A central challenge in cognitive neuroscience is to explain how semantic and episodic memory - two major forms of declarative memory, typically associated with cortical and hippocampal processing - interact to support learning, recall, and imagination. Despite significant advances, we still lack a unified computational framework that jointly accounts for core empirical phenomena across both semantic and episodic processing domains. Here, we introduce the Generative Episodic-Semantic Integration System (GENESIS), a computational model that formalizes memory as the interaction between two limited-capacity generative systems: a Cortical-VAE, supporting semantic learning and generalization, and a Hippocampal-VAE, supporting episodic encoding and retrieval within a retrieval-augmented generation (RAG) architecture. GENESIS reproduces hallmark behavioral findings - including generalization in semantic memory, recognition, serial recall effects and gist-based distortions in episodic memory, and constructive episodic simulation - while capturing their dynamic interactions. The model elucidates how capacity constraints shape the fidelity and memorability of experiences, how semantic processing introduces systematic distortions in episodic recall, and how episodic replay can recombine previous experiences. Together, these results provide a principled account of memory as an active, constructive, and resource-bounded process. GENESIS thus advances a unified theoretical framework that bridges semantic and episodic memory, offering new insights into the generative foundations of human cognition.

Effect of transcranial direct current stimulation on semantic memory depends on anodal stimulation sites in patients with schizophrenia.

Single-neuron recordings from the medial temporal lobe of patients undergoing epilepsy surgery have revealed "concept cells" that respond selectively and invariantly to meaningful stimuli such as specific people, places, or objects. These responses offer a unique window into how individual neurons encode high-level, multimodal representations-the building blocks of episodic memory-that differ from the more distributed, often hierarchical representations supporting semantic memory in the neocortex. Episodic and semantic memory, the systems for storing past experiences and conceptual knowledge, have traditionally been regarded as distinct. However, converging evidence from neuroimaging, lesion studies, and electrophysiological recordings challenges this strict dichotomy. This review synthesizes findings from human single-neuron recordings to re-examine the traditional distinction between episodic and semantic memory. We propose that the primary difference between the two systems lies in the structure of the associations they support: sparse, arbitrary links supporting episodic memory in the hippocampus versus ordered, hierarchical representations supporting semantic memory in the neocortex.

Arabic verbal fluency in mesial temporal and generalized epilepsy: evidence from letter and category tasks with healthy controls.

Convergent thinking, the ability to find a single optimal solution to a well-defined problem, is considered a core component of creativity, and is often assumed to rely on controlled, deliberative processes. We tested this assumption using the Compound Remote Associates (CRA) test, where participants have to find a word that connects three seemingly unrelated words (e.g., "river, note, account"; solution: "bank"). We implemented a two-response paradigm wherein participants provided an initial, intuitive response (under cognitive load and time constraints to minimize deliberation), followed by a final, deliberate response. Our findings reveal that, in most cases, extended deliberation was not necessary for sound thinking-correct final responses were typically preceded by accurate intuitive responses produced under time pressure and cognitive load. By using large language models and semantic network modeling, we found that items with a smaller semantic search space are better solved intuitively, and that participants with a more efficient and flexible semantic memory structure display higher intuitive performance on the CRA. These results suggest that effective problem-solving in creative tasks may often rely on fast, automatic associative processes within semantic memory, without necessarily requiring extended deliberation.

Memory for multimodal picture-sound items: Influences of congruency, identifiability and cognitive ageing.

Environmental sound processing in primary progressive aphasia: behavioral deficits and their neural correlates.

Pouring water from a bottle into a glass requires representing their relative spatial positions (spatial information) and knowing that such objects are found together, or one by one (temporal dynamics), in typical contexts, such as a kitchen (semantic information).While spatial information is encoded using egocentric (body-centered) or allocentric (object-centered) reference frames, semantic knowledge is organized in context frames, which help predict object-context associations. This study investigates whether daily-like semantic contexts and temporal dynamics affect egocentric and allocentric spatial representations. Participants were immersed in semantically meaningless (abstract) or meaningful (bathroom, kitchen) virtual environments. They memorized triads of objects presented either simultaneously (static condition) or sequentially (dynamic condition) and then provided egocentric or allocentric spatial judgments. Results revealed an improvement of allocentric accuracy in meaningful rather than meaningless environments, whereas no differences emerged for egocentric spatial judgments. Crucially, this semantic benefit for allocentric encoding emerged only in the dynamic condition, where spatial information had to be integrated over time. These findings show that semantic context selectively facilitates allocentric spatial representations, particularly when perceptual input unfolds dynamically. The results support the interaction between semantic and spatial memory to shape how we perceive the surrounding world.

