Mechanisms
Memory Processes · Retrieval and Surfacing
Recognition (Familiarity and Recollection)
Example
Someone greets you by name at a conference and you know instantly that you know them. The certainty is immediate and total, and you have nothing else. No name, no place, no occasion. That is familiarity without recollection, and it is a mode of remembering rather than a failed one.
Recognition in Humans
Definition
Recognition is the judgment that something has been encountered before. Dual-process accounts hold that it can proceed by either of two routes. Recollection retrieves qualitative information about the encoding episode, meaning not only that the item occurred but something about when, where, or alongside what, and it is modelled as a threshold process, one that either succeeds or fails, producing a highly confident response when it succeeds (Yonelinas, 1994). Familiarity yields a graded sense of priorness with no accompanying content, and is modelled by signal detection theory, in which a continuously varying strength signal is compared against a criterion the person can shift (Yonelinas, 1994). The two routes are held to be behaviourally, neurally, and phenomenologically separable, and are measured principally by three methods: the process dissociation procedure, receiver operating characteristic analysis, and the remember/know procedure, which asks the person to say whether they consciously relive the item's presentation or merely know that it occurred (Yonelinas, 2002).
Boundaries: The two-route claim is contested at the level of mechanism rather than description. The dual-process signal detection model's specific commitment, that recollection is a threshold process yielding only high-confidence responses while familiarity alone varies with confidence, is what allows recollection and familiarity to be estimated from an ROC, and it is also what its critics reject. Wixted (2007) reviews the literature that pits that model against the unequal-variance signal detection model, which accounts for the same ROC data with a single continuously distributed strength dimension whose variance is greater for studied than for unstudied items. He reports that the unequal-variance model fits better in the large majority of individual-participant datasets, that associative recognition and source memory ROCs originally reported as linear have since been shown to be curvilinear, and that source accuracy rises in graded fashion with confidence in the preceding old/new decision rather than only at the highest confidence level. His proposal is not that there is one process, but that recollection is itself continuous and combines additively with familiarity into a single strength signal, which keeps two processes while discarding the assumption that any individual decision rests on one or the other. This row is scoped to the functional distinction, meaning a judgment of priorness that carries retrieved content versus one that does not, without committing to whether the two are separable within a single decision. That scoping is what makes the AI comparison possible, because the systems assessed below produce a priorness judgment with no accompanying content at all, which is a condition the human debate does not address.
Neural Substrate
The evidence here is human. Reviewing functional neuroimaging studies, Diana, Yonelinas, and Ranganath (2007) report that within the medial temporal lobe, hippocampus and posterior parahippocampal gyrus are disproportionately associated with recollection while anterior parahippocampal gyrus is disproportionately associated with familiarity. On their binding of item and context model, parahippocampal cortex supports recollection by encoding and retrieving context, the hippocampus supports it by binding item to context, and perirhinal cortex supports familiarity by encoding and retrieving item information. The authors stress that the mapping is not simple, and that which region is engaged depends on task demands and on the type of information involved. The full text of this review was not available, so the row draws only on what its abstract states.
Human lesion evidence complicates the mapping. Wais, Wixted, Hopkins, and Squire (2006) tested six patients with bilateral damage thought to be limited to the hippocampal region, whose hippocampal volume was reduced by a mean of 44% (SEM 2.9) with adjacent parahippocampal gyrus intact, against eight age-matched controls, alongside 115 undergraduates tested at retention intervals from one hour to eight weeks. In the young adults the ROC was asymmetric when memory was strong (slope 0.63 at one hour) and grew more symmetric as memory weakened (slope 1.03 at eight weeks). The patients tested on 50-item lists produced a symmetric ROC (slope 1.14), which on the dual-process model implies a recollection estimate of zero. But when the same patients were tested on 10-item lists, which raised their overall memory strength to the controls' level, their ROC slope was 0.83, matching the controls' 0.83, and the model-derived recollection estimates were 0.22 for patients and 0.23 for controls. The authors read this as showing that ROC symmetry tracks memory strength rather than the presence or absence of recollection, and therefore as evidence against the hippocampus selectively supporting recollection.
The two accounts do not merge, and which regional conclusion is drawn depends on which model is fitted. Wais et al. report that across the eight ROCs in their study the standard signal detection model fitted better than the dual-process model in every case, summed chi-square 72.2 against 107.2, which is the neural-side expression of the dispute set out in the Boundaries block above. They also report that the earlier finding of a selective recollection deficit in hypoxic patients (Yonelinas et al., 2002) was argued by Wixted and Squire (2004) to rest on the aberrant recognition performance of a single control subject out of 55. The Yonelinas et al. (2002) full text could not be retrieved, so that characterisation is given here as Wais et al. state it and has not been checked against the original.
