The most common phrases in reels (on thesession.org)

The most common phrases in reels (on thesession.org)

I’ve been playing around with the session’s data outputs (trying to generate AI tunes but that’s another story). This lunch time I decided to see what the most common phrases are in reels. Rules are that it has to be a full bar, has to be from a reel and I’m getting rid of ornaments etc as best as I can. Well without further ado here is the top 16 list (each phrase followed by the number of times found.)

G2BG dGBG : 195
cAGE EDD2 : 170
d2fd Adfd : 165
d2fd adfd : 122
G2BG DGBG : 121
bgaf gfed : 111
gedB BAA2 : 110
A2FA DAFA : 109
defdc AFA : 91
defg afdf : 90
afge fdec : 89
dBAc BGG2 : 86
gedB GABd : 79
eaag a2ga : 77
E2BE dEBE : 72
dBAG FDD2 : 70

Any surprises in there, or anything you feel should have been in there? You can arrange them all into a tune but I won’t pollute the site by posting it! I was wondering where (3cBAe AfAeA stood, one of my faves, it has a count of 40, not as high as I would like.

Why did I do this? Well just like they say you can learn a language by learning the top 100 words, you can now learn irish reels more efficiently by learning the top phrases. (Joke.) Any other stat questions, I’m sure I can have a go!

Re: The most common phrases in reels (on thesession.org)

I couldn’t help but use this to try to find the most ‘generic’ reels. For each reel I am taking the number of times each phrase occurs in other reels, and dividing by the number of phrases, to give a score of how common it’s phrases are (balancing out tunes which are very long). Here’s the top 10:

The Killavil
https://thesession.org/tunes/1723/15150

The Lilies In The Field
https://thesession.org/tunes/1750/1750

The Humours Of Buckfast
https://thesession.org/tunes/2011/2011

Mama’s Pet
https://thesession.org/tunes/902/902

The Famous Ballymote
https://thesession.org/tunes/390/13226

The Old Pigeon On The Gate
https://thesession.org/tunes/5629/5629

Eileen O’Callaghan’s
https://thesession.org/tunes/1219/38945

Jack Broke Da Prison Door
https://thesession.org/tunes/1260/31417

McGoldrick’s
https://thesession.org/tunes/1360/1360

Green Garters
https://thesession.org/tunes/3458/16498

Thanks Jeremy for making all this data available!

Re: The most common phrases in reels (on thesession.org)

Nice idea and a pile of work but, without looking at the rest of the tunes, the selection you have for the Killavil Reel is just one part of the tune so, if you are dividing by the total number of bars, that is going to be more dramatically impacted.

Also, given that you have picked up that variation of that tune (as added to the website by Slainte, as it happens), then I presume your count of repeated phrases is also per setting of a tune rather than per tune. There therefore probably won’t be 195 tunes with the bar G2BG dGBG, but 195 settings of tunes within thesession.org database.

Re: The most common phrases in reels (on thesession.org)

Yep good points! - Like you’ve inferred I’ve been tallying per setting, so tunes with a lot of settings have added more to the count of each phrase. The other thing is that if a tune repeated a common phrase a few times throughout then its count was boosted, which also didn’t seem fair.

I’ve now corrected for those two things - I now only count a phrase once per tune (not per setting), and a tunes
score for the uniqueness of its phrases is based on the sum of its unique phrases, so it isn’t penalised for repetition or length (as far as can gather).

New top 10 tunes:

Green Garters
https://thesession.org/tunes/3458/32336
The High
https://thesession.org/tunes/44/44
Sporting Paddy
https://thesession.org/tunes/430/13292
The Swallow’s Tail
https://thesession.org/tunes/105/28532
The Curragh Races
https://thesession.org/tunes/1760/35509
The Lilies In The Field
https://thesession.org/tunes/1750/1750
After The Sun Goes Down
https://thesession.org/tunes/14581/26860
McGoldrick’s
https://thesession.org/tunes/1360/32699
Da Spirit O’ Whisky
https://thesession.org/tunes/7887/7887

Top 10 phrases - number of tunes:

bgafgfed 64
G2BGdGBG 64
cAGEEDD2 59
afgefdec 52
d2fdAdfd 51
d2fdadfd 50
dBAGFDD2 40
defgafdf 36
dBAcBGG2 35
(3Bcdefg2fg 33

Some really great tunes there. I’m sure theres other corrections to the process to be made too…

Re: The most common phrases in reels (on thesession.org)

This is an interesting project! Regarding the phrase "(3cBA eA fAeA" in your OP, is your algorithm taking only exact matches, or allowing for variance like sixteenth notes or the conversion of triplets into pairs of eighth notes? Two equivalences (parenthetical numbers are results in a tune search without setting deduplication) could be:
c/B/A eA fAeA (7)
cAeA fAeA (46)

It looks like you’re stripping spacing out, so the distinction of whether triplets are separate from adjacent notes appears to be covered. Would stripping capitalization distinctions out of the analysis also help? In the recalibrated analysis, two adjacent examples (d2fdAdfd and d2fdadfd) could be seen as equivalent, and both are often present in variant settings under the same tune.

Re: The most common phrases in reels (on thesession.org)

How about about the opposite? Least common phrases. Do they stand out in a tune and make a tune stand out as unusual?

I wonder if they will come from the numerous non-Irish tunes.

Re: The most common phrases in reels (on thesession.org)

@Bob Sturm - wow this looks exactly what I’ve been tentatively working on (Magento also), I couldn’t find it via google search, folk_rnn - much better results than I have had so far will have to look at how its all set up!

