Aya Gomaa will lecture on "Representational Similarity Analysis on Prediction in Language Comprehension"
Abstract: Predictive processing stands as a cornerstone of human cognition, profoundly influencing various activities, including language comprehension. Over time, researchers have employed diverse methods to delve into predictive mechanisms within language comprehension, ranging from behavioral measures focusing on eye-tracking measures to electroencephalography (EEG) studies, particularly focusing on the N400 component. Recently, the advent of the Representational Similarity Analysis (RSA) framework has offered new avenues to explore the timing and nature of prediction (pre-activation) signals. Notably, RSA has primarily been applied in conjunction with EEG data in language comprehension studies. One critical aspect yet to be explored is how the number of channels used to quantify the representational signal of the RSA can impact the observed similarity across different conditions. For example, RSA is a technique that correlates spatial distributions of representations across the scalp, enabling the discernment of similarities between representations of different stimuli. By harnessing the power of EEG, by varying or increasing the number of EEG channels, we can enrich the spatial distributions that RSA operates on, thereby enhancing its efficacy.
To address this gap, we embarked on a two-fold investigation. Firstly, we sought to replicate key findings from a previous study by Hubbard and Federmeier (2021). By comparing EEG activity patterns elicited by pre-final and final words in sentences with varying levels of constraint, we observed a peak similarity in an early time window following the presentation of the pre-final word. This peak similarity varied based on the constraint level of the sentences. This suggested that rapid pre-activation occurs following certain cues. Subsequently, we explored the effects of varying the number of EEG channels from 30 to 61 channels. Interestingly, we noted an increase in the peak of the similarity signal across both strongly and weakly constrained conditions. These findings suggest that the maximal power of RSA can be harnessed with an extensive array of electrodes. However, further investigations are warranted to determine the ideal or minimum number of electrodes required to observe the maximum signal. Overall, our study sheds light on the potential impact of spatial dynamics on predictive processing in language comprehension.