Human word concept learning leverages prior knowledge to generalize from limited exemplars, yet its neural implementation remains unclear. We developed a Neural Bayesian Model (NBM) that incorporates neural representational priors to explain word concept learning. Using fMRI, we first constructed a neural prior space from activity elicited by familiar objects (with novel shapes as controls), and then examined neural responses during novel word concept learning based on these objects. The NBM integrating priors from ventral occipitotemporal cortex predicted both neural representations and behavioral generalization, outperforming control models lacking neural priors. In contrast, hippocampal activity supported learning for novel shapes without benefit from the NBM, consistent with prior-free associative mechanisms. Large language models showed weaker alignment with human generalization patterns. These findings dissociate prior-based cortical inference from hippocampal exemplar-associative learning, providing a neural instantiation of Bayesian concept acquisition and clarifying the interplay between semantic and episodic memory systems.

Examining the neural dynamics of semantic memory integration using noninvasive brain stimulation with concurrent EEG

Studies of brain lesions or volumes indicate that the integrity of medial and lateral temporal lobe structures is important for news event memory accuracy, but the relationship between cortical thickness and news event memory accuracy has not yet been investigated in older adults. In a mixed sample of 70 older adults with variable cognitive abilities but without dementia, we investigated the relationship between cortical volume, hippocampal volume, and cortical thickness with news event recognition memory accuracy across the entire adult lifespan using the Retrograde Memory News Events Test. Partial Least Squares analysis was used to identify brain regions where news event memory accuracy scores significantly correlated with cortical volume, hippocampal volume, and cortical thickness. We found that mean news event memory accuracy significantly correlated with volume/thickness for a network of regions that included the hippocampus, medial/lateral temporal lobe, medial/lateral parietal lobe, and specific areas within the medial/lateral prefrontal cortex. Poorer performance was associated with a thinner cortex (and smaller volumes). Almost all regions in this network exhibited decreasing brain-behavior correlations as the age of memory increased; thus, retrieval of remote memories was less reliant on the network. We also found regions in this network that were not identified by the Retrograde Memory News Events Test posttest (a measure of episodic anterograde memory for the Retrograde Memory News Events Test content) nor traditional neuropsychological tests. The regions identified as uniquely contributing to news event memory overlap with regions known to exhibit increasing Alzheimer's disease pathology and cortical thinning when pathology begins to spread outside of the medial temporal lobe. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

Abstract A pivotal debate in Alzheimer’s disease (AD) revolves around the clinical utility of a purely biomarker-based diagnostic framework, making it imperative to identify early functional signs that are both clinically meaningful and sensitive to biological biomarkers in the preclinical stages. While the development of the Preclinical Alzheimer’s Cognitive Composite (PACC) addresses this need, its lack of language-specific markers might limit its sensitivity, given that language deficits are among the earliest manifestations of AD. To address this limitation, we developed PACC variants incorporating proper names delayed recall, a novel lexical-semantic marker robustly linked to PET and CSF biomarkers. We analyzed 824 dementia-free participants (mean age ≈ 62 at baseline) from the Wisconsin Registry for Alzheimer’s Prevention, stratifying them by plasma phosphorylated tau 217 (p-tau217) levels (using established amyloid PET positivity cutoffs: negative [<0.40 pg/mL; N=539], intermediate [0.40–0.63 pg/mL; N=163], positive [>0.63 pg/mL; N=122]). We constructed six PACC variants: (1) traditional PACC4 (Rey Auditory Verbal Learning Test immediate learning, Logical Memory II [LM II; delayed recall], Digit Symbol, Mini-Mental State Examination [MMSE]); (2) PACC3 (excluding MMSE to mitigate ceiling effects); (3–4) PACC4_PN and PACC3_PN, replacing LM II’s total score with proper-name recall; and (5–6) PACC4+VF and PACC3+VF, adding animal fluency. Cognitive composites were calculated as the equally weighted sum of component z-scores. We compared PACC variants` sensitivity (i.e., longitudinal decline rates) associated with plasma p-tau217, using linear mixed models adjusted for age, gender, literacy, and practice (total visits minus one). We showed individuals with positive plasma p-tau217 status exhibited significantly faster cognitive decline across all PACCs. Notably, the proper name enhanced-PACC, especially the parsimonious PACC3_PN, demonstrated the steepest decline—6.4%–19% faster than the traditional and animal fluency-appended versions. Our optimized measure offers critical advantages in clinical utility: an 82% reduction in scoring burden for clinicians (9-point proper names vs. 50-point total LM score), finer semantic memory granularity within an episodic memory task, and enhanced sensitivity to AD biomarkers. Our findings position proper name-enhanced PACC as a clinically practical tool for tracking longitudinal cognitive trajectories, highlighting proper names delayed recall as a sensitive marker due to AD-type brain changes. Future work will validate its utility for trial enrichment and clinical screening.

