Earlier this year, I was commissioned to research the history of the BRIT Awards’ Song Of The Year, to celebrate 40 years from the first award (Tainted Love, Soft Cell, 1982) to the most recent (Watermelon Sugar, Harry Styles, 2021). We did a similar thing back in 2020 for Netflix with 10 Years of Eurovision. My team and I analyzed the 40 winners and the 289 nominees, for a total of 329 songs. Our aim was to look at trends over time, to see what we could learn about the evolution of British songwriting and music tastes. Because we used some human-encoded data (chord loops, tonality, lyric theme, vocalist gender) we were able to look at qualitative factors to investigate songwriting trends over time – at least in the most popular songs (i.e. the nominees). The 40 winners are playlisted below; for those with 21 hours and 41 minutes to spare, there’s also a playlist of all 329 nominees, in date order 1982-2021.
Having immersed ourselves in these songs and their metadata for an intense period, we observed some well-known (and a few not-so-well-known) trends in British songwriting over the 40 years. For example, the number of co-writers in a song has been steadily rising; songs are getting shorter; heartbreak songs are disproportionately popular among the winners; 60% of all 329 songs feature a four-chord loop in the chorus. At the time of writing, we’re still exploring the dataset, but I’m sharing some of our initial findings here in advance of the 2022 BRIT Awards ceremony.
This being a blog, I’ve bumped the ‘Methodology’ section to the end of this post, because most people will just want to read the summary of findings. We haven’t fully interrogated the dataset yet, but our initial queries have yielded some really interesting results.
Research team: Joe Bennett (JBMS LLC/Berklee USA), Simon Troup (Digital Music Art UK), William Weston Bennett & Jacob Sunshine (PhD musicology researchers, Harvard University USA). Research funded by MasterCard (sponsors of The BRITs).
Songs are getting shorter.
The shortest song in our dataset was Blur’s Song 2 (1998) at [2:02]; the longest is Dire Straits’ Money For Nothin’ (1986, album version) at [8:24]. Interestingly, the second longest, Tender (2000) was also by Blur, at [7:42]. Although the overall average is an unremarkable [3:58], the trend over our 40 year timeline is towards shorter songs, with some apparent cyclical trends. After a short-song start in 1982, with Adam And The Ants’ Prince Charming and Stand and Deliver combined with Soft Cell’s Tainted Love providing an average of [3:09], song length for the rest of that decade stayed high, with an average of [4:24]. Long songs made a brief reappearance near the turn of the century (Angels, Road Rage, Teardrop, Praise You), before durations shrank again. Since 2017, there has been a dramatic reduction in duration, with the 2021 average at [3:04] – the shortest songs in BRITs history. Several commentators have ascribed this phenomenon to the rise of TikTok and the algorithmic behaviours of Spotify.
Chorus-first songs are becoming more popular.
In the early 80s, most of the songs began with the first verse; by 2021, 50% of the SOTY nominees began with the chorus (Rain, Rover, Secrets, Don’t Rush, and Ain’t It Different). Chorus-first songs are also common among the SOTY winners, including Pass Out (Tinie Tempah, 2011), The Promise (Girls Aloud, 2009), and Relax (Frankie Goes To Hollywood, 1985).
Most choruses use chord loops.
Of the 329 finalists, 198 (60%) use a four-chord loop in the chorus – that is, a pattern of four chords played across 2, 4, or 8 bars, repeating at least once. In fact, eight songs use the same chord loop – the famous 6-4-1-5 (e.g. Am, F, C, G): these are Leave a Light On, Physical, Patience, All Time Low, Number 1, Parachute, Burn, I Need Your Love, Break Your Heart, and How Deep Is Your Love, which are all from 2010-2019. The chart below shows the distribution of the eight most common chord loops (among the 40 songs that contained these loops), with examples of the songs that they appear in. For simplicity, these are shown here as chord progressions in C major / A minor, replacing the Nashville chord numbers that were actually used to code each song.
Song titles are getting shorter.
