When predictions about the "what", "where" and "when" interact with statistical learning, from a behavioural and neural perspective.
MetadataShow full item record
As listeners we tend to detect patterns in what we hear regardless of whether this is music, language or abstract sounds. The fundamental underlying mechanism that supports this process is called statistical learning. Prediction lies at the heart of statistical learning, in the sense that accuracy of predictions influences learning, whether this is implicit or explicit. Numerous studies have investigated the predictive processes during statistical learning from a behavioural or neural perspective. However, it remains unknown how listeners respond to events that violate their predictions which are based on prior implicit learning. In other words, if exposure to a patterned sequence of sounds builds up specific expectations about upcoming sounds, then how does the brain respond to events that violate these expectations? This was the primary research question of the present thesis, and different aspects of the main question were tackled over three experiments. Experiment 1 examined how listeners respond to events that violate predictions that regard the content or the location of the sounds. We used a variant of an established statistical learning paradigm, and electroencephalography (EEG) as a means of measurement. Participants were exposed to a continuous auditory stream of sound triplets with deviants that were either (a) statistical, in terms of transitional probability, (b) physical, due to a change in sound location (left or right speaker) or (c) double deviants, i.e. a combination of the two. Therefore, statistical deviants violated predictions about the content of the stimuli, that is given the prior two sounds what will be the next sound. On the other hand physical deviants violated predictions about the sound location, that is given the location of the prior sounds from where will the next sound be played. Statistical and physical deviants elicited a statistical mismatch negativity (MMN) and a location MMN respectively, reflecting that participants’ prediction about the content or the location of the stimuli was violated. Most importantly, when a double deviant occurred, namely an unexpected sound from an unexpected location, participants’ sensitivity to unexpected “what” was reduced. Our results show that processing of physical properties (location) of the sounds suppresses processing of the structural properties (content) of the sounds. Experiment 2 further examined how predictions about different attributes of a stimulus interact with each other, and whether such an interaction affects learning. In that experiment the same experimental paradigm as in Experiment 1 was employed, except that the pause between the sounds was not constant anymore but varied randomly. In that way it became harder for participants to predict accurately the timing of the stimuli. By comparing the results from Experiments 1 and 2 we found that learning was impaired in Experiment 2. Temporal unpredictability due to random pauses reduced the neurophysiological responses to statistical and location deviants, as indexed by the statistical MMN and the location MMN. Our results suggest that when it comes to learning capabilities, the human brain requires isochronous, or perhaps at least regular, stimulation because processing of the “what” is tightly interconnected to the processing of the “when”. In other words, if we cannot make predictions on the “when”, it impedes the precision of our predictions about the “what”. The last experiment of the present thesis aimed to reveal the neural underpinnings of processing violations of predictions, using functional magnetic resonance imaging (fMRI). The experimental paradigm was simpler than the one used in the other two experiments because it presented only statistical deviants. Participants underwent a learning phase outside the scanner followed by an fMRI session. Processing of statistical deviants activated a network of regions encompassing the superior temporal gyrus bilaterally, the right deep frontal operculum including lateral orbitofrontal cortex, and the right premotor cortex. Our results demonstrate that violation of predictions about the “what” within a statistical learning paradigm, involved areas over different levels of the cortical hierarchy. Interestingly, processing of the occurring irregularities resembled the processing of local syntactic structures in language. Within predictive coding framework the human brain may be seen as an inference machine, yet our results show that the way inferences are produced may not be so mechanistic. We humans tend to perceive stimuli in their totality taking into account all their aspects, and perception of one stimulus aspect influences perception of another stimulus aspect. Perceiving, predicting and learning are operating together to help us capture the salient regularities in an uncertain world and survive in a constantly changing reality.
Has partsPaper 1: Tsogli, V., Jentschke, S., Daikoku, T., & Koelsch, S. (2019). When the statistical MMN meets the physical MMN. Scientific reports, 9, 5563. The article is available at: https://hdl.handle.net/1956/20559
Paper 2: Tsogli, V., Jentschke, S., & Koelsch, S. Unpredictability of the “when” impedes learning of the “what” and “where”. The article is not available in BORA.
Paper 3: Tsogli, V., Skouras, S., & Koelsch, S. Brain-correlates of processing local dependencies within a statistical learning paradigm. The article is not available in BORA.