Towards an LLM-based method for quantifying the sexual content in song lyrics
This paper introduces a reproducible large language model-based method for quantifying thematic content in song lyrics and applies it to a corpus of 1,259 reggaeton tracks to analyze sexual explicitness across artists, time, and against Spotify's explicit flags.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine the world of music as a giant, noisy library where every song is a book. For a long time, when people wanted to know what was inside the books of a specific genre called Reggaeton, they had to read them one by one, taking notes by hand. This is like trying to count the number of red cars in a city by standing on a street corner and writing them down as they drive by. It's slow, it's tiring, and you can only look at a few cars before you get tired. This field of study is called "content analysis," and it's how scientists try to turn feelings and themes—like love, anger, or partying—into actual numbers so they can compare them. The big question everyone has been asking is: "Is Reggaeton really all about sex, or is that just a rumor?" To answer this, we need a way to measure the lyrics without reading every single one of them manually. That's where a new kind of helper comes in: a super-smart computer brain called a Large Language Model (LLM). Think of an LLM as a robot that has read almost every book in the library and can instantly tell you, "This story has a lot of romance," or "This one is full of violence," giving it a score from zero to ten.
This paper is like a recipe for using that robot brain to measure the "spiciness" of Reggaeton lyrics. The author, Ignacio Sticco, built a method to feed 1,259 songs from 12 famous artists into the computer. Instead of just asking "Is this song dirty?", the computer was taught to look at nine different flavors of the song, like "party vibes," "sad love," "street crime," and two specific types of "sex talk": one that hints at it (suggestive) and one that says it directly (explicit). The robot was first trained on a small group of songs that humans had already graded, so it learned to give the same scores a human would. Once the robot was ready, it graded the whole library of songs, creating a massive spreadsheet of data that shows exactly how much of each theme appears in every track.
The results are a bit surprising and definitely not what you might expect if you just listen to the radio. First, the study found that while Reggaeton is indeed full of sexual themes, the "hinting" kind (suggestive) is actually twice as common as the "blunt" kind (explicit). It's like finding that most of the party songs are about the idea of a wild night, rather than describing the wild night in graphic detail. Second, the artists are not all the same. Some, like Plan B, are like a spicy pepper, scoring very high on direct sexual content. Others, like Camilo, are more like a sweet dessert, with lyrics that are almost entirely about romantic love and barely any sexual talk at all. The study shows that treating all Reggaeton artists as a single group is a mistake; they are as different from each other as a heavy metal band is from a jazz trio.
The paper also looked at how things have changed over time, from 2002 to 2025. Contrary to the idea that the genre is getting "wilder" in every way, the study found that the "street crime" theme is actually fading away, while the "romantic emotion" theme is growing stronger. However, the "directly explicit" sexual content has doubled over the last two decades. This means the genre isn't just getting more suggestive; it is getting more direct in its language, but it's also becoming more emotional and less about crime.
Finally, the author checked the computer's work against Spotify's own "Explicit" label (the little "E" you see next to a song). The computer found that Spotify is missing a huge chunk of the explicit content. Out of every four songs the computer identified as having explicit sexual lyrics, Spotify only flagged one of them. It's like a security guard who only checks for big, loud weapons but misses the smaller, sharper knives. This suggests that the current way streaming services label music is too blunt an instrument to catch the specific kind of sexual content found in these songs.
In short, this paper doesn't just say "Reggaeton is sexy." It uses a clever computer method to prove that the genre is a complex mix of romance, partying, and direct sexual talk, and that it has changed in specific ways over the last 20 years. It also shows that the tools we currently use to label music (like Spotify's flag) are not very good at catching the specific details that researchers and listeners might care about. The author has shared the code and the data, inviting anyone else to use this same robot-brain method to study other music genres or even different languages, turning the slow art of reading lyrics into a fast, measurable science.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.