Skip to main content icon/video/no-internet

Music involves patterns of sound; without pattern, there would be no music. Statistical learning is one approach to understanding how humans derive patterns from perceptual input, based on the underlying probabilistic properties of transitions between one element in the perceptual stream and the next, and also between nonadjacent elements. These approaches were developed for understanding children's linguistic development, but have also been applied to music, looking at melodies, harmonic progressions, and other stimuli.

Imagine a newborn infant hearing language. Sounds cascade around her, produced by the constantly changing dynamic articulatory system of the human vocal tract. These sounds contain some elements that seem to recur, but never sound exactly the same twice, and are embedded in ever-varying contexts. How can the child make sense of this stream of sounds? The approach of statistical learning is to look at the probability of one sound following another, and use these differing probabilities as guides to inducing rules as to which sounds belong together. This could result in phonemes judged as belonging to one syllable or syllables belonging to one word because the probabilities of them being adjacent to one another in a particular order are relatively high. These high-likelihood pairings would be opposed to the low-probability transitions, indicating segment boundaries of some kind, such as those that separate one word from the next. Probability theory is applicable not only to language, but also to music.

Researching the Phenomena

The typical way that such phenomena are studied is through constructing small artificial grammars, the transition probabilities of which are precisely controlled. Participants are first trained on stimuli constructed with these grammars, and are then tested with grammatical and nongrammatical sequences to determine whether or not the statistical properties of transition probabilities in the grammar have been learned, so that sequences can be correctly categorized as grammatical or not. Such studies have demonstrated that 8-month-old infants learn and make use of such statistical information in making use of speech sounds; adults show similar abilities.

With these findings in regard to language as heard, the potential parallels to learning about musical structures were obvious to researchers. Melodies were the first musical materials to be studied; because they employ sequences of single sounds, much like speech, they provide a good parallel. Musical tones were combined into small units called “tone words” by the authors—musicians would call such units motives, figures, or gestures—whose inter- and intraword transition probabilities were the same as those in the earlier speech stimuli. Results statistically indistinguishable from those of the earlier study were obtained with both infants and adults, suggesting that inducing regularities from the transition probabilities was a general property of human auditory cognition, underlying both speech and music perception.

Later studies have extended this approach in a variety of ways. For example, researchers have examined listeners' ability to learn a new kind of musical idiom by constructing melodic stimuli in a newly created microgrammar, using a scale quite different from any to which their adult listeners would have been exposed and enculturated. After about 30 minutes of passive exposure, participants not only exhibited learning of specific melodies, but also generalization of the grammar, distinguishing between novel grammatical melodies and lures, using a similar but different grammar.

...

  • Loading...
locked icon

Sign in to access this content

Get a 30 day FREE TRIAL

  • Watch videos from a variety of sources bringing classroom topics to life
  • Read modern, diverse business cases
  • Explore hundreds of books and reference titles

Sage Recommends

We found other relevant content for you on other Sage platforms.

Loading