Skip to main content
European Commission logo print header

Human interaction and the evolution of spoken accent

Article Category

Article available in the following languages:

How human interaction sparks the evolution of spoken accent

Human accents are incredibly diverse, yet we don’t know exactly how they form. A new project used computer modelling and magnetic resonance imaging scans to find out more.

Society icon Society

If a group of individuals found themselves stranded on a desert island for a number of years, they would form their own unique ‘island accent’. We know about this phenomenon, but currently lack the data necessary to completely understand it. In the InterAccent project, which was funded by the European Research Council, researchers investigated how random, local interactions between individuals leading to group-specific spoken accents can push the sound patterns of languages between stable and changing states. The team sought to plug the gap in data, sourcing recordings of spoken language from children in rural and urban settings, from languages that differ markedly in their sound structure, and from groups of adults isolated together for several months during an Antarctic winter. “The optimal kind of data that we have recorded for InterAccent is longitudinal, that is speech recordings from the same individuals at multiple time points,” says Jonathan Harrington, professor of Phonetics and Digital Speech Processing at Ludwig Maximilian University of Munich and InterAccent project coordinator.

Sourcing spoken word and building a computational model

The project included re-recordings of ‘winterers’: scientists and support staff from the British Antarctic Survey who spent several months cut off from their home environment in Antarctica. “The Antarctica study was the first to confirm that a group of speakers even over a 3-month period begin to show a shared common accent,” notes Harrington. The team also recorded isolated words from primary school children in Albania and in southern Germany at yearly intervals over a 4-year period. They also recorded the movement of the tongue using ultrasound. Using these data, the researchers built a computation model that analyses recorded sounds, then replicates and analyses them to then determine how accents shift over time. The model uses pairs of software ‘agents’ based on real speakers that communicate with each other. The model then forms a link between imitation – something thought to be key to accent formation – and sound change. “Several experiments in the last 20 years have shown that we imitate each other subtly in conversation, even without realising it,” Harrington explains. “Just this subliminal imitation is the basis for sound change or new accent formation.”

Scans of the moving vocal organs

In collaboration with the Max Planck Institute, Göttingen, the InterAccent project also used real-time magnetic resonance imaging (MRI) to investigate the way two accents in contact over a prolonged period of time pull each other in certain directions. The team analysed vowel nasalisation in American English and Standard Southern British English – and showed for the first time evidence that Americans nasalise their vowels more than British English do. “We showed that American English has moved closer to a language like French in which the vowel is fully nasal and the following nasal consonant has been lost completely,” Harrington adds.

Scientific impact

The computational framework could be used to better understand how migration and exposure to other accents are changing the sound of our languages. It could also be used in a range of scientific fields to understand how interactions in the distant past such as climate affected spoken language. “The most important use will be in predicting the changes to speech sounds that take place when two dialects come into contact with each other or when one speaker migrates to a new dialect area,” he says.


InterAccent, accent, language, spoken, word, scientific, computational, model, vocal, organs

Discover other articles in the same domain of application