Modelling the dynamics of phonetic variation and phonological change

One of the most fundamental observations about language is that it changes over time. An English speaker from the 1950s sounds a world away from how a young person speaks in the 2020s. At the same time, we would recognise that both people are speaking the same language. This duality represents a key characteristic of language: the constant pull between stability and change. In this fellowship, I will develop a new account of how change in pronunciation develops out of the variation that exists between speakers, grounded in knowledge about the physical movements of the vocal tract during speech.

Decades of research have uncovered detailed knowledge about how the vocal tract moves, how this varies between speakers, and how variation can lead to sound change. Despite this, there are fundamental gaps in our understanding about how these different elements are linked together. My project will unite these elements into a single computational model using insights from mathematical research in dynamical systems, which has proven to be an immensely powerful approach to understanding complex interactions in the natural world. This will allow us to make clear predictions about how detailed aspects of speech can vary and how this variation could lead to the sound changes of tomorrow.

This model will be applied to two case studies. The first will track variation in the production of vowels in two dialects: Standard Southern British English and West Yorkshire English. I will test whether the model can capture this real-world variation and accurately reproduce differences between dialects and speakers. I will then turn to one of the best-known sound changes in the history of English: the Great Vowel Shift, which happened between the 1400s and 1700s but still has lasting consequences for today. Before the Great Vowel Shift, the word 'bite' would have been pronounced like modern-day 'beet' and the word 'out' would have rhymed with modern-day 'boot'. The change between these older pronunciations and their contemporary versions is called diphthongisation.

Diphthongisation over long timescales has proven to be very challenging to understand, because it involves a vowel with a single tongue position, such as 'bee' (one tongue target), transforming into a vowel where the tongue moves from one position to another, such as 'buy' (two tongue targets). Adding a target seems like a dramatic change, because it requires an additional element of planning the tongue's movement. However, we often see that an additional target may be variably present when we compare different accents of English or different speakers of the same accents. This raises a question: how do speakers change a one-target vowel into a two-target vowel? Is it a gradual process, or a more abrupt change where a second target is suddenly added? These questions raise major issues about how sound systems work, such as how variation between speakers becomes long-term sound changes. Indeed, these are some of the most challenging questions in linguistics because they span such diverse timescales, from the hundreds of milliseconds it takes to produce a vowel to the hundreds of years over which a vowel's pronunciation can change.

I will use the new computational model to simulate how the above sound change might happen. This will start with a scenario based on what the start of the Great Vowel Shift looked like and simulate what might happen if speakers interact many times, allowing their speech to be influenced by those around them. I will track the evolution of sounds over these interactions to identify the potential ways in which the Great Vowel Shift might have occurred. I will reconstruct various stages of the sound change and compare this to real data to assess the plausibility of the model's predictions. Finally, a series of workshops and online tutorials will show other people how to use the model and allow them to test their own theories about the nature of sound systems.