AI at a Glance
In humans, recognition can proceed either by recollection, which retrieves content about the encoding episode and produces high-confidence responses, or by familiarity, which yields a graded sense of priorness with no content, and the two are separable behaviourally and, less cleanly, neurally (Yonelinas, 1994; Diana, Yonelinas, & Ranganath, 2007). No AI system has been evaluated on that distinction: recognition as a composite has been tested once, where models accepted semantically related lures at roughly a seventh of the human rate (Cao, Schooler, & Zafarani, 2025), a graded priorness signal over the weights is close to undetectable (Duan et al., 2024), and the closest analogue to recollection, source attribution in a retrieval memory layer, has never been assessed as such.
Recognition in AI
How Far It Has Been Measured
Measurement in AI: Indirectly Measured
AI Memory Performance: Not Yet Compared
Where It Lives in the System
Familiarity: Context; Weights Recollection: Memory layer
What Has Been Measured
- A 2023 preprint gave 19 language models (GPT-2, 2019; GPT-Neo and GPT-J, 2021; OPT and BLOOM, 2022; 125 million to 13 billion parameters) a sentence test modeled on Shepard's: each model was trained once on 600 new sentences, then shown each studied sentence with a new one, but it made no old/new choice; the author counted a trial correct when the model's loss (how surprising it found the text) was lower for the studied sentence, explicitly as "a proxy for familiarity" (Orhan, 2023).
- After one exposure, every model's loss comparison picked the studied sentence on roughly 94% (smallest GPT-2) to nearly 100% (best BLOOM, OPT and GPT-J models) of pairs, plotted against a human band of about 88–91%; the human value is not stated in the text, no human sample or spread is given, and no statistical test against people was run (Orhan, 2023).
- The human baseline is only partly confirmed: Shepard's abstract gives a median of 88% for sentences and a 1971 summary gives 612 sentences, 68 test pairs and 89% correct, which fits Orhan's line, but the number of participants and the spread across them could not be confirmed because the original article was not accessible (Shepard, 1967).
- The design matches Shepard's in its essentials (one exposure, randomly chosen new foils, a two-choice test straight after study), but the scoring does not: people made a recognition decision, while the models' familiarity-like strength signal was read off by the experimenter, learning settings were searched to maximize memory (about 20,000 runs across the study), models were scored on 600 pairs rather than 68, and the sentences differed; a result favouring the side these differences helped cannot be counted (Orhan, 2023).
- No study located tests the recollection route in AI (remember/know judgments, an ROC recollection estimate, or retrieving the study context of a recognized item); the nearest is a 2024 study that trained a language model on synthetic documents to cite the document a fact came from, a source-memory task with no human baseline, so a familiarity-only AI signal is being set against a human score that can draw on both routes (Khalifa et al., 2024).
- Recognition of text met once in full-scale pretraining is far weaker: across Pythia models (2023) of 160 million to 12 billion parameters trained on the Pile, familiarity-based detectors of whether a passage was in the training data performed near chance (AUC below 0.6, where 0.5 is chance) in most domains; no human baseline was run (Duan et al., 2024).
- When chat models made an explicit choice of the verbatim passage from a likely-trained book among four options (the original plus three paraphrases, 25% chance), ChatGPT scored 72.0% and Claude 2.1 (2023) 73.4%; the 10 human raters had not studied the books and scored near chance, so this checks the task, not human memory (Duarte et al., 2024).
- In the current conversation, an item recognition test run on Pythia models (14 million to 12 billion parameters) and GPT-2, scored from how strongly the model predicted list items rather than from yes/no answers, found strong primacy but weak or absent recency, unlike people; the list stayed visible to the model and no human data were collected (Clark & Schuler, 2026).
- For saved memory, benchmarks test only something close: LongMemEval includes 30 false-premise questions requiring a chat assistant to recognize that information never appeared in past sessions, with no human baseline; with recognition measured in training and the conversation but not saved memory, and familiarity but not recollection, the AI level is indirect and no comparison with documented human results has yet been made (Wu et al., 2025).
Where a Mechanism Can Live
04 Product
The whole configuration, and the only thing a benchmark observes directly.
01
Weights
Across deployments
02
Context Window
Within a session
03
Memory Layer
Across sessions
Which of the three a result belongs to is inferred from it, not seen in it.
Read more
Hippocampus
Cortex in the hippocampal formation
A curved allocortical structure in the medial temporal lobe whose bilateral damage reliably produces dense anterograde amnesia in humans.
A paired, seahorse-shaped fold of three-layered cortex running front to back inside the temporal lobe, connected to the rest of the brain mainly through entorhinal cortex and the fornix. What it does is genuinely contested and has been for fifty years: the human lesion literature reads it as necessary for forming new declarative memories, the rodent recording literature as a map of space, and a third position treats both as instances of binding arbitrary elements into relations. The accounts differ on which species and which task they explain best, and the question of which is right remains open.
