Nature: Tunes create context like language

Whistle
1

http://www.nature.com/nsu/040614/040614-11.html

I’d say the article is a bit overblown. This guy Dr. Zanette simply found that the notes in classical music follow a Zipf distribution. Words in literature also follow a Zipf distribution. Therefore music is like language??

I got news for the folks at Nature: the air pressure in car tires follows a bell curve, and so do IQ scores. Therefore brains are like radial tires on some deep level.

The article elaborates that people must like tunes that return frequently to certain notes, creating a kind of context. Again, no big surprise there: any book on music theory or composition will tell you that. This is why we call one of the notes a “tonic.”

BTW, I could run the same experiments on Irish music. I have some software I wrote that chops up tunes from ABC archives and computes various statistics, whatever I want—relative frequencies of keys, or notes, or pairs of notes, which fingers are used more often when playing the concertina, etc. Let me know if you have any IRtrad- or whistle-related questions that could be aided by some number-crunching.

Caj

2

Nice one Caj.
That sounds really interesting, do you know anything about Eigenfunctions ? I understand that it is possible to make an Eigenfunction of the Human face such that changing one variable changes the face from male to female and visa versa. It would be great to have Eigon functions of say Scotish/Irish, Jig/reel, to play with. I think the problem might be representing the time factor in the music as apposed to a static 3D object.

John S

3

I do have a question, what three and four note combinations are most common in popular slow airs? In writing my own music, I find that three note and four note combos tend to have a lot of impact at a slow tempo. Again, I am most intereted in analysis of slow airs, but analysis of fast dance tunes might also prove interesting.

The way I might approach the question from a programming standpoint is to generate a hash table value for each unique sequence found and keep a count.

  • Bill
4

Just to play the devil’s advocate (or not :smiley: ), folk musics in particular have recurrent formations and tendencies that make them what they are, which is arguably not unlike language. The combinations of these formations either make sense or they don’t, given the context of the form. Any one who has seriously attempted ITM compositions would be aware of this.

But I’m no mathematician, or a delver into logic.

5

Alas, this is one question that can not be answered by analyzing ABC files! I don’t have any ABCs of laments or slow airs. Such things are generally not written down. Even when they are, they are uncommon enough that perhaps it is better to analyze that by hand.

Caj

6

I’ve heard about the faces, so-called “eigenfaces”; basically they found a small set of “basis faces”, and a lot of human faces can be expressed as a sum of those basis elements.

I think they shouldn’t have called them eigen-whatevers, because in mathematics an eigen-whatever (eigenvector, eigenfunction, etc) is something different. It sounds catchy, but it’s a bit of a misnomer.

An eigen-function is a function that stays more or less the same under some kind of transformation. For example, you play an A on a whistle, and in the next room over I hear a muffled sound—but the pitch is still the same. The wall between us can’t turn an A into a Bb. This is because a pitch is a bunch of pure tones, and a pure tone is an “eigen-function” of the wall’s filtering process. The wall might muffle a tone, or delay it, or completely filter it out; but the tone’s frequency, it’s actual shape, cannot be changed.

Caj

7

There are many available at the tunefinder.
http://trillian.mit.edu/~jc/cgi/abc/findtune

Get a list of titles from a book or website, and in a short time a person can have a large batch of ABCs to analyze. There also may be large collections of ABCs in single files, but I don’t have a link handy. ABCs often include a line indicating the type of tune, e.g. jig, reel, hornpipe, aire.

The precise rhythm will be missing because of the nature of the music, however, the sequence of notes will be there. To me looking for note combos appears to be a straightforward project, especially if you already have a parser coded. It’s been a while since I have done any coding, but the approach I might use would be to assign a unique numeric value for each combo, and use that value as an index into a large integer array that keeps counts. Increment each array element as each combo is found by the parser. Many modern programming tools allow very large arrays, so this approach is to get the project done quickly. Processing time is probably neglible given the speed of modern PCs, and the relatively small size of ABC files.

I hope you are kidding about analyzing by hand. I believe that even doing eight short songs by hand would be an extremely time consuming task. A person might quickly be able to discern the hooks (the most memorable phrases) but unless extremely thorough and detached would probably miss many common combos because of emphasis and rhythm.

You asked for questions, and I did have one that applies directly to number crunching on ABC files. I hope for some analysis, but understand that each person has time and resource limitations. If you need more clarification as far as a basic algorithm, data structure, or more help with data sources, let me know. At first glance it looks like something that an efficient programmer could knock out in a day or so, given some decent tools. I do understand that often times small programming projects take on a life of their own and grow into big ones. At this point in my life, I no longer have access to decent programming tools nor the desire to seek them out. And like I said its been a long time since I have done any coding, so I am far from that standard of being an efficient programmer.

  • Bill
8

(I got carried away. The following is probably unreasonably picky and uninteresting. Still…)

I found the Nature summary rather confusing, whereas the original article is pretty clear.