Supplemental Material for Structural Neuroanatomy of Semantic Retrograde Memory in Older Adults

Baseline Neurocognitive Functioning in Children/Teenagers and Young Adults Treated for Central Nervous System Malignancies Prior to Proton Therapy.

Beck observed overgeneralised thinking as a key vulnerability factor for excessive self-blame/-criticism in major depressive disorder (MDD). Whilst the contribution of reduced access to specific autobiographical memory episodes has often been considered, the role of more abstract semantic memory systems remains elusive. Here, we investigated the individual ability to differentiate between the meaning of abstract social concepts when interpreting behaviour (e.g. "critical" vs "fault-finding") and its contribution to vulnerability to self-blaming biases and MDD using a previously developed cognitive task in 96 participants (n = 60 medication-free remitted MDD and n = 36 controls). Of those, 75 participants also completed an fMRI paradigm in which they viewed self- and other-blame emotion-evoking statements. A priori right anterior temporal lobe (ATL) seed and bilateral anterior subgenual cingulate and dorsolateral prefrontal cortex regions-of-interest were defined. As expected, the MDD group exhibited greater self-blame-selective conceptual overgeneralisation and stronger interdependency of conceptual overgeneralisation with negative emotional valence relative to the control group. Individuals with this interdependency showed higher self-blame-selective right anterior subgenual cingulate and primary motor cortex activations across groups, potentially corresponding to stronger self-blame and self-agency attributions, respectively, when conceptually overgeneralising the interpretation of their negative actions. Individuals with higher self-blame-selective conceptual overgeneralisation displayed lower right ATL activation for self- vs other-blame, suggesting reduced access to differentiated conceptual representations. Future studies are needed to confirm the hypothesis that self-blame-selective conceptual overgeneralisation characterises a distinct neurocognitive subtype of primary MDD vulnerability which may be modulated by depressive state and thus serve as a personalised treatment target.

Aims. This study aims to explore contronyms (Janus words), which are words that have two opposite meanings, in English and Arabic, by examining the default and dominant meaning of certain selected contronyms to gain a deeper understanding of how individuals comprehend such paradoxical meanings. Methods. Forty Jordanian participants (20 males and 20 females) were asked to provide the first meaning that comes to their mind when reading certain selected contronyms and, then They were asked to give the meaning of the same contronyms in contextualized sentences where three sentences are used for each contronym and; the first sentence bias one meaning, the second sentence bias the opposite meaning and a third sentence that is neutral and allows both interpretation. the frequencies of the participants responses were to calculate using SPSS to analyzed the default (dominant) meaning of the selected contronyms. Results. The analysis revealed the dynamic nature of the semantic memory as that there are common and default senses for these contronyms. The results offer valuable insights for translation and language pedagogy strategies. Conclusion. Exploring contronyms provides valuable insights into the interaction between automatic access and context-driven meaning selection in both Arabic and English. One meaning consistently appeared as the most dominant interpretation, reflecting semantic dominance which is shaped by familiarity, and frequency, often overriding default meanings when sentences biased alternative senses. In general, the study emphasize the centrality of context in psycholinguistic models of lexical access and provides empirical support for the notion that contronyms as the polysemous words are processed through a dynamic interplay of default activation and contextual modulation. This study has several implications for translation, language teaching strategies, lexicography, and Natural Language Processing (NLP) systems. Future studies could replicate this study, including participants from various linguistic backgrounds, considering factors such as linguistic proficiency, and socioeconomic background, including more complex contexts, such as narratives, conversations, investigating Online language processing of contronyms using eye-tracking technique, and for Natural Language Processing (NLP) technology, training or even preparing large language models to identify default meanings in neutral contexts, that could lead to a more accurate linguistics models.

