Dr Ediz Sohoglu
Predictive coding and speech perception
Research in my group is focussed on the neural basis of auditory perception and spoken language processing. A defining property of sound is its dynamic nature. Unlike the static stimuli often used to investigate the visual system, sounds (by definition) change over time. This temporal property of sound and the neural processes that support its representation make auditory neuroscience a fascinating area of research. As well as this, two sensory signals that mark us out as a human species – speech and music – are perceived primarily through the auditory system.
Much of my research is motivated by the influential idea of predictive coding. That is, rather than passively processing auditory input, the brain actively generates predictions (or guesses) in order to make sense of the sounds that we hear. While there is broad consensus that prediction (of some kind) supports perception, the precise details of the underlying neural implementation remain unclear (Aitchison and Lengyel, 2017). Revealing these details is a central goal of the research in my group.
This PhD project would acquire and analyse EEG recordings of cortical activity as human volunteers listen to speech. We will capitalise on the fact that certain points in the speech signal are more or less predictable. For example, hearing the sequence of phonemes “snoo...” sets up a strong prediction for “…ker” because “snooker” is the most frequent word that follows (compared with “snooper”). By using recent neuroimaging analysis methods (Brodbeck et al., 2018), we can relate this variation in predictability with the ongoing EEG recordings and adjudicate between alternative computational implementations of predictive processing.
Other avenues of research could be to examine the relationship between prediction and attention (Sohoglu and Chait, 2016) and how predictive processing adapts over time to cope with noisy or degraded speech (Sohoglu and Davis, 2016).
Students will acquire skills in EEG, experimental design, Matlab programming, computational modelling and sound manipulation/analysis.
Key references
- Aitchison L, Lengyel M (2017) With or without you: predictive coding and Bayesian inference in the brain. Curr Opin Neurobiol 46:219–227
- Brodbeck C, Hong LE, Simon JZ (2018) Rapid Transformation from Auditory to Linguistic Representations of Continuous Speech. Curr Biol 28:3976-3983.e5
- Sohoglu E, Chait M (2016) Detecting and representing predictable structure during auditory scene analysis. Elife 5
- Sohoglu E, Davis MH (2016) Perceptual learning of degraded speech by minimizing prediction error. Proc Natl Acad Sci 113:E1747–E1756.