However, I tend to be wary of generalizations about language. Here’s a quote that seems to me to overstate its case:

“In language, context emerges from the mutually interacting meanings of words. As new words are successively added to a text or speech stream, context is built up by the repeated appearance of certain words or word combinations, by the emphasis on some classes of nouns and adjectives, by the choice of tense, etc. These elements progressively establish the situational framework defined by the message in all its details. Thus, linguistic context is a collective expression of the semantic contents of the message.” (Italics mine.)

Everything leading up to the last sentence is reasonable, but the last sentence is an over-simplification of real-world language, because it ignores extra-linguistic elements (physical situation, culture, etc.) in the linguistic context. One of the most difficult parts of translation is getting across those things that are implied by elements outside the language proper–especially cultural factors that may not be obvious to your target audience. On top of that, it ignores syntax altogether, as though we could equate “The dog bit the man”. and “The man bit the dog.”

Here’s another over-simplification:

“In language, words –or short combinations of words– stand for the units of semantic contents, with (almost) unequivocal correspondence with objects and concepts. Moreover, in the symbolic representation of language as a chain of characters, i.e. as a written text, words are separated by blank spaces and punctuation marks, which facilitates their identification –in particular, by automatic means.”

The boy apparently hasn’t spent much time in arguments on sci.lang and sci.lang.japan about just exactly what “word” should apply to–and about whether there is a single concept indicated by “word” that can be meaningfully applied to all languages–or even to any language. (A very simple way to make this point is that in English writing, “bookstore” is rendered as one word, while “grocery store” is written two words, but neither way of counting words is made unambiguously necessary by any facts about the structure of the English language.)

The statement about writing is just plain wrong. Even if we could say that Chinese languages have “words”, they are certainly not distinguished as such in the writing system. (Chinese characters equate to syllables, which do not consistently equate to anything that we can call words.)

Japanese can also be written with long strings of kana in which “words” are not distinguished graphically. In fact, I think it was R.H. Blythe who pointed out an entire paragraph from The Tale of Genji, written entirely in hiragana, which is so ambiguous in terms of word breaks that it can be read either as a description of the onset of Autumn, or as a description of the emotional effects of a romantic break-up–each interpretation reinforcing the mood of the other. (And done entirely on purpose, of course.) How shall we count words in such a piece of text? This may be an extreme example, but it does show that ambiguities exist.

So, context in language is much more subtle and ambiguous than is implied by subjecting it to word frequency counts. This is not to say that such analysis is not interesting, but only that it may not provide any deep revelations about the nature of language.

The application of the same style of analysis to music seems even less exciting, somehow. If language doesn’t really have a clear-cut natural unit (the word) to which such analysis can be usefully applied, music is even more lacking. I’m not sure that the final result cannot be summed up as simply, “We use context in judging sensory input.”

9

Very astute observations on The Tale of Genji.

Sanskrit apparently suffered from the same problem (it has been theorized that hiragana and katakana were invented by a monk who had Sanskrit on the brain), although apparently, no one knows if such ambiguity in Sanskrit was intentional or not.

10

Thanks for Eigen whatsit info.
One thing I’ve tried with many part tunes is to play all the parts simultaneously in Cubase. Sometimes this just sounds a mess but sometimes you learn something fundamental about the melodic structure.
To my ears some of the tunes seem to have a sort of conversational psychodynamic which requires me as the player to animate the protagonists.
Unlike spoken language however both question and answer have coupled meaning and cannot be treated as completely separate.
Music of cause operates in the sphere of the emotions directly without the coding of words.

John S

11

Something close to this very observation occurred for me last evening. I’d been thinking about the things being discussed on this thread (except for the Eigenwhatzits and Fortean algorithms or whatever–we all know what I’m not. :smiley: ), and played an air, The Banks of the Moy, for my cat. She did the usual, just settled down and closed her eyes and smiled the way cats are wont to do when they’re contented. Basically I just ran through it the way I’ve been doing it per usual, sort of on autopilot, but it wasn’t devoid of expression. Then I played the air again, but “conversationally”, if you will (or more properly, I thought of the attempt as trying to somehow imitate a persuasive monologue), and my cat, who loves being spoken to, got off her chair and writhed her pleasure on the floor. IMHO the air was improved for it, too, I have to say.

I don’t know how to describe the result, but I can say that the phrasing became more pleasing to my ear, more expressive. For me, the presence of mind that the new-and-improved run required was great. It was easy to slip back into autopilot, and that’s when Mubu would lose interest. I wasn’t present to what I could have been “saying” through the tune, so nothing got “said”, and the moggie stopped listening too until I re-became totally present to the tune and its parts, resumed “saying” things, and the cat would writhe again. I also found that overdoing the “expressive” thing, sounding excruciatingly mawkish, happened as soon as my focus slipped. I was fascinated. I’ve now got a new goal in increasing the level my presence of mind in service of the tune at hand. In other words, not being so lazy sums it up. Another goal would be to tone down using quotemarks. People might suspect me of being someone else. :wink:

This isn’t to suggest that language and music are fundamentally one and the same, but a conversational approach to playing definitely has an effect on how a tune unfolds.

Side note: I was discussing this thread with a friend, who rejoindered with, “Music IS the universal language.” He’s never heard Japanese court music before. :laughing:

12

Or some of that Buddhist stuff that sounds like they’re torturing cows… :sniffle: