Involuntary and voluntary memory retrieval relies on distinct neural representations and oscillatory processes

Involuntary autobiographical memories are common in daily life and can be highly distressing when they involve traumatic content, as in posttraumatic stress disorder. Phenomenological accounts suggest four main differences from voluntary retrieval: greater sensory richness, generalized accessibility by perceptually similar cues, temporally extended replay, and rapid, automatic occurrence that can interfere with ongoing tasks. Prior neurocognitive work indicates faster retrieval and reduced top-down control during involuntary recall, but has not tracked the content of individual memories to compare representational formats. This study directly contrasts involuntary and voluntary retrieval for neutral visual events, testing whether involuntary recall reinstates sensory feature representations and extended replay, whereas voluntary recall reinstates temporally compressed item-specific, gist-like representations. Using a visual half-field paradigm and EEG-based encoding-retrieval similarity analyses, the authors hypothesize that voluntary retrieval engages goal-directed search and MTL–prefrontal theta processes to reconstruct item-specific representations, while involuntary retrieval relies on cue-driven sensory reactivation, extended temporal replay, and midfrontal theta related to interference.

Involuntary memories can occur as frequently as voluntary ones and are often sensory-rich, rapidly triggered, and intrusive. Clinical theories proposed a separate perceptual memory system for involuntary memories, whereas cognitive accounts emphasize a single episodic system with different retrieval modes. Neuroimaging studies show early posterior ERP components for involuntary recall and reduced lateral prefrontal engagement compared to voluntary recall, but have not disentangled item-specific from sensory-feature reinstatement. Work on representational formats suggests that episodic traces maintain multiple levels (sensory to conceptual) that can be flexibly accessed. Voluntary recall often reinstates higher-order, gist-like representations and can be temporally compressed; cue-driven reactivation can support rapid retrieval under favorable conditions. Theta oscillations have been linked both to strategic search and recollection (prefrontal–MTL) and to interference detection (midfrontal/ACC). This study builds on visual half-field paradigms and representational similarity analysis to probe sensory-feature versus item-specific reinstatement, compression, and oscillatory signatures in involuntary versus voluntary retrieval.

Participants: Thirty-one right-handed adults (17 female; mean age 23.87 years, SD 4.86) participated; 12 were excluded from EEG analyses (final EEG sample n=19) due to handedness, technical issues, or insufficient full-hit trials. Ethics approval and informed consent were obtained. Stimuli: 288 cue–object pairs (abstract line drawings as cues; everyday objects balanced for complexity/category) segmented into six blocks. Each block had 16 old pairs and 16 new pairs presented in both retrieval phases. Procedure: Six blocks of encoding, voluntary retrieval, and involuntary retrieval in fixed order. Encoding: fixation (1,000–1,500 ms), cue in center (4,000 ms), object in left or right visual hemifield (200 ms) to enforce contralateral processing; learning difficulty rating; 16 s backward counting to prevent rehearsal. Voluntary retrieval: old/new recognition of cues (1,500 ms); for “old” responses, source memory for hemifield (<5,000 ms) and forced-choice item recognition (<3,000 ms). Feedback for false alarms; missed old cues were re-presented to increase subsequent involuntary intrusions. Involuntary retrieval: same cue set with a speeded visual discrimination task (line crossings) prioritized for speed/accuracy (feedback for slow/incorrect responses). After each trial, participants reported whether an involuntary memory occurred (<3,000 ms); if so, they completed source and forced-choice item tasks. EEG acquisition: 64-channel cap (10–20 system), 500 Hz sampling, FCz reference (offline common average), impedances <10 kΩ. Artifact rejection via visual inspection and ICA. Analyses implemented in FieldTrip and MATLAB; ERS via custom pipeline. Only fully remembered trials (hits with correct source and item) entered EEG analyses. Behavioral analyses: D′ for discrimination, Criterion C for bias; 2×2 ANOVA on response times for retrieval condition (voluntary vs involuntary) × memory (hits vs misses), with post hoc tests. Representational similarity analyses: Encoding–retrieval similarity (ERS) computed as Fisher z-transformed Spearman correlations between concatenated electrode×time vectors from overlapping 300 ms windows (20 ms steps), across encoding and retrieval. Two contrasts: sensory feature reactivation (same VF ERS vs different VF ERS using contralateral electrodes) and item-specific reactivation (same item ERS vs same VF ERS across all electrodes). Contralateral electrode localizer used encoding–encoding similarity (EES) to identify time window (50–950 ms) and lateralized electrodes sensitive to hemifield. Cluster-based permutation tests corrected for multiple comparisons. Reactivation compression: Within-subject cluster detection over ERS matrices using subject-specific thresholds (mean of positive t-values; robustness tested with stricter thresholds). Compression score defined as number of encoding time points per retrieval time point within largest cluster; compared between involuntary sensory-feature reactivation and voluntary item-specific reactivation. Time–frequency analysis: Induced theta power (2–8 Hz) computed via Morlet wavelets (width=5) after ERP subtraction; baseline −500 to −100 ms; cluster-based permutation tests over 0–1,500 ms comparing full-hits vs correct rejections separately for each retrieval phase. Source localization via LCMV beamforming on individualized BEM models; theta power averaged over significant sensor-level time windows; nonparametric spatial cluster tests. Control analyses separated slow-theta (2–4 Hz) and fast-theta (4–8 Hz), and assessed aperiodic slope versus oscillatory components.

