regarding the structure of Germanic folksongs, or even of the Essen Collection itself pertaining to geographical locations still need to be tested. However, hopefully these figures will spur further research questions regarding the relationship between musical features and the places where such features can be found.

108

Figure 8: Bins represent instances of perfect fifth intervals across folksongs in the Essen

Collection where lower instances are more green and higher instances are more red.

Transparent black points represent tallies of quarter notes taken from individual folksongs and appear at the point indicated by their latitudinal and longitudinal coordinates. Larger circles represent higher instances of quarter notes while darker shades represent multiple different coordinates in a similar location.

109

Let’s take Figure 8 above as an example perhaps no less interesting than the ones following below. Notice that there are both bins and points outside of the borders of modern-day Germany appearing in parts of France, Switzerland, Austria, Poland,

Denmark, and the Low Countries. The songs included in this sample come only from the original, Germanic portion of the Essen Collection but, as we can see in this figure, obviously there are songs included in the original Collection that appear in other regions.

This topic has been covered previously (see Chapter 2), but it is no less interesting to see this fact firsthand in these figures. Additionally, notice that there are many parts of the

German region for which we have no data. Northern Germany in particular is rather sparse, as is the Southern Alpine region. This indicates that the geographical representation of the Essen Collection is still lacking coverage and that scholars should be careful to note that the Essen Collection is more representative of Central Germany than it is of other portions of the country.

As can be seen in the above figure, there appears to be relatively little difference in counts of perfect fifths overall across the entire German region represented by folksongs in the Essen Collection. The overabundance of green throughout this graph means that the spread of perfect fifth intervals is relatively even, indicating that there is not any specific region that contains more or less of these intervals than any other region.

Conversely, instances of quarter notes are diverse. In particular, there seems to be an overabundance of folksongs including quarter-notes around the Frankfurt region but songs with higher numbers of quarter notes appear in other regions like east of

Amsterdam, Dresden, Berlin, and Stuttgart. Notice also that the instances of quarter notes

110 tend to be rather small in regions outside of urban centers. However, the darker shades around urban centers, indicating higher numbers of folksongs in those regions, may negatively affect our impressions of the data. In order to address this problem, figures with standardized counts rather than raw counts have also been included in this chapter.

Figure 9: A reproduction of Figure 8 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of quarter notes from the overall mean of quarter notes across all folksongs. 111

Figure 9 above provides a slightly different perspective on the information from

Figure 8. In this figure, all quarter-note data (all point-data for the following figures) has been standardized based on distance from the overall average instances of quarter note per folksong. In this graph, larger circles represent areas where the counts of quarter notes are vastly different from the average quarter notes per folksong. The size of the circle does not indicate whether the folksong contains more or less than the average, only that it is different from the average. Notice that the area around Berlin and east of

Amsterdam both have rather large circles just as they did in the previous graph. This suggests, as expected, that the remarkably high number of quarter notes in these places is different from the average instances of quarter notes. Additionally, notice that the circles around Frankfurt are all rather small. This indicates that, while there may be many folksongs in these regions contributing to darker circles as shown in Figure 8, these folksongs typically contain close to the average number of quarter notes per song. Rural areas also tend to contain average-sized circles as well. Regarding of quarter note information, it seems to be the case that the largest number of quarter notes per folksong in the specific regions noted above also have the largest deviation from the mean per folksong. This suggest that, while there are some folksongs with an abnormally large number of quarter notes, there does not seem to be many examples of folksongs with a lower-than-average number of quarter-notes (at least out of those folksongs that actually contain quarter notes).

One major reason for including binned heatmap and point information on the same graph for these figures is so that we can compare counts of the two features in

112 question. Regarding quarter notes and perfect fifths, notice that overall counts of both features tend to hover around the average across all folksongs. There are more exceptions to this phenomenon in the quarter-note data than in the perfect fifth data, but overall these two features seem, on the surface, to be relatively normal. It does seem to be the case that quarter-note and perfect fifth interval data are relatively similar in that both features tend to appear in relatively average amounts in the same places across the German region.

Interestingly, the few places where quarter note data is well above average seem to have a relatively normal number of perfect fifth intervals. Given that tallies of perfect fifths in folksongs seem to be spread evenly across the German region, the comparison between perfect fifths and quarter-note data provides an interesting perspective on the structure of

Germanic folksongs.

113

Figure 10: Presence of scale degree 1, represented by a binned heatmap, and quarter-note tallies, represented by shaded circles.

In Figure 10 above, we are now comparing tallies of scale degree 1 to tallies of quarter notes across the German region. Nothing has changed regarding the quarter note information but notice that the heatmap of scale degree 1 provides more variation in shading as opposed to the perfect fifth data in the previous figures. Specifically, a few

114 key places in particular stand out as containing a large number of instances of scale degree 1. These areas include northwest of Bremen, east of Rostock, west of Dortmund, and most of all, in Leipzig. The area around Leipzig in particular stands out as having an abnormally higher presence of scale degree 1 than in other German regions. There also appears to be a slightly higher presence of scale degree 1 overall in the Northern region than in the Southern region, though accurate testing would need to be done in order to confidently assert this claim.

115

Figure 11: A reproduction of Figure 10 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of quarter notes from the overall mean of quarter notes across all folksongs.

116

Figure 11 above provides a bit more insight into the relationship between quarter notes and presence of scale degree 1 across the German region. As in the previous graphs including quarter-note information, only a few places contain an above average number of instances. Interestingly, the area around Leipzig, which contains a higher number of instances of scale degree 1 seems to have a relatively normal amount of quarter-note information, leaning only slightly above average. In general, none of the regions with above average quarter-note information also contain higher tallies of scale degree 1, other than Leipzig. Instead, some of the regions that are a slightly darker shade of red tend all to have relatively average instances of quarter notes. In sum, Figures 10 and 11 above show some interesting properties related to regional differences of feature information.

The most obvious property between these two figures is the strange over-abundance of scale degree 1 around Leipzig and the seeming lack of relationship between regions that deviate from the average in numbers of quarter notes and the overall distribution of scale degree 1 across the German region.

117

Figure 12: Presence of scale degree 1, represented by a binned heatmap, and eighth-note tallies, represented by shaded circles.

Figure 12 above provides a slightly different perspective on the relationship between scale degree 1 and durational values. As can be seen in the graph, there is a relatively large number of eighth notes across the German region compared to number of

118 quarter notes. However, not many places stand out as having more eighth notes than other places. The regions with the largest instances of eighth notes seem to be west of

Amsterdam and along the southern border with Austria. In general, higher concentrations of eighth notes seem to lie along the western side of the country. Comparatively, the regions with the higher tallies of quarter notes fell, for the most part, on the Eastern border with Poland and around the Frankfurt area. It seems to be the case that, regardless of durational value, there seems to be a rather large number of folksongs coming from the

Frankfurt region overall. In order to better understand the deviations in eighth notes across the German region, let us move forward.

