The role of prediction in language development: perspectives from neuroscience

Many cognitive scientists believe that children and adults are able to predict what the next word in a sentence will be, allowing them to process and respond more rapidly in conversation. This observation has come to influence many key theories in cognitive science. One such theory is the error-based theory of language acquisition. It suggests that people always compare their predictions to what they hear, and use any mistakes to update their knowledge, making predictions more and more accurate. This theory is highly influential as it can explain puzzling linguistic phenomena such as how children learn to produce the correct plural of 'mouse' ('mice' rather than 'mouses') on their own. However, there is surprisingly little direct evidence that listeners' incorrect predictions actually play a role in language learning. My research aims to provide concrete evidence for the existence of such a mechanism, and by observing it in action, better understand how and why it works.
My PhD reported three sets of studies seeking evidence for this theory by developing new experimental approaches. While we found some strong support for error-based learning, other results cast doubt on the nature of the mechanism proposed by this theory. For example, when we assessed whether adults show prime surprisal (increased likelihood of repeating surprising structures, which is often taken as evidence for the error-based learning theory) we found that this effect only appears with certain structures, suggesting that these mechanisms might not operate under all circumstances. We also used neuroimaging to measure the brain's electrical activity while adults listened to surprising or predictable sentences. This study showed that there was a greater processing cost while listening to unpredictable sentences, but no evidence that this difficulty is related to failed predictions.
The current fellowship would allow me to consolidate these results and increase their impact. I plan to carry out three sets of activities to achieve this: engaging with new audiences, building academic impact, and developing the skills to carry out neurolinguistics research with young children.
Public engagement: The fellowship will allow me to utilise the ESRC LuCiD Centre's extensive research dissemination network. This collaboration between the Universities of Liverpool, Manchester and Lancaster not only conducts state-of-the art research on language acquisition, but also provides evidence-based advice about how best to foster language development in preschool children, and has direct links to a wide range of non-academic beneficiaries.
Academic impact: While the first two sets of studies have already been written up and presented at conferences, the third dataset requires further work to allow me to publish it as a high-impact journal article. This dataset, resulting from our neuroimaging work, has important implications for error-based theory, as it questions whether active predictions or some other processing difficulty drives learning. However, the novel methods developed for this study mean that we need to carry out further analysis and some limited additional data collection at the original site, which was the University of California, Davis. By maximising the potential of this study, I will lay crucial groundwork for my subsequent career plans.
Developing skills: My proposed mentor Dr Perrine Brusini is an expert in neurolinguistic research with young infants and will provide training in the software and methods necessary to work with this population. By the end of the fellowship I will be in a position to carry out my own infant neuroimaging study, under my mentors supervision. This will not only broaden my skillset but also provide pilot data for an ESRC New Investigator grant proposal aimed at investigating the role of predictive coding in infant language learning - work that will allow us to better understand the processes that enable language acquisition.