Previous evidence has identified the ventromedial prefrontal cortex (vmPFC) as crucial for implementing high-level semantic memory structures (schemas) during event construction. If this is the case, one would expect reduced semantic coherence in events mentally constructed by vmPFC patients compared to healthy and brain-damaged controls. We tested this prediction by having participants mentally construct events using objects as cues and reanalyzing a published dataset using sentences as cues. In both cases, we measured the semantic coherence of patients’ mental constructions and their semantic coherence with the cue, using transformer-based sentence embeddings (S-BERT), and further corroborated the findings with E5 Multilingual and E5 Italian embedding models. Our results reveal that the hypothesized impairment in semantic coherence following vmPFC damage is, in fact, task-dependent. With minimal (object) cues, vmPFC patients’ reports exhibited reduced local coherence, increased connectedness to the cues, and reduced lexical diversity. In contrast, with extended (sentence) cues, they showed preserved- or even enhanced-local and global coherence. We suggest that vmPFC integrity is necessary to trigger schema activation under minimal cue conditions. Although extended cues may facilitate schema activation, schemas are degraded and essentialized following vmPFC damage, thereby constraining patients’ mental constructions within a narrower—hence overly coherent—semantic space.

Three decades of studying false semantic memory

Transient global amnesia (TGA) is a striking model of isolated amnesia. While hippocampal lesions are well described, the network-level mechanisms and the precise neuropsychological profile remain debated. Our objective was thus to characterize functional and neuropsychological correlates of acute TGA and their longitudinal evolution. Prospective, single-center case-control study of 20 patients with acute TGA and 20 age- and sex-matched healthy controls. All participants completed neuropsychological testing and underwent structural and functional MRI at three time points: acute phase (< 24 h from onset), day 3, and 3 months. Primary outcomes were neuropsychological performance across episodic, semantic, and metamemory domains and resting-state fMRI connectivity within the episodic memory network. Secondary outcomes were functional connectivity within the Default Mode (DMN), Executive (ECN), and Salience (SN) networks. A total of 40 participants were included (20 patients with TGA, mean age 65.5 years, 45% women; 20 controls, mean age 64.3 years, 45% women). In patients, median delay from symptoms' onset to MRI was 6.67 h. Neuropsychologically, patients showed profound multimodal anterograde amnesia during the acute phase, resolving by 3 months. This deficit was largely isolated, sparing semantic memory and metamemory. Structurally, small bilateral lesions were present in most patients. Functionally, acute hypoconnectivity was observed within the extended hippocampal system, particularly between parahippocampal and cingulate cortices, normalizing by 3 months. No consistent disruption was found in large-scale networks (default mode, executive control, salience). TGA is associated with transient, selective hypoconnectivity within the mesiotemporal-cingulate episodic memory network, aligning with previous reports and further precising the functional anatomy. The finding of a profound anterograde amnesia was replicated and its recovery timecourse was elucidated. Semantic memory and metamemory remain preserved, clarifying inconsistencies in prior reports. These findings suggest that TGA reflects a transient limbic dysconnectivity syndrome rather than a diffuse network disorder, reconciling structural lesions with clinical and functional data.

Multi-agent systems are increasingly deployed in complex, dynamic environments requiring sophisticated coordination, rapid decision-making, and extensive knowledge sharing. However, as the number of agents and the complexity of tasks grow, inter-agent communication and memory constraints become critical bottlenecks. This paper proposes a novel framework integrating communication compression memory management with cooperative reasoning to address these challenges. We introduce an adaptive latent-space compression algorithm that dynamically reduces communication overhead while preserving mission-critical semantic information. Furthermore, we design a hierarchical memory management system that efficiently allocates episodic and semantic memories, facilitating rapid retrieval and bounded storage requirements. Building upon this optimized infrastructure, our cooperative reasoning module enables agents to align their belief states and execute joint logical inferences without requiring full state broadcasting. Extensive empirical evaluations across diverse multi-agent scenarios demonstrate that our proposed architecture reduces communication bandwidth consumption by significant margins while maintaining or improving collective task performance compared to state-of-the-art baselines. We provide a rigorous theoretical analysis of the compression bounds and reasoning convergence. Our findings indicate that integrating memory pruning with selective communication is essential for the scalable deployment of autonomous multi-agent networks in bandwidth-constrained environments.