119

Figure 13: A reproduction of Figure 12 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of eighth notes from the overall mean of eighth notes across all folksongs.

120

Figure 13 above provides some interesting insight onto the overall presence of eighth notes. It seems to be the case that, across the German region, there are relatively few places that have an abnormally higher-than-average number of eighth notes. In fact, the only region worth strongly noting is along the northwestern border with the

Netherlands. Recall that this specific region was also noteworthy for its abnormally large number of quarter notes as well. It could easily be the case that there is one or two folksongs from this region that are especially long, thus allowing for a higher number of both durational values compared to the overall mean. It is worth noting that this region lies within the German county of Bentheim, a district in Lower Saxony. Interestingly, this region is rather rural compared to other parts of Germany and has strong ties to Dutch culture. While there is no obvious reason why this region would contain such an overabundance of these durational values in its folksongs, scholars have argued that the people and culture surrounding this region is distinctly different from the Western Dutch and Eastern Germanic peoples (Harger & Lemmen, 2013).

121

Figure 14: Presence of scale degree 1, represented by a binned heatmap, and sixteenth- note tallies, represented by shaded circles.

Figure 14 above provides the last comparison between instances of scale degree 1 and a specific durational value across the German region. Notice first that, while the overall counts for sixteenth notes is considerably smaller than that of eighth- and quarter notes, there are several more regions that feature as prominent regions for the presence of 122 sixteenth-note durations. In particular, the Lubeck region seems to have the highest concentration of sixteenth notes, followed closely by the region of Mannheim to the south of Frankfurt, and the region of Augsburg. There are also some places outside of the

German region that contain higher concentrations of sixteenth notes. Most prominent of these regions is east of Bern within the Swiss region, east of Amsterdam within the Dutch region, and around Wrocław in Poland. As with both the eighth note and quarter note figures, there does not seem to be any strong relationship between regions with higher concentrations of scale degree 1 and regions with higher concentrations of sixteenth notes.

123

Figure 15: A reproduction of Figure 14 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of sixteenth notes from the overall mean of sixteenth notes across all folksongs.

In the case where the sixteenth note data has been standardized across the overall mean, as in Figure 15 above, there seems to be relatively little change between the two

124 figures. As might be expected, those regions that contain higher instances of sixteenth notes also feature prominently as regions with higher-than-average instances of sixteenth notes. Again, notice the lack of any obvious relationship between instances of scale degree 1 and presence of sixteenth note durations.

Across three common durations in the Essen Collection, quarter notes, eighth notes, and sixteenth notes, there does not appear to be any strong relationship between regions that feature higher concentrations of these durational values and regions with higher concentrations of scale degree 1. There are certainly many theories that can be drawn from this information, but is important that such theories be rigorously tested before any substantial claims are made. In the meantime, we can only guess as to why there does not appear to be any strong relationship between common durations and the tonic scale degree. Given the importance of the scale degree 1 in Western music, one might have imagined that areas with higher concentrations of the common durations would also have had higher, or at least moderately high, concentrations of scale degree 1.

Instead, it seems as though there is something interesting happening with folksongs around the Leipzig region and that, for the most part, all other regions have a relatively similar concentration of scale degree 1.

Regarding regions like Leipzig, Bentheim, Frankfurt, and others that contained higher-than-average concentrations of various features, one possible explanation for the deviations is that there are a small number of folksongs in those regions that are acting as outliers. However, it is important that we distinguish the criteria for outlier before dismissing the deviations across these regions. With the exception of the more urban 125 areas like Frankfurt and Berlin, many of the regions represented in the Essen Collection only contain a small number of folksongs. As an example, if one out of three folksongs in a region has an abnormally large number of a given feature, than the higher concentration of a feature in the region in question is solely the product of the one “outlying” folksong.

The problem with designating that one folksong as an “outlier” though is the lack of data from the region. Without more data and more rigorous testing, it is difficult to say with confidence that it is the “outlier” itself that does not represent the feature-content of the region of interest. It could just as easily be the case that the other folksong(s) from that region, containing a relatively normal amount of the given feature, is instead not representative of music from that region (thus acting as the true “outlier”). Therefore, we should be careful when noticing regions that deviate in their distributions of a given musical feature. For now, it is perhaps best to simply note that these deviations exist and to pay attention to both how these deviations relate to one another, within and across different features, and to where these deviations appear across the German region.

126

Figure 16: Presence of scale degree 5, represented by a binned heatmap, and quarter note tallies, represented by shaded circles.

Moving slightly in a different direction, Figure 16 above provides information regarding instances of scale degree 5 (represented by a binned heatmap) and quarter notes

(shaded circles). Notice that, unlike scale degree 1, scale degree 5 seems to appear in differing amounts across the German region. While there is not one place with an

127 abnormally high concentration of instances of scale degree 5, there are plenty of places with a higher-than-average concentration. These regions include the areas around

Brunswick, Essen, and Nuremberg. Outside of the German region, south of Wrocław and north of Bern both seem to contain more presence of scale degree 5. On the surface, there does not appear to be anything linking together these separate regions. We might notice that, with the exception of Brunswick, there seems to be an overall lower concentration of scale degree 5.

128

Figure 17: A reproduction of Figure 16 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of quarter notes from the overall mean of quarter notes across all folksongs.

129

After standardizing the distributional data, there again does not appear to be any significant relationship between scale degree and durational value. There are plenty of regions that have a higher concentration of scale degree 5, but none of those regions also contain a higher-than-average number of quarter notes. It is worth noting here that, according to previous scholars (Aarden, 2003), scale degree 5 appears most frequently across all folksongs in the Essen Collection. It is interesting that the most prevalent scale degree also tends to congregate in specific places across the German region, unlike scale degree 1. The geographical information encountered thus far suggests that relationships between the most common durational values and the most common scale degrees, at least in terms of location, are not explicitly strong.

130

Figure 18: Presence of scale degree 5, represented by a binned heatmap, and eighth-note tallies, represented by shaded circles.

As one last examination of the relationship between common durations and common scale degrees, Figure 18 above provides information on scale degree 5 and eighth notes. Again, there does not appear to be any strong relationship between location of concentration between these two features. Standardizing the distribution data does not 131 provide any additional insight into this phenomenon. While this issue is certainly intriguing, perhaps it is best for now to leave this issue behind lest we arrive at unreliable conclusions.