@benhockenberry - those equivalencies wouldn’t be processed as equivalent currently! This will require a bit more effort than what I’ve currently done, but I may have a go. Perhaps it would be worth converting single note triplets (or triplets that go up and doen) to crotchets, and like you suggest removing the middle note of other triplets…

@David50 - The problem with that is that there are so many tunes that are completely unique, so you can’t differentiate them in anyway further. I have though been able to correlate a tunes ‘uniqueness’ factor with its popularity on thesession.org (# times it appears in tunebooks). I find a negative correlation, people slightly prefer to tunes with more common patterns, but its hard to see whether this is just down to a large number of obscure and unique tunes - to be explored further…. I have found some tunes from picking the extremes of the ratio of uniqueness to popularity:

Popular tunes with common patterns

Drowsy Maggie
https://thesession.org/tunes/27

The High
https://thesession.org/tunes/44

Sporting Paddy
https://thesession.org/tunes/430

The Swallows Tail
https://thesession.org/tunes/105

The Fermoy Lasses
https://thesession.org/tunes/219


Rare tunes with common patterns (note: only tunes in 10+ tunebooks included)

The Lilies In The Field
https://thesession.org/tunes/1750

Green Garters
https://thesession.org/tunes/3458

McGoldrick’s
https://thesession.org/tunes/1360

Murphy’s Hound
https://thesession.org/tunes/4318

Da Spirit O’ Whisky
https://thesession.org/tunes/7887

Popular tunes with uncommon patterns

Pressed For Time
https://thesession.org/tunes/2589

The Derry Air
https://thesession.org/tunes/6867

Music For A Found Harmonium
https://thesession.org/tunes/346

Over The Waterfall
https://thesession.org/tunes/5353

The Honeymoon
https://thesession.org/tunes/2064

Re: The most common phrases in reels (on thesession.org)

jamesa - interesting stuff.

Just one small point - those phrases might need a key assigned.

eg the first one : G2BG dGBG …looks like key of G, but for the others, there could be some ambiguity.

Re: The most common phrases in reels (on thesession.org)

"The most common phrases in reels " - "How does that b*st**d start, again ?"

Posted by .

Re: The most common phrases in reels (on thesession.org)

Ha ha. Good one, Kenny.

Also, "Do you have a name for that?"

Re: The most common phrases in reels (on thesession.org)

The thing is, just take the top phrase

G2BG dGBG : 195

Each half of that phrase can be played in a large number of same-sounding/similar-sounding ways which could be notated in ABC differently:

G2BG dGBG
~G2BG dGBG
GGBG dGBG
(3GGG BG dGBG
(3GFG BG dGBG
Gz1 BG dGBG
G2 (3BAG dGBG
GABG dGBG
FGBG dGBG
G<G BG dGBG

it could go on much longer! And that’s just with the first half of the phrase.

Then each of these have to be multiplied by all the possible ways of playing the second half of the phrase.

So instead of 195 tunes using that phrase it’s probably several hundred, or a thousand.

The thing is, a group of players could be playing a tune that had that phrase, and all of them play all the permutations I wrote out above, and the overall sound would be of unison playing.

The same player might play that phrase several different ways as he/she goes along.

But unless the ABC search engine has changed, it can’t recognise same-sounding phrases, only identically-typed phrases.

Re: The most common phrases in reels (on thesession.org)

"How about about the opposite? Least common phrases."

For sure it seems like when people are playing "the hot new reel" and it’s merely familiar sections of a couple other well-known reels put together in a slightly, very slightly, different way, I do wonder why bother.

Then there are tunes with quirky phrases, "crooked" tunes, that don’t readily fall under the fingers.

Highland pipe tunes often have phrases like that, here’s the start of a Scottish jig:
6/8, key A Mixolydian
| e2A GAA | dcG AAA | e2A GAd | ccc dcd |

Re: The most common phrases in reels (on thesession.org)

Re: Richard’s comment

"Each half of that phrase can be played in a large number of same-sounding/similar-sounding ways which could be notated in ABC differently"

You could probably build a preprocessing step to standardise the ABC and get rid of/replace ornaments, triplets, < and > signs etc. I did something like that with Java years ago to clean up cluttered transcriptions and give a ‘bare bones’ version of the tune. You could also build octave sensitivity (or insensitivity if that made sense) within that preprocessing.

Wouldn’t it be interesting to see if a clustering algorithm could show you clusters of similar tunes based on common phrases? and by extension show the outliers in the tunes database.

Re: The most common phrases in reels (on thesession.org)

Thats a really interesting series of blog posts Bob Sturm, I wondered if the DL-distance incorporates only single character edits or copy and pasting of phrases/sections too? I think what you’ve done to address the problems with the similar phrases would work well - I might try and do it on mine. 🙂 Would be interesting to see your analysis with reels too, I feel that reels have more signature patterns somehow, but if DL-distance doesn’t allow chunking it might not capture it?

This all started for me because I learnt the Pope’s toe, followed by The High followed by O’Rourkes…. (All of which extensively features cAeA fAeA)

Re: The most common phrases in reels (on thesession.org)

Thanks @jamesa. The DL distance on ABC is very musically naive. If we have two strings, "CEG" and "CEG", but one is in the context of C-major and the other in C-minor, the DL distance between them will be zero even though it shouldn’t be. Also, if we have two strings "CEG" and "C/D/E/F/G", both in the context of the same mode, their DL distance will be 6, even though harmonically they are equivalent. It’s like using the DL distance between two news stories to determine if they are about the same topic. That’s why in later blogposts of that series I move to time-interval representations that are 8-bar sections, accounting for any anacrusis.