Abstract: Emotion fundamentally reconfigures the allocation of attentional resources. While the normalization model successfully characterizes how emotional valence modulates spatial attention—predicting response or contrast gain based on the attentional field—its applicability to rhythmic temporal attention remains unexplored. Unlike spatial orienting, rhythmic attention relies on the dynamic entrainment of neural oscillations. Grounded in the normalization framework, the present study investigates whether emotional states regulate temporal attention through specific gain computations and how they alter the trajectory of expectation formation. We hypothesized that distinct emotional valences would recruit qualitatively different strategies (response gain versus contrast gain) to modulate steady-state processing, while simultaneously reshaping the temporal dynamics of attentional deployment to accommodate environmental uncertainty.
Two experiments were conducted involving 26 healthy Han Chinese university students (13 in Experiment 1; 13 in Experiment 2). In Experiment 1, a 3 (Emotional valence: positive, neutral, negative) x 2 (Rhythmic validity: valid, invalid) x 5 (Stimulus contrast) within-subjects factorial design was employed. Emotional states were induced via validated audiovisual clips. Participants completed an orientation discrimination task using Gabor targets embedded within rhythmic streams. To mathematically characterize the attentional gain profiles, the Naka-Rushton equation was fitted to the discriminability (d') data for each participant. This allowed for the precise decomposition of modulation into response gain (amplification of asymptotic response, indexed by d'max) and contrast gain (enhancement of sensitivity, indexed by c50). Experiment 2 investigated the dynamic evolution of attentional allocation using a 3 (Emotional valence) x 5 (Rhythmic strength: 1 to 5 cues) x 2 (Rhythmic validity) design with target contrast fixed at a supra-threshold level. Crucially, we extended the dynamic normalization model by incorporating a parameterized rhythmic strength coefficient to quantify the accumulation of attentional gain over time. This extended model was optimized against behavioral data using Bayesian Adaptive Direct Search (BADS) to estimate the parameters governing the temporal distribution of attentional weights (w_av).
Experiment 1 revealed a robust valence-dependent dissociation in gain mechanisms. Under negative emotional states, the attentional gain (defined as the d' differential between valid and invalid conditions) exhibited a monotonic increase as a function of stimulus contrast. Computational modeling confirmed that negative emotion selectively elevated d'max in valid trials without modulating c50, a signature characteristic of a response gain mechanism. Conversely, positive and neutral emotions yielded a non-monotonic, inverted-U shaped gain function. Model parameters indicated a significant reduction in the semi-saturation constant (c50) for valid trials under positive emotion, with no significant alteration in d'max, indicative of a contrast gain mechanism. Experiment 2 demonstrated that while rhythmic cueing facilitated performance across all conditions, emotion significantly modulated the magnitude and temporal precision of this effect. The reaction time validity effect was markedly attenuated under both positive and negative emotions relative to the neutral condition. The extended dynamic normalization model achieved an exceptional goodness-of-fit (R-squared > 90%) across conditions. Parameter analysis revealed that in neutral states, attentional resources were sharply focused on the predicted time point; however, under emotional arousal (both positive and negative), the distribution of attentional resources became significantly more equipotent between valid and invalid intervals. This finding suggests that emotion flattens the temporal weighting function, effectively broadening the temporal window of attention.
This study constitutes the first empirical and computational instantiation of the normalization model within the domain of emotional modulation of rhythmic temporal attention. The findings establish a dual regulatory framework: emotion selects distinct steady-state gain mechanisms based on valence—amplifying signal magnitude via response gain under negative emotion (supporting a "threat-vigilance" mode) and enhancing perceptual sensitivity via contrast gain under positive emotion (supporting an "opportunity-seeking" mode). Furthermore, emotion promotes dynamic flexibility by expanding the temporal attentional window. This broadening reflects a strategic trade-off wherein the cognitive system sacrifices peak temporal precision to maintain heightened sensitivity to unexpected events in volatile environments. Collectively, these results validate the normalization model as a canonical framework for deciphering emotion-cognition interactions and underscore the critical role of emotion in dynamically reconfiguring the computational architecture of temporal attention.

Key words: emotion, rhythmic temporal attention, normalization model of attention