Figure 19: Presence of scale degree 6, represented by a binned heatmap, and eighth note tallies, represented by shaded circles.

132

The distribution of scale degree 6 across the German region, provided in Figure

19 above, shows a similar story to that of scale degree 1. Once again, there does not appear to be any one specific region where there is an overabundance of the scale degree.

However, unlike scale degree 1 where Leipzig acted as the exception to this observation, scale degree 6 seems evenly distributed across the whole German region without exception. It should be noted that, of all the diatonic scale degrees found in the Essen

Collection, scale degree 6 appears rather infrequently overall. The figure above provides evidence for the notion that even more uncommon scale degrees in the Essen Collection still tend to occur in average amounts across the entire German region. Once again, standardizing the data based on deviations from the overall average does not provide any additional insight into the relationship between the two features presented in Figure 19.

133

Figure 20: Instances of perfect fourth intervals, represented by a binned heatmap, and tallies of scale degree 5, represented by shaded circles.

Figure 20 above provides an example of the relationship between two pitch- related features: the perfect fourth interval, represented by a binned heatmap, and scale degree 5, this time represented by shaded circles. It is plausible that there might be some relationship between the dominant scale degree and the interval that most prefers the 134 dominant as its antecedent pitch. However, notice that that there are relatively few places where perfect fourth intervals appear in higher concentrations. Similarly, notice also that, just as Figure 18 shows, scale degree 5 also tends to be relatively average in tallies across the German region. The exceptions to this observation are that the region around Leipzig seems to have a higher concentration of perfect fourth intervals than elsewhere and that many of the folksongs in the Frankfurt region seem to contain instances of scale degree 5.

135

Figure 21: A reproduction of Figure 20 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of scale degree 5 from the overall mean instances of scale degree 5 across all folksongs.

Looking at Figure 21 above, we see that the high concentration of scale degree 5 in the Frankfurt area seems to be a byproduct of the large number of folksongs in that 136 region. Only a few places stick out as areas with higher-than-average instances of scale degree 5. These include: the Lubeck region, the southern region of Frankfurt, and, to a lesser extent, the region around Dresden. There does not appear, on the surface, to be any strong relationship between regional concentrations of perfect fourth intervals and those of instances of scale degree 5. One possible explanation for this phenomenon might be that, overall, the instances of perfect fourths and scale degree 5 tend to be rather similar across the German region and that the few places where concentrations of these features is slightly higher are simply outliers. If this is the case, then it might be more accurate to think of the relationship between perfect fourths and instances of scale degree 5 across the German region as being fairly normal. Of course, it is important to keep in mind that rigorous testing should be conducted before adhering wholly to this one possible conjecture.

137

Figure 22: Instances of perfect fifth intervals, represented by a binned heatmap, and tallies of scale degree 5, represented by shaded circles.

In order to see whether a similar situation occurs between instances of scale degree 5 and perfect fifth intervals, Figure 22 above was created. Just as in the case with perfect fourths, it might be fine to assume that perfect fifth intervals and instances of

138 scale degree 5 might be related to one another as the dominant scale degree typically features prominently as the consequent pitch of the perfect fifth interval. As can be seen in Figure 22, there does not appear to be any relationship between concentrations of these two features. In fact, standardizing based on deviations from the overall average tells the same story. Even more so than perfect fourths, perfect fifths and instances of scale degree

5 tend to appear in relatively average concentrations across the German region. A few places might possibly be understood as having higher concentrations of one of the two features, but overall the two features seem to be relatively evenly distributed.

139

Figure 23: Instances of scale degree 1, represented by a binned heatmap, and tallies of major second intervals, represented by shaded circles.

As one last example, let us observe the relationship between instances of scale degree 1, represented by a binned heatmap, and major second intervals, represented by shaded circles, in Figure 23 above. Perhaps most interesting in this graph is the fact that the region around Leipzig, which contains a higher concentration of instances of scale

140 degree 1, has an abnormally low number of major second intervals, especially compared to regions like Dresden, south of Frankfurt, and Lubeck. Other regions with more instances of scale degree 1 like west of Berlin seem to also have a higher number of major second intervals. Perhaps there is something interesting about the feature content of folksongs from Leipzig that is contributing to this phenomenon. Instances of scale degree 1 do not seem to be as evenly distributed across the German region as other scale degrees are, but areas with higher concentrations of this scale degree do not always align with areas containing more major seconds.

141

Figure 24: A reproduction of Figure 23 in which the data has been standardized based on standard deviation from the overall means. Larger circles represent areas where there is a significant difference in instances of scale degree 1 from the overall mean instances of scale degree 1 across all folksongs.

142

After standardizing the data in Figure 24 above the strange relationship between concentrations of the two features does not change. Overall, while the Leipzig region has higher instances of scale degree 1, it seems to hover around the average number of major seconds. Conversely, regions like Dresden and Lubeck have average instances of scale degree 1 with above-average instances of major seconds. Again, regions like west of

Berlin show an even more interesting perspective as being one of the only regions with above-average instances of scale degree 1 and instances of major seconds.

In sum, we might note some general trends in the observations provided by the figures encountered in this chapter. Regarding scale degrees, it seems to be the case that scale degrees 5 and 6 are distributed across the German region relatively evenly with only a few regions acting as potential “hotspots” for higher concentrations of these features. In general, all intervals encountered in these observations also tended to be evenly distributed with only a few exceptions. Interestingly, the rhythmic durations and scale degree 1 exhibited the most deviation across the German region with several different places containing higher-than-average concentrations of these features. More broadly, these figures also show that the Essen Collection is severely lacking in representation of the North German and Alpine regions and that there is a substantial amount of representation from regions outside of the modern-day German borders. If I may be allowed to make one potentially dangerous conjecture regarding these observations, it is that rhythmic values, and potentially meter, seems to represent the most distinct variations in folksongs across the German region. One explanation for this might be that, while Germanic folksongs tend to relate to one another in the realms of pitch and interval

143 content, rhythmic content seems to play the biggest role in differentiating between regional musics.

The graphs provided above provide only a brief glimpse into the various relationships between sonic-musical features in the Essen Collection and geographical location. There are still many more features to examine. It is worth noting that exceptionally large intervals (those a minor sixth and above) appear in drastically low numbers and therefore were not analyzed in this Chapter. Likewise, it did not seem that there was enough data for scale degree 7 and durations other than those encountered in this chapter to make any meaningful observations. However, further studies involving features not encountered in this chapter are encouraged as each observation provides additional insight into the structure of Germanic folksongs as represented in the Essen

Collection.

