IN his study, Martin (2025) created a corpus of 750 country songs spanning 1949-2023 to examine the history of swing within the genre. His findings show that swing in country music falls out of use over the length of the corpus, that duple rhythms are the most prevalent within the corpus, that faster tempos are generally used for songs with first-level subdivisions, and that slower tempos are generally used for songs with second-level subdivisions. [1]
This promising research gives the field greater insight into country music, the usage of swing, and types of swing styles within country music. Further, this corpus highlights several decades, primarily the 1980s-2000s, that are worth further intellectual inquiry into the historical and industry conditions that surround this period. My commentary here will mostly focus on Martin’s compiling and encoding of the corpus, what role that songs from the 1980s-2000s play within it, and finally how the data from these decades raise questions that should be further addressed.
Admittedly, corpus creation is complicated and time-consuming, requiring making contestable decisions about what should be encoded and how to encode it. It takes a massive amount of labor to understand what kind of corpus is needed to answer any given research question. Martin chooses to replace repeating songs across years with the next song down (see Methods), but this does potentially create a different picture for how music in that year was experienced. There are then an equal number of difficult questions that go into how to encode the data that goes into the corpus, including decisions about how repeated songs are handled, how chart sources are combined, and how individual stylistic influences are represented within the dataset. There is no single correct way to approach these tasks and corpus construction necessarily involves a series of informed methodological tradeoffs. What I seek to do in this section is not to invalidate any of the great work done by Martin. Instead, I wish to highlight possible factors at play that might create different data sets. Further, I wish to show the strength of his encoding methods and offer ways to further add credence to his work.
One quirk in the creation of this corpus is that it is not made up of any singular chart, but rather three different charts: Joel Whitburn’s Top Country Singles 1944-1993, The Billboard book of Top 40 Country, and Billboard’s online “Hot Country Songs” year-end charts. Martin is clear that there is a lack of archival data with year-end charts for country music, and the three sources he uses are probably the best he could use. Still, it presents a methodological challenge to use three unrelated sources. While it is helpful that the first two catalogs are by Joel Whitburn, meaning a singular person compiled them, it is then odd to use a third chart that is compiled by radio airplay. While stopping at 2006 (where the “Hot Country Songs” portion of the corpus begins) would present its own set of challenges, that would be one way to get a “cleaner” view of the songs in the corpus. It might also be helpful to consider recent scholarship on corpus compiling. Shea et al. (2024) examine social forces in corpora which are implicit biases that can shape the compiling of a corpus. At the same time, deClereq (2025) pushes back against these approaches, arguing that rebalancing methods have limited explanatory power. [2] Country music is a complicated example of how issues of gender, class, and sexuality play out (Hubbs, 2014). While Martin addresses this by citing Jade Watson’s (2019) research, it could prove useful to adopt some of the approaches taken by Shea et al. to see if the trends of the corpus still hold when accounting for forms of implicit bias, for example their Anti-Discriminatory Alignment System. Martin notes that his corpus is limited by possible selection bias, and methods like these could help combat against them.
A possible confounding factor in the corpus, particularly in the more recent portion of the dataset, is the role of the producer(s). While Martin controlled for songs that were repeated in subsequent years, this control only applies to those performing the song. If the performing artist is not the individual writing the song and a producer is, then any effect of the producer's style is not accounted for if it is the artist who is dropped when songs are repeated. [3] While I am not sure this would produce any new trends, it is something to consider.
Martin writes that encoding the songs into an “objective numerical format” was difficult, partly due to the need to separate his biases. However, the justification for his encoding process, and how he went about it, in my view, are quite strong. He identifies the potential tactus by using the rate between the kickdrum and the snare drum, what Martin calls the downbeat/backbeat, a common musical features across songs. This approach is musically grounded and avoids the potential of having to use an AI software to find a tempo, which can often prove problematic. And while not every song makes use of kickdrums and snare drums, Martin is able to consider other musical features as stand ins, for example, fiddle “chucks” or palm muted strings.