Medial temporal lobe
Cortex in the temporal lobe
The inner fold of the temporal lobe (hippocampus, the cortices wrapped around it, and the amygdala), removed together in the classic amnesia cases.
A regional grouping rather than a single structure: the hippocampal formation, the entorhinal, perirhinal and parahippocampal cortices that feed it, and, on some definitions, the amygdala. It became the canonical "memory region" because the surgeries and infarcts that produce dense anterograde amnesia (H.M. most famously) take out most of it at once. Whether the parts constitute one memory system or several with different jobs is an open argument, and lesions in patients often cross the boundaries that would settle it.
Parahippocampal cortex
Cortex in the parahippocampal gyrus
The posterior part of the parahippocampal gyrus, responding strongly to scenes and spatial layouts and feeding medial entorhinal cortex.
Cortex behind the entorhinal region on the medial temporal surface, connected to retrosplenial and posterior parietal areas and projecting into medial entorhinal cortex. Human imaging finds a patch here that responds far more to images of places than of objects or faces: the parahippocampal place area. What that response encodes is unsettled: candidate accounts include scene geometry, spatial context for navigation, and the presence of large landmark-like objects, and each explains some of the contrast that named the region.
Parahippocampal gyrus
Cortex in the medial temporal lobe
The medial temporal gyrus below the hippocampus, running from the uncus back to the isthmus of the cingulate gyrus.
The parahippocampal gyrus forms the medial edge of the temporal lobe's ventral surface, bounded laterally by the collateral sulcus and above by the hippocampal fissure. Its anterior part carries entorhinal cortex, which supplies the hippocampus with most of its cortical input through the perforant path; its posterior part is parahippocampal cortex proper, which continues into retrosplenial cortex at the isthmus. The uncus is its hooked front tip. Posterior parahippocampal cortex contains the parahippocampal place area, activated by scenes and spatial layouts; whether that response is about scenes, space, or contextual associations more generally is debated.
Perirhinal cortex
Cortex in the medial temporal lobe
Cortex along the rhinal sulcus that relays object information to the hippocampus and is itself implicated in item familiarity.
Areas 35 and 36, flanking the rhinal sulcus and receiving heavy input from ventral visual stream regions carrying object identity. It projects onward to lateral entorhinal cortex and so into the hippocampus. The contested question is whether it is only a relay: one account holds that perirhinal cortex supports familiarity judgments independently of hippocampal recollection, which would split recognition memory into two processes with two substrates; another treats the dissociation as an artifact of memory strength. Both camps have patient data.
Cao et al. (2025)
Analyzing Memory Effects in Large Language Models through the lens of Cognitive Psychology
In
arXiv
Abstract
Memory, a fundamental component of human cognition, exhibits adaptive yet fallible characteristics as illustrated by Schacter's memory "sins".These cognitive phenomena have been studied extensively in psychology and neuroscience, but the extent to which artificial systems, specifically Large Language Models (LLMs), emulate these cognitive phenomena remains underexplored. This study uses human memory research as a lens for understanding LLMs and systematically investigates human memory effects in state-of-the-art LLMs using paradigms drawn from psychological research. We evaluate seven key memory phenomena, comparing human behavior to LLM performance. Both people and models remember less...
Find it
Diana et al. (2007)
Imaging recollection and familiarity in the medial temporal lobe: a three-component model
In
Trends in Cognitive Sciences, 11(9), 379-386
Abstract
The medial temporal lobe (MTL) plays a crucial role in supporting memory for events, but the functional organization of regions in the MTL remains controversial, especially regarding the extent to which different subregions support recognition based on familiarity or recollection. Here we review results from functional neuroimaging studies showing that, whereas activity in the hippocampus and posterior parahippocampal gyrus is disproportionately associated with recollection, activity in the anterior parahippocampal gyrus is disproportionately associated with familiarity. The results are consistent with the idea that the parahippocampal cortex (located in the posterior parahippocampal...
Find it
Duan et al. (2024)
Do Membership Inference Attacks Work on Large Language Models?
In
arXiv
Abstract
Membership inference attacks (MIAs) attempt to predict whether a particular datapoint is a member of a target model's training data. Despite extensive research on traditional machine learning models, there has been limited work studying MIA on the pre-training data of large language models (LLMs). We perform a large-scale evaluation of MIAs over a suite of language models (LMs) trained on the Pile, ranging from 160M to 12B parameters. We find that MIAs barely outperform random guessing for most settings across varying LLM sizes and domains. Our further analyses reveal that this poor performance can be attributed to (1) the combination of a large dataset and few training iterations, and...