144

Conclusion

Research involving the intersection of musical feature analysis and geographical location has just recently begun to unfold. This dissertation looked at the ways that scholars have dealt with regional musics around Germany through an intense investigation of the Essen Folksong Collection. In the years since the Essen Collection’s original conception, more and more databases for European folk music have been appearing. Databases like the Meerten’s Tune Collection are beginning to gain in popularity to rival that of the Essen Collection. However, as a long-standing tool for music research, the Essen Collection occupies a certain space in the folksong research community that current archivists and encoders hope to enter. After all, the Essen

Collection is not without its own issues. Questions regarding the original conception of the Essen Collection still plague its history, with overuse and potential misunderstandings of representation adding to the Collection’s controversy. The Essen Collection is certainly not a perfect database for musical information, but it has provided many scholars the means with which to carry out a plethora of academic studies. As we reach the end of this dissertation, it is important that we keep in mind some of the key concepts surrounding the Essen Collection.

As we saw in Chapter 2, it is not entirely clear just how many of the purported primary sources were used by Schaffrath as materials for the Essen Collection. There is 145 strong evidence to suggest that some of the subsections (like “erk” and “fink”) were encoded from primary sources that have been made known to the scholarly community.

However, other subsections (like the “ballad” subsection) each hold their own issues concerning how closely Schaffrath worked with their supposed primary sources. It is also debatable just how accurately the folksongs of the Essen Collection represent 19th- century Germanic folksong. As is the case with some subsections (like the “altdeu” subsections), there is a large potential for some folksongs in the Essen Collection to originate from later periods than the nineteenth century. Additionally, it is difficult to say with confidence that the original archivists of the primary sources that make up the Essen

Collection knew exactly from what time period their materials came. As we saw in

Chapter 4 and in the investigation of the Essen’s primary sources, there exist more than a few folksongs in the Essen Collection that come from regions far removed from heavy

German influence. In cases like this, taking into consideration the historical situation of the Germanic cultures of the nineteenth century, it is difficult to define what constitutes a

“Germanic” folksong. Finally, the diversity of genre representation combined with the potential for more Classically-influenced materials present in the Essen Collection suggest that there is a distinct possibility that the Essen Collection is not completely representative of “folksong” materials.

The fluidity and cloudiness in what exactly the Essen Collection’s materials represent is something that scholars should keep in mind when working with the

Collection. As we saw in Chapter 3, well over a hundred studies have been conducted using the Essen Collection as their preferred testing/training material and only a scarce

146 few attempt to address the Essen Collection’s representativeness, origins, or structure.

We would do well to be careful in assuming that the Essen Collection is a perfect representation of 19th-century Germanic folksongs and to recognize the potential pitfalls and biases inherent when choosing to adopt this assumption. Communities like those in music information retrieval should be careful with using the Essen Collection too frequently for the same or similar research topics. The potential for lack of generality and type I and II errors are increased when the only material used across all studies on a topic comes from the same exact database. Overall, scholars involved in all areas of musical research should keep in mind that the Essen Collection is a collection of a few different sources on folksong materials, that each subsection of the Essen Collection is pulled either from a different author entirely, or from a different archival source. Few, if any, scholars attempted to control for confounds that may have come up as a result of unequal sampling of the Essen Collection’s twelve different subsections. Of those scholars who elected to create a random sampling of the Essen Collection, none checked to see whether their sample might have over- or under-represented one or more of the Essen’s subsections.

This dissertation is not meant to dissuade scholars from working with the Essen

Collection. Instead, as mentioned in Chapter 1 and further explored in Chapter 4, we should recognize that collections of folksong materials like the Essen Collection still have much to offer in terms of furthering our understanding of Germanic folk music. When archivists, collectors, encoders, and the like put together and make publically available these large databases of musical material, we should be thankful to have the opportunity

147 to work with new materials and uncover new and exciting musical phenomena. One of the primary goals of this dissertation is to show how careful future scholars should be when choosing to work with folksong materials. Archivists and collectors should be especially careful to be a clear as possible when organizing their materials, to be transparent in where, how, and why the materials were collected, and to make sure that their work is followed up on by other experts in the field. Likewise, encoders and transcribers should make sure to record as much of their process as possible so that we might know exactly how they arrived at their materials and what makes their materials different from, or similar to, the originals. Finally, scholars interested in conducting studies with the collections and databases created and organized by others should be cautious that the materials they have chosen are the best fit for their current research project. Researchers should make sure that they understand the materials they are working with and that they make clear to the community the kinds of assumptions and biases that they necessarily encounter when electing to make use of any database of musical material. If we can all make sure to work together in careful handling of this precious material then we all can benefit from the rewards of scholarly pursuits free from the gravity of potential misuse.

148

Bibliography

Aarden, B., & Huron, D. (2001). Mapping european folksong: Feographical localization

of musical features. Computing in Musicology: A Directory of Research, 12, 169–

183.

Aarden, B. J. (2003). Dynamic melodic expectancy [PhD Thesis]. The Ohio State

University.

Alvarez, A. A., & Gómez-Martin, F. (2019). Distributed Vector Representations of

Folksong Motifs. International Conference on Mathematics and Computation in

Music, 325–332.

Ammirante, P., & Russo, F. A. (2015). Low-skip bias: The distribution of skips across

the pitch ranges of vocal and instrumental melodies is vocally constrained. Music

Perception: An Interdisciplinary Journal, 32(4), 355–363.

Arronte-Alvarez, A., & Gomez-Martin, F. (2019). An Attentional Neural Network

Architecture for Folk Song Classification. ArXiv Preprint ArXiv:1904.11074.

Bailes, F. (2010). Dynamic melody recognition: Distinctiveness and the role of musical

expertise. Memory & Cognition, 38(5), 641–650.

Bausinger, H. (1973). Schlager und Volkslied (“Hit Song and Folk Song”). Handbuch

Des Volksliedes, 1(Die Gattungen des Volksliedes), 679–690.

149

Beckerman, M. (1986). In search of Czechness in music. Nineteenth-Century Music,

10(1), 61–73.

Bell, D. A. (2009). The cult of the nation in France: Inventing nationalism, 1680-1800.

Harvard University Press.

Bernardes, G., Cocharro, D., Guedes, C., & Davies, M. E. (2016). Harmony generation

driven by a perceptually motivated tonal interval space. Computers in

Entertainment (CIE), 14(2), 1–21.

Blacking, J. (1974). How musical is man? University of Washington Press.

Bod, R. (2001). Probabilistic grammars for music. Belgian-Dutch Conference on

Artificial Intelligence (BNAIC).

Bod, R. (2002a). A unified model of structural organization in language and music.