The longest song title in SOTY BRITs history is I Wish I Was a Punk Rocker (with Flowers in My Hair) (Sandi Thom, 2008), followed by If You Tolerate This Your Children Will Be Next (Manic Street Preachers, 1999). Short titles have become much more common in recent years, with half of the 2021 nominees having one-word titles – Lighter, Physical, Rain, Rover, Secrets. Of the 329 nominees, 75 have only one word. This phenomenon is observable beyond the BRITs corpus, and has been noted by other commentators.
Collaboration is becoming more common.
The average number of co-writers has been steadily rising throughout the 40-year period. One- and two-writer songs were common in the early 1980s, and became rarer over time. 1982’s winner Tainted Love was written by a single writer (Ed Cobb in 1964), and its two co-nominees Prince Charming and Stand And Deliver were written for Adam And The Ants by co-writing partnership Adam Ant and Marco Pirroni. During the 1990s the number of writers per song began to rise, through band co-writes (Snow Patrol’s Chasing Cars, 2007), bands with other co-writers (Take That’s Shine and Patience, 2007 and 2008 winners), artists with large songwriting and production teams (Mark Ronson feat. Bruno Mars’s Uptown Funk, 2015), and at least one band who decided to spread royalties equally for presumably political reasons (Chumbawamba’s Tubthumping, 1998; they were, after all, an anarcho-communist collective). By 2021 the average number had risen to 5.1, helped by team efforts such as Head And Heart (8 writers) and Ain’t It Different (13 writers). This is a general trend in songwriting, due partly to the rise of Songwriting Camps, where artists and writers get together for intensive co-writing sessions .
Explicit lyrics have become more common in the late 2010s.
Explicit lyrics do not appear in the BRITs corpus until 1994 (Radiohead’s Creep) although it should be noted that the infamous ‘Parental Advisory’ RIAA policy was not introduced in the US until 1985, and was not adopted by the UK’s BPI until 2011. The first controversial lyric content appeared in Frankie Goes To Hollywood’s Relax (1985), which was briefly banned by the BBC due to its sexual (and homosexual) themes and cover art. From 2017 to 2021, the proportion of explicit lyrics among nominees has been trending upward. In 2021, 50% of the nominated songs featured explicit lyrics.
“Winter” love songs make up 12% of nominees but 20% of winners.
59.7% of the lyrics were about themes related to romantic love. To divide these further, we classified love songs according to a calendar scheme: “Spring” songs are about early days, looking for love, flirting, and falling in love; “Summer” songs are about being in a relationship where all is well; “Autumn” songs cover love on the rocks and relationship issues, and “Winter” songs are about the aftermath, after the lovers have parted ways. Of all the lyric categories, “Winter” songs saw the biggest jump from the pool of nominees to the select set of winners—12% to 20%.
We extracted metadata from the corpus using a combination of computer-automated and human-coding methods. For the 329 nominees, we coded the following information:
- Originality category (original song, cover version, remix, or sample-based)
- Length of intro (time from start of track to the main vocal entering)
- Chorus time (time from start of track to the first chorus)
- Explicit lyric yes/no
- Spotify’s AI-generated Valence (aka ‘happiness score’)
- Spotify’s AI-generated Loudness (the average loudness of the track across its duration)
- Tempo (beats per minute)
- Lyric theme – Character, Dance Party, History, Home / Family, Love (Spring, Summer, Autumn or Winter), Money, Places, Self-Belief, Social, Spirituality, Youth, Other.
- Key signature
- Tonality (the main musical scale used in the vocal melody and underlying chords)
- Chorus chord loop (for choruses that used looping materials)
- Vocalist gender (Female, Male, Female Group, Male Group, Mixed group)
- Genre/style descriptors
Coding the data
Publicly available data (composer, title, release date, songwriter credits, lyric text) were human-researched from Internet sources, including Musixmatch and ASCAP.com.
Machine-read data were sourced from the Spotify API, including Duration, Valence, Explicitness, and Loudness.
