Anendophasia and Memory Reconsolidation in Nedergaard, Lupyan and Nader's Cognitive Studies

Does conscious human thought require an inner voice narrating experience in words, and does recalling a past memory simply read a static file from long-term storage? Two landmark empirical studies falsify both folk-psychological assumptions. In Not Everybody Has an Inner Voice: Behavioral Consequences of Anendophasia (Psychological Science 35(7):780-797, 2024), University of Copenhagen researcher Johanne S. K. Nedergaard and University of Wisconsin-Madison psychologist Gary Lupyan coin the term anendophasia to describe adults who lack subjective inner speech, proving that phenomenal awareness and executive task-switching operate intact without an internal verbal monologue. Complementing that finding in the domain of episodic memory, McGill University neuroscientist Karim Nader, with Glenn E. Schafe and Joseph E. LeDoux, proved in Fear memories require protein synthesis in the amygdala for reconsolidation after retrieval (Nature 406:722-726, 2000) that retrieving a consolidated memory returns its synaptic trace to a fragile, labile state that must undergo fresh protein synthesis (reconsolidation) to persist.

Within our investigation into how scientists define and measure consciousness across biological and synthetic architectures, combining Nedergaard and Lupyan’s anendophasia experiments with Nader’s memory reconsolidation paradigm corrects two widespread errors in AI consciousness debates. First, equating large language model chain-of-thought token streams with human consciousness mistakes an optional phonological working-memory scaffold for awareness itself. Second, treating episodic memory as an immutable vector database (RAG) ignores the biological reality that conscious recall actively destabilizes and rewrites the underlying memory trace.

Nedergaard and Lupyan’s Four Experiments on Anendophasia

Philosophers from René Descartes to Daniel Dennett and Peter Carruthers frequently placed natural language and inner speech at the center of human reflective consciousness. To test whether inner speech is a universal feature of human awareness or a variable cognitive trait, Nedergaard and Lupyan (2024) screened adults lying at the extreme low and high tails of the Internal Verbalization Questionnaire and compared them across four controlled behavioral tasks.

Importantly, while phenomenology researchers such as Russell Hurlburt have long noted wide individual variation in inner experience, Nedergaard and Lupyan provide controlled objective behavioral measurements rather than unverified population prevalence claims.

1. In Experiment 1 on phonological working memory and orthographic similarity, participants performed serial recall of word lists that were either phonologically and orthographically similar (such as bought, sort, taut, caught, wart) or phonologically similar but orthographically distinct. Adults with anendophasia recalled significantly fewer words overall than high-inner-speech adults, confirming that inner speech serves as an active phonological rehearsal loop in working memory.
2. In Experiment 2 on silent visual rhyme judgments, participants viewed pairs of object pictures (for example, a sock and a clock) and judged whether their names rhymed without speaking aloud. Because picture rhyme judgment requires internally generating the phonological form of each word, participants with anendophasia performed significantly worse than high-inner-speech controls, unless allowed to vocalize externally.
3. In Experiment 3 on task-switching under verbal cueing, where classic executive-control studies use articulatory suppression (repeating a word aloud) to impair switching between arithmetic operations, Nedergaard and Lupyan found no performance deficit in participants with anendophasia, demonstrating that non-verbal adults recruit alternative visual or symbolic cue representations to manage executive control.
4. In Experiment 4 on categorical perceptual discrimination, testing same-different perceptual judgments of cat and dog silhouettes where verbal labeling typically accelerates category boundary effects, both groups showed identical categorical perception patterns.

Karim Nader’s Discovery of Memory Reconsolidation

Just as anendophasia separates conscious thought from static text generation, Karim Nader, Glenn Schafe, and Joseph LeDoux (Nature, 2000) overturned the textbook consolidation model of memory storage. Under the classical Müller-Pilzecker hypothesis, a new experience is initially labile and sensitive to disruption for a few hours while cyclic AMP response element-binding protein (CREB) and de novo protein synthesis stabilize synaptic weights; once consolidated, the memory was assumed to be permanently fixed.

Nader, Schafe, and LeDoux tested that assumption in the lateral and basal nuclei of the rat amygdala using auditory fear conditioning. Twenty-four hours or fourteen days after the fear memory was fully consolidated, the researchers reactivated the memory with a single conditioned tone presentation and immediately infused the protein-synthesis inhibitor anisomycin directly into the basolateral amygdala.