Journal of Artificial Intelligence Research, 17, 289–308.

Bod, R. (2002b). Memory-based models of melodic analysis: Challenging the Gestalt

principles. Journal of New Music Research, 31(1), 27–36.

Bod, R. (2005). Towards Unifying Perception and Cognition: The Ubiquity of Trees.

Prepublication.

Bod, R. (2008). The data-oriented parsing approach: Theory and application. In

Computational Intelligence: A Compendium (pp. 307–348). Springer.

Bod, R. (2002c). A general parsing model for music and language. International

Conference on Music and Artificial Intelligence, 5–17.

Bod, R. (2002d). Combining Simplicity and Likelihood in Language and Music.

Proceedings of the Annual Meeting of the Cognitive Science Society, 24(24).

150

Bohlman, P. V. (1988). The study of folk music in the modern world. Indiana University

Press.

Bohlman, P. V. (2013). Central European folk music: An annotated bibliography of

sources in German. Routledge.

Böhme, F. M. (1895). Volksthümliche lieder der Deutschen im 18. Und 19. Jahrhundert.

Breitkopf und Härtel.

Brinkman, A., & Huron, D. (2018). The leading sixth scale degree: A test of Day-

O’Connell’s theory. Journal of New Music Research, 47(2), 166–175.

Bronson, B. H. (1969). The ballad as song. Univ of California Press.

Bruderer, M. J. (2008). Perception and modeling of segment boundaries in popular

music.

Cenkerová, Z., Hartmann, M., & Toiviainen, P. (2018). Crossing Phrase Boundaries in

Music. Proceedings of the Sound and Music Computing Conferences.

Chai, W., & Vercoe, B. (2001). Folk music classification using hidden Markov models.

Proceedings of International Conference on Artificial Intelligence, 6.

Chan, A. B., & Hsiao, J. H. (2016). Information distribution within musical segments.

Music Perception: An Interdisciplinary Journal, 34(2), 218–242.

Cherla, S. (2016). Neural Probabilistic Models for Melody Prediction, Sequence

Labelling and Classification [PhD Thesis]. City, University of London.

Cherla, S., Garcez, A. d’Avila, & Weyde, T. (2013). A neural probabilistic model for

predicting melodic sequences. International Workshop on Machine Learning and

Music.

151

Cherla, S., Tran, S. N., Garcez, A. d’Avila, & Weyde, T. (2015). Discriminative learning

and inference in the Recurrent Temporal RBM for melody modelling. 2015

International Joint Conference on Neural Networks (IJCNN), 1–8.

Chippindale, C. (2000). Capta and data: On the true nature of archaeological information.

American Antiquity, 65(4), 605–612.

Conklin, D. (2013). Folk tune classification with multiple viewpoints. Proceedings of the

Third International Workshop on Folk Music Analysis (FMA2013), 99.

Conklin, D., & Anagnostopoulou, C. (2006). Segmental pattern discovery in music.

INFORMS Journal on Computing, 18(3), 285–293.

Conklin, D., & Bergeron, M. (2008). Feature set patterns in music. Computer Music

Journal, 32(1), 60–70.

Davis, J. A. (2019). Maryland, My Maryland: Music and Patriotism During the

American Civil War. University of Nebraska Press.

Dean, R. T., & Pearce, M. T. (2016). Algorithmically-generated corpora that use serial

compositional principles can contribute to the modeling of sequential pitch

structure in non-tonal music. Empirical Musicology Review, 27–46.

Eck, D. (2005). Meter and autocorrelation. 10th Rhythm Perception and Production

Workshop (RPPW’05).

Eerola, T. (1999). Kohti kulttuurienvälistä kognitiivista musiikintutkimusta. Musiikin

Suunta, 1(1999), 21.

Eerola, T. (2016). Expectancy-violation and information-theoretic models of melodic

complexity. Empirical Musicology Review, 11(1), 2–17.

152

Eerola, T. (2004). Data-driven influences on melodic expectancy: Continuations in North

Sami Yoiks rated by South African traditional healers. Proceedings of the Eighth

International Conference of Music Perception and Cognition, 83–87.

Eerola, T., & Bregman, M. (2007). Melodic and contextual similarity of folk song

phrases. Musicae Scientiae, 11(1_suppl), 211–233.

Eerola, T., Himberg, T., Toiviainen, P., & Louhivuori, J. (2006). Perceived complexity of

western and African folk melodies by western and African listeners. Psychology

of Music, 34(3), 337–371.

Eerola, T., Jäärvinen, T., Louhivuori, J., & Toiviainen, P. (2001). Statistical features and

perceived similarity of folk melodies. Music Perception: An Interdisciplinary

Journal, 18(3), 275–296.

Eerola, T., Toiviainen, P., & Krumhansl, C. L. (2002). Real-time prediction of melodies:

Continuous predictability judgements and dynamic models. Proceedings of the

7th International Conference on Music Perception and Cognition, 473–476.

Elowsson, A. (2012). Statistical Analysis of vocal folk music. Stockholm: Royal Institute

of Technology.

Elowsson, A., & Friberg, A. (2012). Algorithmic composition of popular music. The 12th

International Conference on Music Perception and Cognition and the 8th

Triennial Conference of the European Society for the Cognitive Sciences of

Music, 276–285.

153

Erk, L. (1893). Deutscher Liederhort: Auswahl der vorzüglicheren deutschen

Volkslieder, nach Wort und Weise aus der Vorzeit und Gegenwart (Vol. 1).

Breitkopf und Härtel.

Erk, L. (1894). Deutscher Liederhort: Auswahl der vorzüglichern deutschen Volkslieder

aus der Vorzeit und der Gegenwart mit ihren eigenthümlichen Melodien (Vol. 3).

Breitkopf.

Eyers, T. (2013). The Perils of the" Digital Humanities": New Positivisms and the Fate of

Literary Theory. Postmodern Culture, 23(2).

Flossmann, S. (2012). Expressive performance rendering with probabilistic models:

Creating, analyzing, and using the Magaloff Corpus. na.

Frieler, K. (2014). Exploring Phrase Form Structures. Part I: European Folk Songs. Proc.

Int. Workshop Folk Music Anal., 108–110.

Gardiner, E., & Musto, R. G. (2015). The Digital Humanities: A primer for students and

scholars. Cambridge University Press.

Gkiokas, A., Katsouros, V., Lattner, S., Flexer, A., & Carayannis, G. (2015). Towards an

Invertible Rhythm Representation. Proceedings of the 18th Int. Conference on

Digital Audio Effects.

Golumbia, D. (2009). The cultural logic of computation. Harvard University Press.

