Hidden Brain States Reveal the Temporal Dynamics of Neural Oscillations During Metaphor Generation and Their Role in Verbal Creativity

Abstract

We investigated the oscillatory brain processes while people generated metaphors for science concepts. Applying a hidden Markov model, we extracted brain states, representing temporally disentangled oscillatory processes, from EEG data. By associating the trial-by-trial occupancy of brain states with the creative quality, novelty, and aptness of the generated metaphors, we identified oscillatory processes that played a role in creative ideation in a data-driven manner. Metaphor novelty was positively associated with occupancy in a state featuring widespread alpha-band synchronization during the early trial stage and occupancy in a state featuring alpha-band desynchronization during the later trial stage. In addition, metaphor novelty was negatively associated with gamma-band power. Our results not only extend previous literature on the role of oscillatory processes in creative ideation but also highlight the importance of temporal dynamics in understanding the brain mechanisms during sustained cognitive task performance.

Keywords: EEG; creativity; hidden Markov model; idea generation; metaphor.

FIGURE 1

Hidden states during metaphor generation. (A) Each row represents the spatial‐spectral signature of a brain state. Each topography shows the spectral power distribution in the frequency band indicated at the top of the column. The red and blue colors on the scale indicate synchronization (above‐average power) and desynchronization (below‐average power). Brain states that can be mapped to either oscillatory synchronization or desynchronization were highlighted with a box. Each state was given a label for convenience. S1: alphaD = desynchronization across the spectrum with peaks in the lower and upper alpha bands and posterior sites. S6: alpha = wide‐spread alpha‐band synchronization. S4: interim = an interim state mediating transitions. S3: gamma = widespread gamma‐band synchronization with peaks in frontal, temporal, and occipital sites. (B) Distribution of fractional occupancy in each state across participants. (C) Each (directed) edge represents the transition probability between two states with the thickness proportional to the transition likelihood. The color of an edge indicates its origin node. Transitions that occurred less than 2 standard errors below the uniform probability are not shown.

FIGURE 2

Time course of state occupancies (probability of visiting) grouped by generated metaphor novelty. Temporal course of the state occupancy (instantaneous probability of visiting each brain state), grouped by a behavioral measure. The bands indicated one standard error. (A) Trials were grouped by high‐(blue) and low‐(orange) Crowd‐Novelty. Each subplot (row) represents a different brain state. The left plot was onset‐locked with 0 indicating the audio‐onset. The right plot was response‐locked with 0 indicating button press. (B) Trials were grouped by high‐(blue) and low‐(orange) Self‐Novelty.

FIGURE 3

The contrast of high versus low novelty trials in the transition probability. Differences in the transition probability between high‐ and low‐novelty trials. The arrow indicates the direction of the transition. The edge thickness is proportional to the magnitude of the difference. The color of an edge indicates its origin node. The solid edge indicates more transitions during the high‐, compared to the low‐, Self‐Novelty trials, and the dotted edge indicates fewer transitions. Only differences that exceeded 2 standard errors are shown. (A) difference between high Crowd‐Novelty and low Crowd‐Novelty. (B) Difference between high Self‐Novelty and low Self‐Novelty.