Speaker
Description
Human brain employs sensory inputs to predicts future events [1]. Research has delved into linguistic expectations and alpha oscillations (8-12Hz), where alpha power modulation was linked to anticipatory mechanisms [2]. Studies that manipulated contextual constraint revealed stronger predictions associated with lower pre-stimulus alpha power than weaker predictions [3]. Yet, such results could reflect specific processes of the controlled manipulations. Here, we investigated the alpha power fluctuation in a naturalistic comprehension context. We recorded cortical activity of 25 English participants who listened to 45 minutes stories in English. Each words was assigned entropy values (GPT-2 extracted) and split into two conditions depending on contextual constraint (low entropy = predictable; high entropy = unpredictable). We built a linear mapping model between the words’ entropy and the brain signals. On this, we calculated alpha power in the time window preceding (-300ms–0ms) and following (0ms–300ms) word onsets. Results showed that, in contrast with previous research, alpha power was greater in the low entropy than the high entropy condition, with larger difference in the pre-onset time-window than the following window. This confirms that alpha power modulation reflects anticipatory mechanisms but the effect depends on the task manipulation. In task that do not prompt prediction (simple comprehension), high alpha power reflect automatic word expectation. Conversely, when words are not predictable (same context), expectation update is likely paused, leading to a decrease in alpha power.
[1] Clark et al., (2013). Behavioral and Brain Sciences.
[2] Gastaldon et al., (2020). Cortex.
[3] Lago et al., (2023). Psychophysiology.