As a final note in regard to the creation of Martin’s corpus, having one or two other individuals validate a small subset of the corpus could have been an effective method for reducing subjectivity, and could still serve this purpose in future work. Even having the rate of agreement amongst two or three people across ten to twenty songs could go a long way in showing how much others would agree with Martin’s encoding practices. Still, as stated above, I find his method of encoding to be both valuable and musically sophisticated.
In looking at the findings of the corpus, I continually found myself intrigued by varying differences that appeared largely within the late 1980s-2000s. Martin first highlights this in showing that the 1990s and 2000s had extremely low swing usage, with some years at zero songs that used non-duple rhythms. Further, while we see tempo going down over time, and this is supported by a linear regression, Martin does point to some slight increases in tempo in the late 1980s and early 1990s. Finally, Martin sees the year 1990 as a cutoff point between earlier songs that are often above 110 bpm and swing at the first metric layer, and songs appearing after that are more likely to be notated in a compound meter. Martin concludes with saying these moments need more inquiry, and I would like to speculate as to some possible factors that played a role.
While it is only in the late 1980s that we start to see some oddities within Martin’s corpus, I am left wondering about the possible effect of MTV (debuting in 1981) and Country Music Television (debuting in 1983). MTV is certainly known for its profound effect on reshaping the music industry, at least from a visual perspective (Goodwin, 1992). However, I wonder to what degree some of the influences of seeing music might have had on shaping the sounds that artists played. If swing is a historical style, perhaps in trying to appear more visually appealing artists also tried to have a more modern style to their song writing. And, it would take a few years for these differences to appear within a corpus, so the dates decently line up.
It is also worth speculating about the effect that Garth Brooks, George Strait, and Shania Twain had on music at that time, being dominant through the late 80s and 90s. Their first albums debuted in 1989, 1984, and 1993 respectively, and at the time of writing have respectively sold 170.3, 110.8, and 87.4 million albums. Within Martin’s corpus, George Strait occurs 17 times, 11 of which are from the 90s, which is more than any other artist within that timeframe. [4] While Martin tries to control for songs being repeated across years, it would make sense that the influence these artists had over other artists at the time allows for their musical styles to appear in the songwriting of others. While influence might not be found directly in Martin’s corpus, a constant trope within the music industry is finding ways to replicate current success, and these artists certainly did.
Overall, Martin’s corpus is an excellent contribution to the field and offers a look into the role of swing music from 1949-2023. While the corpus was built to only consider swing music and types of swing styles, the trends found throughout his article can allow for new ways of thinking about swing in country music, new trends to consider, and possible experimental follow-ups. While any corpus is open to being influenced by the biases of the creator, Martin does a great job at controlling for his own biases.
This article was copyedited by Jonathan Tang and layout-edited by Niels Chr. Hansen
de Clercq, T. (2025). What does a corpus of music represent? Commentary on “Diversity in Music Corpus Studies.” Music Theory Online, 31(4). https://doi.org/10.30535/mto.31.4.9
Goodwin, A. (1992). Dancing in the Distraction Factory: Music Television and Popular Culture. University of Minnesota Press.
Hubbs, N. (2014). Rednecks, Queers, and Country Music. University of California Press.
Martin, S. (2025). Swing usage in country music (1949-2023) via corpus analysis. Empirical Musicology Review, 20(2), 72-85. https://doi.org/10.18061/emr.6602
Shea, N., Reymore, L., White, C. W., Duinker, B., VanHandel, L., Zeller, M., & Biamonte, N. (2024). Diversity in music corpus studies. Music Theory Online, 30(1). https://doi.org/10.30535/mto.30.1.8
Watson, J. (2019). Gender on the Billboard Hot Country Songs Chart, 1996-2016. Popular Music and Society, 42(5), 538–560. https://doi.org/10.1080/03007766.2018.1512070