Fixation-Related Potentials (FRPs)

• Fixation-Related Potentials are EEG signals aligned with eye fixations, offering a precise measure of cognitive processing during natural tasks.
• They enable detailed analysis of visual and semantic processing via synchronized eye-tracking and EEG, as shown in reading and code comprehension studies.
• Recent research extends FRPs to compare human reasoning with AI models, providing insights into brain-aligned language and abstract reasoning systems.

Fixation-related potentials (FRPs) are a class of neural signals obtained by time-locking electroencephalography (EEG) data to the precise onset of an eye fixation event. Unlike conventional event-related potentials (ERPs), which typically synchronize EEG recordings with discrete external stimulus presentation, FRPs align the neural signal with the observer’s self-paced, attention-guided interaction with visual stimuli. This property renders FRPs a critical tool for investigating the neural dynamics underlying naturalistic behavior, including sentence reading, code comprehension, and abstract reasoning.

1. Conceptual Framework of Fixation-Related Potentials

FRPs are defined as EEG potentials that are epoched relative to the moment the gaze first fixates a particular region of interest in a visual scene. This approach enables the paper of cognitive processes as they unfold in ecologically valid, continuous tasks. The technical distinction from traditional ERPs is central: in FRP methodology, the “event” of interest is an endogenous attention shift, rather than the exogenous onset of a stimulus.

The extraction of FRPs typically involves co-registration of eye-tracking and multichannel EEG, followed by the alignment of neural signals with the timestamp of fixations on task-relevant regions. The resulting FRPs allow for direct examination of cognitive processing, lexical access, and memory updating at the moment of information uptake.

2. FRPs in Sentence Reading and Linguistic Processing

FRPs have been utilized as sensitive neural indices of word and sentence comprehension under natural reading conditions (Liu et al., 2021). By synchronizing EEG traces with fixations during self-paced reading, research demonstrates distinctive temporal dynamics:

• A pronounced occipital ERP peak is observed at approximately 162 ms after sentence onset, corresponding with rapid retrieval of lexical and semantic visual features.
• This ERP peak can be modeled by a Gaussian function:

$$ERP(t) = A \cdot \exp\left(-\frac{(t-t_0)^2}{2\sigma^2}\right), \quad t_0 \approx 162\,\text{ms}$$

which characterizes the brief, localized neural activation.

• Event-related spectral perturbations (ERSPs) reveal that, around 200 ms post-fixation, power increases in the high alpha (10–12 Hz), high beta (25–30 Hz), and high gamma (40–50 Hz) bands, while power decreases occur in low beta (13–25 Hz) and low gamma (30–40 Hz).
• These temporal and spectral signatures support the view that visual and semantic information extraction is tightly coupled and predominantly occurs within a 200 ms window of fixation.

A plausible implication is that robust FRP and ERSP markers allow cognitive NLP models to be quantitatively evaluated against neural correlates of semantic processing, potentially informing the development of bio-inspired language architectures.

3. FRPs in Program Code Comprehension and Ambiguity Resolution

Application of FRP methodology to program code comprehension allows for the detailed analysis of how programmers process ambiguous code elements, such as “atoms of confusion” (Bergum et al., 13 Dec 2024). In this context:

• EEG is synchronized to the initial fixation on predefined regions containing ambiguous or unambiguous code fragments. The temporal precision of FRPs is essential due to large variance in fixation onset across trials.
• Ambiguous code elicits a sustained, late frontal positivity between 390 ms and 660 ms post-fixation. This component is interpreted as reflecting neurocognitive mechanisms for processing unexpected but plausible input and updating the reader’s situation model.
• The difference waveform is formalized as

$$\Delta FRP(t) = FRP_{ambiguous}(t) - FRP_{unambiguous}(t)$$

where significant positive values in the specified time window denote increased cognitive load.

• Notably, this late positivity bears strong resemblance to ERP components seen in natural language studies, particularly those triggered by unexpected words in high-constraint contexts.

These findings indicate that both code and natural language comprehension recruit similar neural processes for expectancy violation and model updating, emphasizing a domain-general mechanism for information integration during reading.