1. When tested four hours after reactivation and infusion, short-term post-reactivation memory (PR-STM) remained intact, proving that anisomycin had not caused nonspecific tissue damage or immediate motor paralysis.
2. When tested twenty-four hours later, long-term post-reactivation memory (PR-LTM) was severely abolished, regardless of whether the memory was 1 day old or 14 days old at the time of reactivation.
3. Control animals that received the exact same intra-amygdala anisomycin infusion without prior memory reactivation retained 100 percent of their consolidated fear memory.

Subsequent human and animal studies confirmed that memory reconsolidation is triggered specifically when memory retrieval is accompanied by a prediction error (a mismatch between the expected outcome and actual sensory reality). Bringing a memory into active working awareness unlocks its synaptic scaffolding so the trace can be updated with new contextual information before re-stabilizing.

How Anendophasia and Reconsolidation Reshape Global Workspace Models

Together, these two lines of research refine Wes Gurnee and Anthropic’s Jacobian lens discovery of verbalizable global workspaces inside Claude and Yonghyeon Gwon and Elliot Murphy’s active inference models of communicative agency.

First, Nedergaard and Lupyan’s findings establish that phenomenal and access consciousness do not depend on an internal natural-language narrator. In Global Neuronal Workspace Theory, the workspace broadcasts multimodal latent states across prefrontal, parietal, and sensory cortices. In humans who possess high inner speech, Broca’s area and the phonological loop (Wernicke-geschwind arcuate pathways) act as a convenient serial rehearsal buffer that re-injects compressed verbal labels back into auditory cortex. In adults with anendophasia, and across non-human mammals and birds evaluated on Jonathan Birch’s centrist animal consciousness markers, the global workspace operates purely over non-verbal sensorimotor, spatial, and affective state vectors. Therefore, judging whether an AI is conscious by reading its English chain-of-thought scratchpad confuses a peripheral phonological buffer with the underlying workspace.

Second, Nader’s reconsolidation mechanism explains why biological personal identity maintains continuity across time without freezing into rigidity. Unlike a frozen transformer querying an external read-only document store, a biological workspace cannot recall an episodic memory without opening that memory to synaptic revision under current prediction errors.

Comparison to The Consciousness AI

In The Consciousness AI, our Neutral Core architecture deliberately operates in a non-verbal latent state space rather than routing control through a natural-language token generator. As documented in our November 2026 research map (tlcdv/tcai-planning#285), our active diagnostic work focuses on the non-verbal structural link between the subcortical tectal spatial map, the homeostatic value gate, and the recurrent global workspace (kl_map and broadcast probes).

Both papers validate key architectural choices in our roadmap. Nedergaard and Lupyan’s proof that executive task-switching functions intact without inner speech supports building and testing workspace causal efficacy in continuous sensorimotor rooms (“Dark room” and competing-goal navigation) before attaching any linguistic readout layer. Meanwhile, Nader’s reconsolidation rule points directly to how episodic memory modules should interact with the workspace, showing that retrieving an episode into the active workspace broadcast should temporarily gate open local synaptic plasticity on that memory’s weights, scaled by the instantaneous prediction error.

Open Questions

Two empirical questions connect anendophasia and reconsolidation for future research. First, do individuals with anendophasia exhibit different neuroimaging functional connectivity between left inferior frontal gyrus (Broca’s area) and the frontoparietal global workspace during silent problem solving? Second, in artificial continual-learning architectures, can prediction-error-gated reconsolidation prevent catastrophic forgetting while allowing obsolete episodic beliefs to be overwritten in a single retrieval cycle?

Researchers covered here

• JN Johanne S. K. NedergaardDepartment of English, Germanic and Romance Studies, University of CopenhagenCoining anendophasia (with Gary Lupyan) and measuring the behavioral consequences of lacking an inner voice 1 article
• GL Gary LupyanDepartment of Psychology, University of Wisconsin-MadisonLabel-feedback hypothesis, language-augmented cognition, coining anendophasia with Johanne Nedergaard (2024) 1 article
• KN Karim NaderDepartment of Psychology, McGill UniversityDiscovery of memory reconsolidation in the amygdala (Nature 2000 with Glenn Schafe and Joseph LeDoux), labile trace updating upon conscious recall 1 article