A corpus of the most popular American songs from 1900 to 1999
The Pop20c Corpus includes 100 songs of American popular music: the most popular song from each year of the 20th century. The songs from 1900 through 1954 were selected based on the charts from Joel Whitburn's book A Century of Pop Music; these charts in turn were based on other charts reflecting record sales as well as other media such as sheet music sales, radio play, and jukebox play. Songs from 1955 through 1999 were selected from the Billboard Hot 100 chart. Click here to view a table that displays the year, title, and artist of every song in the corpus.
The Pop20c Corpus was developed by Joseph VanderStel (email@example.com) and David Temperley (firstname.lastname@example.org). For more about our methodology, see our article “The Evolution of Syncopation in Twentieth-Century American Popular Music” in Journal of New Music Research (2022). (PDF) We originally developed the corpus in connection with that study, but it may also be useful for other music researchers and enthusiasts who wish to study the rich tradition of American popular music.
Songs are encoded as complete vocal melodies using a modified version of a format used in the Rolling Stone Corpus. Encodings are based on transcriptions from recordings rather than sheet music.
There are three files associated with each song. Below is a description of each file type, using the opening measures of John McCormack's “It's A Long Way To Tipperary” (1915) as an example:
(View GitHub source; download zipped folder with all 100 transcription files)
The scale degrees and rhythms of the melody are encoded using a format that enables rapid transcription of the song in near real time. Here are the opening measures of the transcription file for “It's A Long Way To Tipperary”:
1..3 2~1~ 6~5~ 3~45 | ~~6~ 5~3~ 5~~~ .... |
This is almost identical to the format that is used in the Rolling Stone Corpus (click here for an overview), with one exception: in the Rolling Stone Corpus, both continuations and rests are encoded with periods (
.), but the present corpus differentiates continuations with a tilde (
Melismas are encoded with parentheses around the notes of the melisma. This follows a convention established in an extension of the original Rolling Stone format.
(View GitHub source; download zipped folder with all 100 lyric files)
The corpus includes lyrics for each song, which were crosschecked with multiple sources on the internet. Here are the lyrics of the above excerpt:
Up to mighty London came An Irish man one day
(View GitHub source; download zipped folder with all 100 note list files)
A “note list” file is generated from the raw transcription and lyric files using a custom script, and is designed to be machine-readable. The output file represents each melody as a list of note statements. Here is the note list for the above excerpt:
4.0000 4.0625 60 0 0 UP 4.1875 4.2500 64 4 0 TO 4.2500 4.3750 62 2 1 MIGHTY 4.3750 4.5000 60 0 0 MIGHTY 4.5000 4.6250 57 9 1 LONDON 4.6250 4.7500 55 7 0 LONDON 4.7500 4.8750 52 4 1 CAME 4.8750 4.9375 53 5 0 AN 4.9375 5.1250 55 7 1 IRISH 5.1250 5.2500 57 9 0 IRISH 5.2500 5.3750 55 7 1 MAN 5.3750 5.5000 52 4 0 ONE 5.5000 5.7500 55 7 1 DAY
Each line contains six values:
5.2500means the second quarter-note beat of the sixth measure.
1being stressed, and
MIGHTYrefers to the second syllable of the word “mighty.”
Syllabic stress values are calculated by mapping each syllable from the lyric file to one of three stress values according to the Carnegie Mellon University Pronouncing Dictionary: 0 for unstressed syllables (to-ma-to), 1 for stressed syllables (to-ma-to), and 2 for syllables with secondary stress (to-ma-to). Following the convention of prior studies, all monosyllabic function words (such as articles, pronouns, and prepositions) are assigned an unstressed syllable, and 2's are converted to 1's, i.e. treated as stressed syllables.
In order to create a one-to-one relationship between notes and syllables, note lists only include the first note of a melisma.
N.B.: Several of the songs in the corpus (especially in the very early years of the century) contain lyrics that are negative and offensive towards African-Americans. While we apologize for this, we felt it was best to maintain objective, consistent criteria in selecting the corpus. Many popular songs from the very early 1900s contain racist lyrics, and we did not wish to deny this problematic aspect of American musical history.