Behavioral: Participants endorsed more old than new items in both phases (voluntary: t30=9.177, p<.001, d=1.65; involuntary: t30=10.89, p<.001, d=1.95). D′ was higher during involuntary than voluntary retrieval (M=1.15 vs 0.67; t30=8.51, p<.001, d=1.53), driven by lower false alarms in involuntary (t30=7.80, p<.001, d=1.40). Response bias C was more conservative in involuntary (M=0.55) than voluntary (M=0.18) conditions (t30=5.29, p<.001, d=0.95). Reaction times showed a condition×memory interaction (F1,30=5.12, p=.031): involuntary intrusions slowed visual discrimination (hits vs misses: t30=2.96, p=.006, d=0.53), whereas voluntary RTs did not differ (t30=0.39, p=.703). ERS—sensory feature reactivation: Significant sensory feature reinstatement during involuntary retrieval (same VF > different VF) across contralateral electrodes, with retrieval 250–1,150 ms and encoding 470–1,350 ms (tsum=668.51, Pcorr=.006, d range ≈0.61–1.00). No sensory feature reinstatement during voluntary retrieval (Pcorr>.935). Direct comparison showed stronger sensory feature reinstatement in involuntary vs voluntary with retrieval 350–990 ms and encoding 490–1,190 ms (tsum=1172.50, Pcorr<.001, d range ≈0.80–1.15). ERS—item-specific reactivation: Involuntary retrieval showed no significant item-specific ERS cluster (Pcorr>.674). Voluntary retrieval exhibited a significant item-specific cluster with retrieval 210–810 ms and encoding 250–1,050 ms (tsum=963.92, Pcorr=.018, d range ≈0.71–0.87). Item-specific ERS was higher in voluntary vs involuntary in retrieval 370–790 ms and encoding 530–1,310 ms (t≈−802.11, p=.029, d range ≈0.69–0.92). Electrode subset analyses confirmed voluntary item-specific reinstatement in hemifield-independent electrodes (tsum=890.68, Pcorr=.018), but not in lateralized sensory electrodes. Temporal compression: Group-level clusters suggested less compression during involuntary. Within-subject compression scores confirmed that involuntary reactivation was less compressed than voluntary (t18=2.92, p=.009, d=0.69), with no difference in overall cluster size (t18=0.79, p=.452). Effects were robust across threshold variations. Theta oscillations: Involuntary retrieval (full-hits vs correct rejections) showed an early, sustained theta power increase peaking at ~550 ms with midfrontal topography (tsum=3415.48, Pcorr<.001, d range ≈1.04–1.14). Source localization implicated left precentral and bilateral midcingulate cortex (tsum=195.11, Pcorr=.012), with trends in right occipitotemporal and left midtemporal regions. Effects were present in both slow- and fast-theta and included contributions from oscillatory theta and a steeper aperiodic slope. Voluntary retrieval showed later theta increases (450–1,350 ms; peak ~650 ms) with right frontal and left parietal distribution (tsum=826.22, Pcorr=.012, d range ≈0.84–0.98). Sources included right dorsolateral prefrontal cortex (tsum=63.50, Pcorr=.025) and a trend in left medial temporal lobe (hippocampus; tsum=19.70, Pcorr=.068). Direct between-condition theta differences were not significant (Pcorr>.124).

Findings demonstrate that involuntary and voluntary episodic retrieval access distinct representational formats and engage different control processes. Involuntary recall is characterized by cue-driven reinstatement of item-unspecific sensory features over extended time, aligning with reports of sensory-rich content, cue generalization, and protracted replay. Midfrontal theta increases localized to midcingulate and precentral regions suggest rapid detection of interference and initiation of control demands, consistent with slowed performance in the concurrent visual task. In contrast, voluntary recall leverages goal-directed search and reconstructive processes, reinstating distributed item-specific representations with greater temporal compression and engaging dorsolateral prefrontal and medial temporal systems via theta mechanisms associated with strategic retrieval and recollection. Together, results support a single episodic memory system that maintains multiple representational formats; retrieval mode determines which format is accessed and which control mechanisms are recruited. The study reconciles clinical and cognitive perspectives by proposing that involuntary memories preferentially reactivate sensory features, while voluntary retrieval targets event-specific, conceptual representations.

This work identifies distinct neural signatures and representational formats for involuntary versus voluntary retrieval. Involuntary memories rapidly reinstate sensory feature representations via extended replay and midfrontal theta linked to interference, whereas voluntary recall compresses time and reinstates item-specific representations supported by prefrontal–medial temporal theta dynamics. These insights advance understanding of episodic retrieval modes and have implications for psychological well-being and conditions featuring intrusive memories. Future research should generalize beyond retinotopic hemifield features to other sensory modalities and features, employ higher spatial-resolution methods (e.g., intracranial EEG) to localize item-specific reinstatement, apply frequency-resolved RSA or coupling analyses to identify spectral sources of reinstatement, and translate findings to clinical populations and traumatic content to test alterations in sensory feature reactivation and control mechanisms.

Generalization is currently restricted to retinotopic visual hemifield features; other sensory features and modalities were not tested. Scalp EEG’s low spatial resolution limits precise localization of item-specific reinstatement and theta sources, with hippocampal involvement remaining tentative. Time-domain RSA constrains inferences about underlying spectral components; frequency-resolved approaches are needed. The study used neutral stimuli in healthy participants, requiring clinical translation to traumatic content and PTSD. Task design limitations include fixed order (voluntary preceding involuntary) and a dual-task in the involuntary phase, which could introduce practice or inhibition effects; authors argue these cannot account for observed sensory reinstatement but acknowledge potential influences. Direct theta-power differences between retrieval modes were not significant despite phase-specific patterns.