Find it
Duarte et al. (2024)
DE-COP: Detecting Copyrighted Content in Language Models Training Data
In
arXiv
Abstract
How can we detect if copyrighted content was used in the training process of a language model, considering that the training data is typically undisclosed? We are motivated by the premise that a language model is likely to identify verbatim excerpts from its training text. We propose DE-COP, a method to determine whether a piece of copyrighted content was included in training. DE-COP's core approach is to probe an LLM with multiple-choice questions, whose options include both verbatim text and their paraphrases. We construct BookTection, a benchmark with excerpts from 165 books published prior and subsequent to a model's training cutoff, along with their paraphrases. Our experiments show...
Find it
Khalifa et al. (2024)
Source-Aware Training Enables Knowledge Attribution in Language Models
In
arXiv
Abstract
Large language models (LLMs) learn a vast amount of knowledge during pretraining, but they are often oblivious to the source(s) of such knowledge. We investigate the problem of intrinsic source citation, where LLMs are required to cite the pretraining source supporting a generated response. Intrinsic source citation can enhance LLM transparency, interpretability, and verifiability. To give LLMs such ability, we explore source-aware training -- a recipe that involves (i) training the LLM to associate unique source document identifiers with the knowledge in each document, followed by (ii) an instruction-tuning stage to teach the LLM to cite a supporting pretraining source when prompted....
Find it
Orhan (2023)
Recognition, recall, and retention of few-shot memories in large language models
In
arXiv
Abstract
The training of modern large language models (LLMs) takes place in a regime where most training examples are seen only a few times by the model during the course of training. What does a model remember about such examples seen only a few times during training and how long does that memory persist in the face of continuous training with new examples? Here, we investigate these questions through simple recognition, recall, and retention experiments with LLMs. In recognition experiments, we ask if the model can distinguish the seen example from a novel example; in recall experiments, we ask if the model can correctly recall the seen example when cued by a part of it; and in retention...
Find it
Shepard (1967)
Recognition memory for words, sentences, and pictures
In
Journal of Verbal Learning and Verbal Behavior, 6(1), 156-163
Find it
Wu et al. (2024)
LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory
In
arXiv
Abstract
Recent large language model (LLM)-driven chat assistant systems have integrated memory components to track user-assistant chat histories, enabling more accurate and personalized responses. However, their long-term memory capabilities in sustained interactions remain underexplored. We introduce LongMemEval, a comprehensive benchmark designed to evaluate five core long-term memory abilities of chat assistants: information extraction, multi-session reasoning, temporal reasoning, knowledge updates, and abstention. With 500 meticulously curated questions embedded within freely scalable user-assistant chat histories, LongMemEval presents a significant challenge to existing long-term memory...
Find it
Yonelinas (1994)
Receiver-operating characteristics in recognition memory: Evidence for a dual-process model.
In
Journal of Experimental Psychology: Learning, Memory, and Cognition, 20(6), 1341-1354
Abstract
Evidence is presented that recognition judgments are based on an assessment of familiarity, as is described by signal detection theory, but that a separate recollection process also contributes to performance. In 3 receiver-operating characteristics (ROC) experiments, the process dissociation procedure was used to examine the contribution of these processes to recognition memory. In Exps 1 and 2, reducing the length of the study list increased the intercept (d') but decreased the slope of the ROC and increased the probability of recollection but left familiarity relatively unaffected. In Exp 3, increasing study time increased the intercept but left the slope of the ROC unaffected and...
Find it
Yonelinas (2002)
The Nature of Recollection and Familiarity: A Review of 30 Years of Research
In
Journal of Memory and Language, 46(3), 441-517
Abstract
To account for dissociations observed in recognition memory tests, several dual-process models have been proposed that assume that recognition judgments can be based on the recollection of details about previous events or on the assessment of stimulus familiarity. In the current article, these models are examined, along with the methods that have been developed to measure recollection and familiarity. The relevant empirical literature from behavioral, neuropsychological, and neuroimaging studies is then reviewed in order to assess model predictions. Results from a variety of measurement methods, including task-dissociation and process-estimation methods, are found to lead to remarkably...
Find it
Yonelinas et al. (2002)
Effects of extensive temporal lobe damage or mild hypoxia on recollection and familiarity
In
Nature Neuroscience, 5(11), 1236-1241
Abstract
Memory for past events can be based on recollection or on assessments of familiarity. These two forms of human memory have been studied extensively by philosophers and psychologists, but their neuroanatomical substrates are largely unknown. Here we examined the brain regions that are involved in these two forms of memory by studying patients with damage to different temporal lobe regions. Our results come from (i) structural covariance modeling of recall and recognition, (ii) introspective reports during recognition and (iii) analysis of receiver operating characteristics. In sum, we found that the regions disrupted in mild hypoxia, such as the hippocampus, are centrally involved in...
Find it