Hagen, E. H., & Bryant, G. A. (2003). Music and dance as a coalition signaling system.

Human Nature, 14(1), 21–51.

154

Hannon, E. E., Snyder, J. S., Eerola, T., & Krumhansl, C. L. (2004). The role of melodic

and temporal cues in perceiving musical meter. Journal of Experimental

Psychology: Human Perception and Performance, 30(5), 956.

Harger, S., & Lemmen, L. J. (2013). The County of Bentheim and Her Emigrants to

North America. S. Harger.

Harker, D. (1985). Fakesong: The manufacture of British" folksong" 1700 to the present

day. Open University Press.

Harrison, P. M., Collins, T., & Müllensiefen, D. (2017). Applying modern psychometric

techniques to melodic discrimination testing: Item response theory, computerised

adaptive testing, and automatic item generation. Scientific Reports, 7(1), 1–18.

Hedges, T., & Wiggins, G. A. (2016). The prediction of merged attributes with multiple

viewpoint systems. Journal of New Music Research, 45(4), 314–332.

Hillewaere, R., Manderick, B., & Conklin, D. (2012). String Methods for Folk Tune

Genre Classification. ISMIR, 2012, 13th.

Honingh, A., & Weyde, T. (2008). Integrating Convexity and Compactness into the

ISSM: Melodic Analysis of Music.

Hoos, H. H., Renz, K., & Görg, M. (2001). GUIDO/MIR-an Experimental Musical

Information Retrieval System based on GUIDO Music Notation. ISMIR, 41–50.

Huron, D. (1995). The Humdrum Toolkit: Reference Manual. Center for Computer

Assisted Research in the Humanities.

Huron, D. (1996). The melodic arch in Western folksongs. Computing in Musicology, 10,

3–23.

155

Huron, D. (2002). Music Informaiton Processing Using the Humdrum Toolkit: Concepts,

Examples, and Lessons. Computer Music Journal, 26(2), 15–30.

Huron, D. (2013). On the virtuous and the vexatious in an age of big data. Music

Perception: An Interdisciplinary Journal, 31(1), 4–9.

Huron, D., & Royal, M. (1996). What is melodic accent? Converging evidence from

musical practice. Music Perception: An Interdisciplinary Journal, 13(4), 489–

516.

Huron, D., Yim, G., & Chordia, P. (2010). The effect of pitch exposure on sadness

judgments: An association between sadness and lower than normal pitch.

Proceedings of the 11th International Conference on Music Perception and

Cognition, 63–66.

Jan, S. (2018). “The Two Brothers”: Reconciling Perceptual-Cognitive and Statistical

Models of Musical Evolution. Frontiers in Psychology, 9, 344.

Jasmin, K., Dick, F., Holt, L., & Tierney, A. T. (2018). Degeneracy makes music and

speech robust to individual differences in perception. BioRxiv, 263079.

Juhász, Z. (2009). Automatic segmentation and comparative study of motives in eleven

folk song collections using self-organizing maps and multidimensional mapping.

Journal of New Music Research, 38(1), 71–85.

Juhász, Z., & Sipos, J. (2010). A Comparative Analysis of Eurasian Folksong Corpora,

using Self Organising Maps. Journal of Interdisciplinary Music Studies, 4(1).

156

Kaliakatsos-Papakostas, M. (2019). Generative Conceptual Blending of High-Level

Melodic Features: Shortcomings and Possible Improvements. International

Conference of New Music Concepts, ICNMC.

Khoo, S., Man, Z., & Cao, Z. (2012). Automatic han chinese folk song classification

using extreme learning machines. Australasian Joint Conference on Artificial

Intelligence, 49–60.

Khoo, S., Man, Z., Cao, Z., & Zheng, J. (2013). German vs. Austrian folk song

classification. Industrial Electronics and Applications (ICIEA), 2013 8th IEEE

Conference On, 131–136.

Kirschenbaum, M. G. (2016). What is digital humanities and what’s it doing in English

departments? In Defining Digital Humanities (pp. 211–220). Routledge.

Krumhansl, C. L., & Kessler, E. J. (1982). Tracing the dynamic changes in perceived

tonal organization in a spatial representation of musical keys. Psychological

Review, 89(4), 334.

Krumhansl, C., & Schellenberg, E. G. (1990). An empirical test of the implication-

realization model. Conference on Music and the Cognitive Sciences, Cambridge,

UK.

Laade, W. (1988). Musik und Musiker in Marchen, Sagen, und Anekdoten der Volker

Europas: Eine Quellensammlung zum Problemkreis “Musik als Kultur” (“Music

and Musicians in Fairy Tales, Tales and Anecdotes of European Peoples: A

Source Collection for the Problematic ‘Music as Culture’”): Vol. 1: Mitteleuropa.

Valentin Koerner.

157

Lambert, A. J., Weyde, T., & Armstrong, N. (2014). Studying the effect of metre

perception on rhythm and melody modelling with LSTMs. Tenth Artificial

Intelligence and Interactive Digital Entertainment Conference.

Lambert, A., Weyde, T., & Armstrong, N. (2014). Beyond the beat: Towards metre,

rhythm and melody modelling with hybrid oscillator networks. International

Computer Music Association.

Lanigan, R. L. (1994). Capta versus data: Method and evidence in communicology.

Human Studies, 109–130.

Lattner, S., Grachten, M., Agres, K., & Chacón, C. E. C. (2015). Probabilistic

segmentation of musical sequences using restricted Boltzmann machines.

International Conference on Mathematics and Computation in Music, 323–334.

Lattner, S., Grachten, M., & Widmer, G. (2018). A predictive model for music based on

learned interval representations. Proceedings of the 19th International Society for

Music Information Retrieval Conference.

Laufhutte, H. (1991). Deutsche Balladen (“German Ballads”). Reclam.

Li, Y., & Huron, D. (2006). Melodic Modeling: A Comparison of Scale Degree and

Interval. ICMC.

Ling, J. (1997). A history of European folk music. University Rochester Press.

Lomax, A. (1976). Cantometrics: An approach to the anthropology of music. distributed

by University of California, Extension Media Center.

158

Margulis, E. H. (2007). Silences in music are musical not silent: An exploratory study of

context effects on the experience of musical pauses. Music Perception: An

Interdisciplinary Journal, 24(5), 485–506.

McBride, J. M., & Tlusty, T. (2019). Imperfect fifths pack into musical scales. ArXiv

Preprint ArXiv:1906.06171.

McLeod, A., & Steedman, M. (2017). Meter detection in symbolic music using a

lexicalized PCFG. Proceedings of the 14th Sound and Music Computing

Conference, 373–379.