4. FRPs in Human-AI Alignment During Abstract Reasoning

Recent work has investigated the relationship between FRPs recorded during abstract reasoning tasks and the internal representations of LLMs (Pinier et al., 12 Aug 2025):

• During pattern-completion tasks, human EEG was time-locked to fixations over task icons, yielding frontal FRPs relevant to high-order reasoning.
• Representational similarity analysis (RSA) was used to compare human FRP-derived representational dissimilarity matrices (RDMs) with those from mid-level LLM layers, where

$\text{Dissimilarity} = 1 - r$

and $r$ denotes the Pearson correlation between trial-level activation vectors.

• Moderately positive correlations (r ≈ 0.17–0.25) were found between human frontal FRP RDMs and the emergent block structure in the hidden layer representations of performant LLMs (notably Qwen-2.5-72B and DeepSeek-R1-70B).
• Control analyses with response-locked ERPs and resting-state EEG yielded lower or negative correlations, establishing the specificity of the fixation-driven methodological approach.

The aligned representational geometries suggest that, during abstract reasoning, both biological and artificial systems may instantiate convergent strategies for pattern segregation and task structure extraction.

5. Methodological Aspects of FRP Extraction and Analysis

Extraction of FRPs requires high-precision synchronization of eye-tracking and EEG data streams. In both linguistic and code comprehension paradigms:

• The FRP epoch is defined as a window (e.g., –300 ms to +1000 ms) centered on the fixation onset to the area of interest.
• Robust baseline correction and artifact rejection are essential due to the self-paced nature of the fixations.
• Statistical inference commonly employs cluster-based permutation testing across time points and spatial electrodes to detect significant periods and topographies of neural activation (as in late positivity detection for code ambiguity).
• Time-frequency decomposition (e.g., for ERSP measurement) is achieved using windowed Fourier or wavelet transforms, with changes quantified by normalized log-power formulas, such as

$$\Delta P(f,t) = 10 \cdot \log_{10}\left(\frac{P(f,t)}{P_{\text{baseline}}(f)}\right)$$

These methodological refinements ensure that FRPs selectively capture cognitive events as anchored by the participant’s natural visual behavior, minimizing timing confounds common to stimulus-onset ERPs.

6. Cognitive and Applied Implications

FRP signals have elucidated essential aspects of human cognitive processing:

• In reading, FRPs provide objective markers for lexical and semantic retrieval within critical early windows, supporting biologically grounded evaluation of NLP models (Liu et al., 2021).
• In software engineering, late frontal FRP components expose cognitive load associated with ambiguous coding constructs, with direct relevance for programming language design and software maintenance (Bergum et al., 13 Dec 2024).
• In human-AI comparison, the partial alignment of FRP-derived representational geometry with LLM hidden states suggests shared inductive principles for abstract reasoning, with possible consequences for brain-aligned artificial intelligence architectures (Pinier et al., 12 Aug 2025).

A plausible implication is that integrating FRP analysis with computational modeling and AI development could foster mutual enlightenment: providing metrics for evaluating cognitive plausibility in models and deepening mechanistic understanding of human cognition during complex, real-world tasks.

7. Visualization and Schematic Representation

FRP research employs schematic figures and mathematical illustrations to substantiate findings and facilitate methodological transparency. For example:

• ERP peak and ERSP change sketches for occipital FRPs in sentence reading (peak at 162 ms, ERSP changes at 200 ms) clarify the spatiotemporal profiles of neural activation (Liu et al., 2021).
• Experimental timelines marking code snippet presentation, fixation-aligned FRP epochs, and windows of late positivity simplify interpretation of code comprehension dynamics (Bergum et al., 13 Dec 2024).
• Accuracy and representational similarity matrices (RDMs), constructed from both FRP and LLM activations, visually corroborate the block-structured alignment in abstract reasoning tasks (Pinier et al., 12 Aug 2025).

Such documentation bridges neurophysiological evidence and computational modeling, allowing precise translation of empirical observations into formal, quantitative frameworks suitable for technical audiences.