Machine-read musical data (BPM, key) were machine-extracted then human-corrected; Spotify’s API dataset showed a number of musical errors, particularly in modal melodies and compound time signatures for BPM and key respectively.
Chorus chord loops were human-coded and checked by all coders for consistency. We defined a chord loop as any 2, 4, or 8-bar harmonic sequence, repeating at least once during the chorus, and where all the chord’s durations were the same. Any songs that did not adhere to this standard were not considered looping for our purposes. We used a modified form of relative-pitch Nashville harmony notation in the data coding, whereby the root note of the chord is defined according to an imagined major key signature (so C-F-G-Am would be 1-4-5-6m) and chromatic modifications were addressed with b (flat) signs (so Cm-Eb-F-Gm would be 1m-b3-4-5m). Chord names default to major if no suffix is used, and are minor if a lower case ‘m’ is added.
Lyric theme categories
In our experience of analysing pop songs and teaching songwriting, a small number of broad lyric themes are common across eras, styles, and countries. We have summarised these below, and included the criteria that we set for coding the BRITs SOTY songs. If themes overlapped (for example, a heartbreak song where dancing is mentioned) we used the chorus’s main thematic point for coding purposes. The majority of the songs (around 60%) dealt with themes of romantic love between two people; we sub-classified these by relationship status; Spring (the relationship has not yet begun), Summer (the characters are together), Autumn (still together, but having some issues), and Winter (the relationship is over). This was used as a more consistent coding method than subjective lyric interpretations (e.g. whether the relationship was happy or sad).
Lyric theme definitions for categorisation:
- LOVE (Spring): Looking for love: asking someone for love. The characters are not together yet, but at least one of them is thinking about it.
- LOVE (Summer): Being in love, and celebrating the other person. Everything’s fine and joyful.
- LOVE (Autumn): Love on the rocks: partners who are unfaithful, lyrics critical of partners, worrying about love. The characters are still together, but the stability of the relationship is at risk.
- LOVE (Winter): After love: loneliness, post-breakup heartache, love gone bad, baby take me back, missing someone after a breakup or death, or moving on after a relationship.
- DANCE PARTY: Non-romantic dance and partying themes, including songs about music
- SOCIAL: Political, poverty, social injustice/change, social commentary of any kind
- PLACES: Place names
- SELF-BELIEF: Believe in yourself, you can’t push me down etc. (first person e.g. Movin’ On Up, or second person e.g. Put Your Records On)
- SPIRITUALITY: Fantasy themes, or religious and non-religious spirituality
- HOME / FAMILY: Themes of homecoming, travel, leaving home, family
- CHARACTER: Songs about a specific (real or fictional) character
- MONEY: Songs about money (or the lack of it)
- YOUTH: Songs about being young, growing up, and getting older
- OTHER: An outlier/unusual theme.
Data analysis approaches
When the data was collected, we ran the following categories of queries:
- Whole corpus: what is the distribution of characteristics across the 329? (e.g. what were the most common lyric themes?)
- Winners: what is the distribution of characteristics among the winners? (e.g. how many winners started with the chorus?)
- Trends: what changes can we see over a given period (e.g. how have tempos changed over time?)
- 2D pivot trends: what changes show in data mapped to other data? (e.g. lyric themes against male/female vocals).
Studies of this type raise a larger question about songwriting universality – that is to say, what are the characteristics of songs that remain consistent across styles, decades, cultures and markets, and which characteristics change through context? Our research has attempted to explore this area with a manageably small corpus (four researchers analysing 329 songs over a 10-day period).
Further work might include: additional queries from the existing dataset; comparative queries with larger datasets; a larger corpus size (with greater use of Music Information Retrieval and/or larger-scale human encoding). Automated qualitative data extraction from lyric files may be possible, which would enable researchers to explore some of our early hypotheses about lyric universality – for example, that the majority of songs in any mainstream pop corpus deal with themes of romantic relationships. It would be interesting to test these speculations across cultures and repertoires.