Meier, J., Heiske, W., Quellmalz, F., Schewe, H., & Seemann, E. (Eds.). (1935).

Deutsche Volkslieder mit ihren Melodien (“German Folk Songs with Their

Melodies”) (Vol. 1). Walter de Gruyter.

Meyer, L. B. (1989). Style and music: Theory, history, and ideology. University of

Chicago Press.

Morgan, E., Fogel, A., Nair, A., & Patel, A. D. (2019). Statistical learning and Gestalt-

like principles predict melodic expectations. Cognition, 189, 23–34.

Morgenstern, U. (2018). Towards the History of Ideas in Ethnomusicology: Theory and

Methods Between the Late 18th and the Early 20th Century. Musicologist, 2(1),

1–31.

Morrison, S. J., Demorest, S. M., & Pearce, M. T. (2018). Cultural Distance: A

Computational Approach to Exploring Cultural Influences on Music Cognition. In

The Oxford Handbook of Music and the Brain.

159

Müllensiefen, D., Harrison, P., Caprini, F., & Fancourt, A. (2015). Investigating the

importance of self-theories of intelligence and musicality for students’ academic

and musical achievement. Frontiers in Psychology, 6, 1702.

Müllensiefen, D., & Wiggins, G. (2011). Polynomial functions as a representation of

melodic phrase contour. na.

Müllensiefen, D., Wiggins, G., & Lewis, D. (2008). High-level feature descriptors and

corpus-based musicology: Techniques for modelling music cognition. Citeseer.

Narmour, E. (1990). The analysis and cognition of basic melodic structures: The

implication-realization model. University of Chicago Press.

Nettl, B. (2010). The study of ethnomusicology: Thirty-one issues and concepts.

University of Illinois Press.

Packard Humanities Institute’s Center for Computer Assisted Research in the Humanities

at Stanford University. (2015). EsAC. Wiki.Ccarh.Org.

https://wiki.ccarh.org/wiki/EsAC

Patel, A. D. (2010). Music, language, and the brain. Oxford university press.

Pazdera, J. (2016). Computational modeling of musical enculturation: An investigation of

multicultural music learning using self-organizing maps [PhD Thesis].

Pearce, M. T., & Wiggins, G. A. (2004). Rethinking gestalt influences on melodic

expectancy. Proceedings of the 8th International Conference on Music Perception

and Cognition, 367–371.

160

Pearce, M., & Wiggins, G. (2003). An empirical comparison of the performance of ppm

variants on a prediction task with monophonic music. Artificial Intelligence and

Creativity in Arts and Science Symposium.

Pearce, M., & Wiggins, G. (2004). Improved methods for statistical modelling of

monophonic music. Journal of New Music Research, 33(4), 367–385.

Pearce, Marcus T., & Eerola, T. (2017). Music perception in historical audiences:

Towards predictive models of music perception in historical audiences. Journal of

Interdisciplinary Music Studies., 8(1–2), 91–120.

Pearce, Marcus T., Müllensiefen, D., & Wiggins, G. A. (2010a). Melodic grouping in

music information retrieval: New methods and applications. In Advances in music

information retrieval (pp. 364–388). Springer.

Pearce, Marcus T., Müllensiefen, D., & Wiggins, G. A. (2010b). The role of expectation

and probabilistic learning in auditory boundary perception: A model comparison.

Perception, 39(10), 1367–1391.

Pearce, Marcus T., Müllensiefen, D., & Wiggins, G. A. (2008). A Comparison of

Statistical and Rule-Based Models of Melodic Segmentation. ISMIR, 89–94.

Pearce, Marcus T., & Wiggins, G. A. (2006). Expectation in melody: The influence of

context and learning. Music Perception: An Interdisciplinary Journal, 23(5), 377–

405.

Pearce, Marcus T., & Wiggins, G. A. (2012). Auditory expectation: The information

dynamics of music perception and cognition. Topics in Cognitive Science, 4(4),

625–652.

161

Prince, J. B., Tan, S. E. J., & Schmuckler, M. A. (2019). Listeners perceive complex

pitch-temporal structure in melodies. Memory & Cognition, 1–15.

Ren, I. Y. (2016). Closed patterns in folk music and other genres.

Robb, D. (2016). The mobilising of the German 1848 protest song tradition in the context

of international twentieth-century folk revivals. Popular Music, 35(3), 338–359.

Rohrich, L. (1973). Die Textgattungen des popularen Liedes (“The Textual Genres of

Popular Song”). Handbuch Des Volksliedes, 1(Die Gattungen des Volksliedes),

19–35.

Rohrmeier, M., Rebuschat, P., & Cross, I. (2011). Incidental and online learning of

melodic structure. Consciousness and Cognition, 20(2), 214–222.

Ronca, P. (2009). Pattern recognition by simplicial analysis. 2009 Sixth International

Conference on Fuzzy Systems and Knowledge Discovery, 7, 155–158.

Sadakata, M. (2006). Ritme and Rizumu: Studies in music cognition.

Sadakata, M., Desain, P., & Honing, H. (2006). The Bayesian way to relate rhythm

perception and production. Music Perception, 23(3), 269–288.

Savage, P. E., Brown, S., Sakai, E., & Currie, T. E. (2015). Statistical universals reveal

the structures and functions of human music. Proceedings of the National

Academy of Sciences, 112(29), 8987–8992.

Schaefer, R. S., Murre, J. M., & Bod, R. (2004). Limits to universality in segmentation of

simple melodies.

Schäfer, T., Huron, D., Shanahan, D., & Sedlmeier, P. (2015). The sounds of safety:

Stress and danger in music perception. Frontiers in Psychology, 6, 1140.

162

Schaffrath, H. (1995). The Essen Folksong Collection (D. Huron, Ed.). Center for

Computer Assisted Research in the Humanities.

Schneider, M. (1982). Jodler aus Tirol (“Yodels in Tirol”). Eigenverlag des Tiroler

Volksliedwerkes.

Schroubek, G. (1973). Das Wallfahrts- und Prozessionslied. (“The Pilgramage and

Processional Songs”). Handbuch Des Volksliedes, 1(Die Gattungen des

Volkslied), 445–462.

Shanahan, D., & Albrecht, J. (2019). Examining the Effect of Oral Transmission on

Folksongs. Music Perception: An Interdisciplinary Journal, 36(3), 273–288.

Shuker, R. (2012). Understanding popular music culture. Routledge.

Spevak, C., Thom, B., & Höthker, K. (2002). Evaluating melodic segmentation.

International Conference on Music and Artificial Intelligence, 168–182.

