Precision, recall and musicality


a man carrying a pile of documents
Listen to this article
Share it like your embrace

It takes one

A few weeks ago I asked in the Facebook group how you can someone tell if a dancer is musical and what that means. Among the expected answers that a musical dancer can distinguish and dance the rhythm or the melody of a song etc., there was one that I liked very much. The summary of it is that it takes one to know one. To judge someone about their musicality you need to know about musicality and understand what they do.

Frequently, in discussions with friends, they tell me that they like such and such dancers because they find them very musical. When I ask for more details on why they think so, they have a very hard time explaining what exactly it is that they do. The answer is usually quite vague… Something like… “They follow the music” etc. If you would ask me some years ago, I would ask in the same way. Today I can articulate much better what exactly I like and what I don’t in terms of musicality in a dancer.

So the question that arises is: are we really able to spot the musical dancers if we don’t know about musicality? Can we be tricked? Is it an intuitive or instinctive process that guides us or do we think that something is musical but in essence, it is not so much? And in the end, why do musical dancers tend to say that less is more?

Search engines

In IT there is a whole domain of the science called information retrieval (IR). It has to do with researching ways to improve the performance of search engines. In this domain, there are two main metrics for evaluating different methods of searching for documents in a large pool of documents. One is called “precision” and the other “recall”.

In order to explain what each one is, let’s assume that you have a collection of 100 electronic documents. You know beforehand that in this collection 20 documents are relevant in some way to the word “Tango”. So if someone searches for “Tango” in those 100 documents the perfect result from your search engine would be those 20 documents and nothing more. But your engine returns 30 documents. 15 of them are between the 20 relevant ones and the other 15 are not.

The precision is the number of relevant documents found divided by the total number of documents found. So in this case 15 / 30 which means 50% precision. The recall is the number of relevant documents found divided by the total number of relevant documents in the collection. In this case, it would be 15 / 20 which makes for 75%.

Measuring musicality

Imagine that instead of documents you have musical notes played at specific points in time. If your dance is the search engine and you aim to dance to something specific (e.g. the rhythm, the bandoneon etc.) then out of those notes and time points only a specific amount is relevant to what you aim for. The rest is irrelevant. So let’s say you have 100 notes/moments and 20 of them are relevant to your aim. You step/move in 40 out of the 100 and the 20 relevant notes are in between those 40 you step. Let’s see your performance.

In terms of recall, you managed to step on all 20 relevant notes so your recall is 100%. But your precision on the other hand is 20/40 which means 50%. As you see the more notes you hit the lower the precision will get. So more steps less precision. In terms of recall, you can easily get to 100% as long as among the notes you pick, you get the 20 relevant ones. But to get your precision high you need to do fewer steps… Not more.

That is why “Less IS more!!!” in musicality.

Confirmation bias and intention

That’s all nice but then why do we still find musical performances the ones where the dancers step on as many notes as possible? How can we not see that the precision is low? My explanation is that most probably it is based on some confirmation bias. For example, I listen to a song and I get attracted by the piano while someone else is attracted to the bandoneon. If the dancer manages to hit a good amount of notes in both instruments then we will both think that they were dancing to what we paid attention to and neglect the other steps. For me, the steps on the piano would verify that they followed the piano… For someone else, the steps on the bandoneon will verify that they were dancing to the bandoneon. But the critical question is what was the actual dancer aiming for? If they were aiming for both then they might be indeed musical. If they were aiming for the bandoneon or the piano alone then their precision is actually low but because we are primed to neglect the irrelevant steps when we see the dance we both think they were nearly perfect when in reality they were not musical at all.

Real musical dancers have very specific intentions in every part of the music. They have decided on each phrase on which element they will focus and they will remain on that element. Therefore, they will do only the steps/moves that this element dictates. Not more… Not less. Real musical dancers don’t get 100% only on their recall but on their precision as well. Dancers who dance to many elements together and don’t have a focus will look musical to the untrained eye, but for someone who distinguishes musical elements in their mind, they will seem quite boring. Add to this the element of awe and impressiveness from complex moves performed and you are so dazzled that in the end, you think you saw an amazing show when in reality it was nothing like this.

The prerequisite

When I started writing about precision and recall I wrote that we assume that we know that in a collection of documents where we know that some specific documents are related to a keyword that we search with. In information retrieval, there are such collections of documents with an already existing set of keywords and their predefined correct answers so that researchers have a common baseline to run their experiments on. The same stands for Tango songs too. Each song contains several moments and each one is associated with specific aspects (e.g. rhythm, specific instrument, etc.). The musicians already know the answers for each aspect.

To identify a musical dancer you need first of all to know about the different aspects of the music and which moments in a song are associated with each one. From the distinction between rhythm, melodies, countermelodies etc. to the way that different instruments can be played, the sound they produce and what kind of sensation they create. It’s a never-ending exploration and discovery. If your mind can analyze those aspects in a song then it’s like you have your pool of documents and you know which ones are relevant to each specific aspect. It’s only then that you can understand what the dancer’s intention is and judge correctly how good is their precision and recall. It’s only then that you can truly judge the musicality of a dancer.

Tonight’s Goodnight Tango

In tonight’s Goodnight Tango the singer is asking his partner to think about their next step. He is challenging them because it might be a mistake. They might never see each other again. Take this lyric as advice next time you dance and you pay attention to the music. Think about your next step. Do you really need to take it? Is it really the correct moment? Is it helping you increase your precision? Or will it reduce it?

How about you? Are you able to recognise the different aspects of the music? Can you see the difference between dancers with high precision in their dance and the rest? Do you think you know enough to judge the musicality of a dancer? Or do you think that musicality is something more that cannot be measured simply with metrics like precision and recall?

Did you like the post? Spread the word…

Do you have something to say on the topic?


One response to “Precision, recall and musicality”

  1. Lj Avatar

    I like the way you merge your idea!

Leave a Reply


Skip to content