Theoretical and computational work

Modelling early speech sound acquisition

To uncover whether human foetuses learn from their linguistic environment, it is crucial that populations from various language backgrounds are compared. Before a researcher decides to recruit and test near-term foetuses, or newly born babies, in at least two different countries, which is an extremely demanding task, it is crucial that she or he has specific, informed hypothesis about what is to be tested (and how). Computational modelling is very useful in this respect – it allows to test and re-test an infinite number of participants, in a relatively short time, from many varied language backgrounds.

In this project, we compare how a Czech-exposed and a Spanish-exposed fetus would learn the vowels of their language, namely, a e i o u, as well as their short and long variants. These two languages have partly similar phonemic inventories, which differ slightly in their acoustic realizations. Our results to date indicate that a Czech and a Spanish fetus differ in what they learn about the speech sounds of their language. Our model predicts that at birth a Czech infant will be better at discriminating a from o than a from e, while a Spanish infant will distinguish a from both e and o equally well. A paper reporting this virtual experiment can be found here, and will be presented at the virtual meeting of the Cognitive Science society in summer 2020. Further simulations are ongoing, stay tuned for the next results.

Our collaborator on the modelling project is Paul Boersma from the University of Amsterdam.