Steinbeck, W. (1976). The use of the computer in the analysis of German folksongs.

Computers and the Humanities, 10(5), 287–296.

Steinbeck, W. (1982). Struktur und Ähnlichkeit: Methoden automatisierter

Melodieanalyse. Kieler Schriften zur Musikwissenschaft XXV. Kassel, Basel,

London: Bärenreiter.

Suppan, W. (1966). Volkslied: Seine Sammlung und Erforschung: Vol. 2 vols. Stuttgart.

Tanji, M., Ando, D., & Iba, H. (2008). Improving metrical grammar with grammar

expansion. Australasian Joint Conference on Artificial Intelligence, 180–191.

Taylor, R. J. (2017). Die Melodien der weltlichen Lieder des Mittelalters:

Darstellungsband. Springer-Verlag.

163

Temperley, D. (2000). Meter and grouping in African music: A view from music theory.

Ethnomusicology, 44(1), 65–96.

Temperley, D. (2003). End-Accented Phrases: An Analytical Exploration. Journal of

Music Theory, 47(1), 125–154.

Temperley, D. (2004). The cognition of basic musical structures. MIT press.

Temperley, D. (2007). Music and probability. Mit Press.

Temperley, D. (2008). A probabilistic model of melody perception. Cognitive Science,

32(2), 418–444.

Temperley, D. (2009). A unified probabilistic model for polyphonic music analysis.

Journal of New Music Research, 38(1), 3–18.

Temperley, D. (2014). Probabilistic models of melodic interval. Music Perception: An

Interdisciplinary Journal, 32(1), 85–99.

Temperley, D. (2019). Second-Position Syncopation in European and American Vocal

Music. Empirical Musicology Review, 14(1–2), 66–80.

Temperley, D., & Marvin, E. W. (2008). Pitch-class distribution and the identification of

key. Music Perception: An Interdisciplinary Journal, 25(3), 193–212.

Thornton, C. (2011). Generation of folk song melodies using bayes transforms. Journal

of New Music Research, 40(4), 293–312.

Toiviainen, P., & Eerola, T. (2001a). Self-organizing map of the essen collection. ISSCM.

Toiviainen, P., & Eerola, T. (2003). Where is the beat?: Comparison of Finnish and South

African listeners. Proceedings of the 5th Triennial ESCOM Conference, 501–504.

164

Toiviainen, P., & Eerola, T. (2006a). Autocorrelation in meter induction: The role of

accent structure. The Journal of the Acoustical Society of America, 119(2), 1164–

1170.

Toiviainen, P., & Eerola, T. (2006b). Visualization in comparative music research. In

COMPSTAT 2006-Proceedings in Computational Statistics (pp. 209–219).

Springer.

Toiviainen, P., & Eerola, T. (2002). A computational model of melodic similarity based

on multiple representations and self-organizing maps. Proceedings of the Seventh

International Conference on Music Perception and Cognition, Sydney. Causal

Productions, Adelaide, 236–239.

Toiviainen, P., & Eerola, T. (2001b). A method for comparative analysis of folk music

based on musical feature extraction and neural networks. III International

Conference on Cognitive Musicology, 41–45.

Toiviainen, P., & Eerola, T. (2005). Classification of Musical Metre with Autocorrelation

and Discriminant Functions. ISMIR, 351–357.

Toiviainen, P., & Eerola, T. (2004). The role of accent periodicities in meter induction: A

classification study. Proceedings of the Eighth International Conference of Music

Perception and Cognition, 422–425. van der Weij, B., Pearce, M. T., & Honing, H. (2017). A Probabilistic Model of Meter

Perception: Simulating Enculturation. Frontiers in Psychology, 8.

https://doi.org/10.3389/fpsyg.2017.00824

165

Van Kranenburg, P., de Bruin, M., Grijp, L. P., & Wiering, F. (2014). The Meertens Tune

Collections. Meertens Online Reports, No. 2014-1.

Van Kranenburg, P., Garbers, J., Volk, A., Wiering, F., Grijp, L. P., & Veltkamp, R. C.

(2010). Collaboration perspectives for folk song research and music information

retrieval: The indispensable role of computational musicology. Jims, 2009, 030.

Van Kranenburg, P., Garbers, J., Volk, A., Wiering, F., Grijp, L., & Veltkamp, R. (2007).

Towards integration of music information retrieval and folk song research.

Proceedings of the 8th International Conference on Music Information Retrieval,

505–508.

Van Kranenburg, P., & Janssen, B. (2014). What to do with a digitized collection of

western folk song melodies. Proceedings of the 4th Workshop on Folk Music

Analysis (FMA). van Kranenburg, P., Volk, A., & Wiering, F. (2013). A comparison between global and

local features for computational classification of folk song melodies. Journal of

New Music Research, 42(1), 1–18.

Vogel, Joh. N. (Ed.). (1845). Liedertafel: Eine Sammlung von Liedern, Romanzen und

Balladen.

Volk, A. (2012). Melodic segmentation using the jensen-shannon divergence. 2012 11th

International Conference on Machine Learning and Applications, 2, 351–356.

Von Hippel, P. (2000). Redefining pitch proximity: Tessitura and mobility as constraints

on melodic intervals. Music Perception: An Interdisciplinary Journal, 17(3), 315–

327.

166

Von Hippel, P., & Huron, D. (2000). Why do skips precede reversals? The effect of

tessitura on melodic structure. Music Perception: An Interdisciplinary Journal,

18(1), 59–85.

Wiggins, G. A., Pearce, M. T., & Müllensiefen, D. (2009). Computational modeling of

music cognition and musical creativity. na.

Wiora, W. (1949). Concerning the conception of authentic folk music. Journal of the

International Folk Music Council, 1, 14–19.

Wiora, W. (1964). The origins of German Spiritual Folk Song: Comparative Methods in a

Historical Study. Ethnomusicology, 8(1), 1–13.

Wiora, W. (1977). “Gattungen des Volksliedes” und “Gattungen der Musik.” (‘ “Genres

of Folk Song” and “Genres of Music”’). Neue Ethnomusikologische

Forschungen: Festschrift Felix Hoerburger, 37–44.

Yim, G. K. (2014). Two Studies on Assessing Emotional Responses to Music and Mode:

The Effect of Lowered Pitch on Sadness Judgments, and the Affective Priming

Paradigm as an Implicit Measure [PhD Thesis]. The Ohio State University.

Zaane, M. van, Bod, R., & Honing, H. (2003). A memory-based approach to meter

induction.

Zelter, C. F. (Ed.). (1812). Lieder, Balladen und Romanzen für das Piano-Forte: Vol. 3

vols.

167