NEURAL CORRELATES OF SEGMENTATION STRATEGIES DURING ARTIFICIAL LANGUAGE PROCESSING

Compared to firing-rate–based codes, encoding information in the precise timing of spikes can yield neural representations with lower energy consumption. Understanding how the brain exploits spike timing for information encoding is therefore central to both biology and emerging information-processing technologies. While sensory systems are known to encode information through relative spike timing, recent work has shown that spike-timing-based representations also support higher-level cognitive functions. In the human cortex, information can be encoded in reproducible spike sequences rather than firing rate alone (Xie_et_al.,_2024), and temporal codes can provide more stable representations over time than rate-based codes (Zhu_et_al.,_2025).

The rodent hippocampus provides a canonical model for studying higher-level cognition. However, most studies of hippocampal spatial and cognitive encoding rely on firing-rate–based models that average spike counts over time windows. Although hippocampal spiking activity can be described using point-process frameworks defined directly on spike times (Eden_et_al.,_2004; Truccolo_et_al.,_2005), decoding approaches in behaving animals remain largely rate-based.

Here, we demonstrate that hippocampal information can be robustly decoded from the relative timing of spikes across neuronal populations. We further show that decoding performance improves when spike timing is considered compared to firing rate alone. These findings suggest that hippocampal representations may be more efficient than previously assumed, potentially requiring reduced connectivity and lower energy expenditure. Beyond advancing our understanding of hippocampal computation, this work has implications for how high-level information encoding is studied in the brain and for the design of neuromorphic computing systems that exploit spike-timing-based representations (Stanojević_et_al.,_2024; Izhikevich,_2025).