Enhancing learning and retention with distinctive virtual reality environments and mental context reinstatement

The study investigates how leveraging environmental context during learning—via distinctive virtual reality (VR) environments—and guiding mental context reinstatement at retrieval can enhance acquisition, reduce interference, and improve long-term retention of challenging, confusable material (foreign vocabulary from two phonetically similar languages). Grounded in context-dependent memory theory, the authors hypothesized that learning each language in a unique VR context would reduce cross-language interference and improve delayed retention, particularly for participants who experienced a strong sense of presence in VR. They further hypothesized that mentally reinstating the original learning context prior to recall would enhance performance when tested outside the learning environments, and that neural reinstatement of encoding-context patterns (measured with fMRI) during retrieval would predict better recall. The work aims to balance the benefits of contextual support (“contextual crutch”) with transfer across contexts by using mental reinstatement, thereby informing pedagogy and memory enhancement strategies.

Prior work shows memory is context-dependent: matching study-test contexts improves recall while mismatches impair it (e.g., Godden & Baddeley underwater vs land; room changes; noise level changes). Context effects are stronger for recall than recognition and can be pronounced when similar materials are learned close in time, increasing interference. Distinctive contexts can reduce such interference by providing unique cues, especially when contexts are highly distinctive and item sets per context are small. However, reliance on a single stable context can hinder transfer (“contextual crutch”), which can be mitigated by mental context reinstatement that approximates the benefit of physically returning to the context. VR offers controllable, immersive, and distinctive environments with ecological validity; presence (the subjective feeling of being in the VR place) moderates VR’s mnemonic effects. Prior VR studies have demonstrated context-dependent recall and emphasized presence. The authors position their work to harness distinctive VR contexts and mental reinstatement to optimize learning while promoting transfer outside VR.

Design and participants: Two experiments. Behavioural experiment: N=48 adults (26 female, 18–27 years), randomly assigned to single-context (n=24) or dual-context (n=24) learning conditions. fMRI experiment: N=22 adults (12 female, 19–25 years), all dual-context. Behavioural participants were monolingual English speakers; fMRI participants were bilingual (English plus one other language) to boost baseline learning. Inclusion criteria: minimal prior VR exposure, normal/corrected vision and hearing, no learning disabilities, no substance dependence, no psychotropics. Compensation provided; IRB-approved; informed consent obtained.

Learning materials and VR contexts: Participants learned 80 foreign words spanning two phonetically similar Bantu languages (Swahili and Chinyanja): 10 words unique to Swahili, 10 unique to Chinyanja, and 30 learned in both languages. VR contexts were custom desktop VR environments built with OpenSimulator (rendered via Firestorm Viewer on a 27″ 2560×1440 LED display): (1) Fairyland Garden (bright, verdant, open, outdoor ambience), and (2) Moon Base (dark, enclosed, metallic indoors). Each contained nine named rooms. A separate underwater VR environment was used for instructions and practice. Navigation used mouse/keyboard; headphones with microphone enabled audio delivery and recording.

Procedure overview: Across two consecutive days, each language underwent four encoding rounds (one initial study-only, then three test-study cycles with retrieval practice) with expanding retrieval practice. Day 1: instructions and practice; Context A encoding; Language 1 encoding in Context A (Rounds 1–3); Context B encoding; Language 2 encoding (in Context A for single-context group; in Context B for dual-context group). Day 2: Language 1 and 2 Round 4 (T3). After a delay (10 min in behavioural; ~30 min in fMRI due to scanner setup), participants completed a non-VR test (T4) in the lab or MRI scanner. Day 8: surprise telephone recall test (T5). Presence was assessed post-Day 1 using a 10-item scale (1–5).

Context encoding task: In each of nine rooms per context, participants stood on a marker and performed two full clockwise rotations (720°) while imagining being a tourist trying to remember being in that place; room names were announced by the experimenter as participants transitioned between rooms.

Language encoding task: In each round, participants followed a predetermined path to 40 pedestals with 3D objects representing target words. Round 1: read English name aloud, click object to reveal transliteration and hear the foreign pronunciation three times over 10 s, repeating aloud each time. Rounds 2–4: attempt verbal recall before clicking to receive feedback (test-study cycles). Object locations were fixed within a language but differed between languages to avoid simple spatial mapping across languages.

Non-VR test (T4) with controlled mental reinstatement: 80 trials across 10 runs. Each trial: Ready (1 s), Mental reinstatement (10 s; audio cue naming a specific room; eyes closed; button presses to indicate orientation and rotation progress), Language recall (8 s after a 2 s gap; audio cue “Language: English word”; covert retrieval with success/failure button press; at beep, verbalize foreign word; responses recorded), Imagery vividness rating (2 s), and two brief arithmetic tasks (5 s) as active baseline. Reinstatement condition was congruent if the cued room matched the learning context for that word, or incongruent otherwise; trials intermixed.

Surprise phone test (T5): One week later, participants were called for a purported follow-up interview and then given cued recall for all 80 items (English cue + target language); responses were recorded.

Scoring and outcome measures: Verbal responses scored offline by two trained raters using a phoneme-level partial-credit scheme (correct phonemes/total phonemes). Recall scores computed at T1–T5; intrusions coded when responses matched the other language’s translation or similar items. Retention computed as 1 − mean forgetting index between two tests (e.g., T5 relative to T4), considering only items with non-zero prior recall. Presence mean-split into high vs low for analyses. Behavioural analyses used RM-ANOVA with between-subject factors Context Group (single vs dual) and Presence (high vs low), and within-subject factors Times, Reinstatement.

fMRI acquisition and analysis (dual-context only): Siemens 3T Prisma, multiband EPI (TR=1 s, TE=30 ms, flip=52°, 65 slices, 2 mm isotropic, MB=5). Ten runs (330 volumes each). T1 MPRAGE (0.8 mm). Preprocessing: brain extraction, motion correction, high-pass filter (128 s), registration to MNI (FSL/ANTS), no smoothing. For each trial, time windows extracted for Mental reinstatement (imagery; from self-reported orientation to beep) and Language recall (6 s after cue onset), adjusted for hemodynamic lag.

Multivariate analysis pipeline: (1) Whole-brain searchlight MVPA (radius 4 voxels) on imagery periods to classify reinstatement of Moon Base vs Fairyland Garden using linear SVM (libSVM) with leave-one-run-out cross-validation; select top 2000 voxels per participant as feature mask. (2) Create participant-specific context templates by averaging imagery patterns for each context within the mask. (3) Representational Similarity Analysis (RSA): correlate each word’s language-recall pattern with its learned-context template; Fisher z-transform correlations; within-subject mean split to label trials as high vs low representational fidelity. (4) RM-ANOVA on recall performance with factors Times (T4, T5) × Reinstatement prompt (congruent, incongruent) × RSA fidelity (high, low) × Presence (high, low).

• Initial learning: After two exposures (T2), participants recalled 42% ± 17% of the 80 words; performance remained 42% ± 17% at T3 after overnight, with no group differences (p > 0.05). Learning rate exceeded prior non-VR baseline expectations (22–26%).
• Transfer via mental reinstatement at T4: Congruent reinstatement improved recall relative to incongruent reinstatement (52% ± 18% vs 47% ± 19%; RM-ANOVA p = 0.009, ηp² = 0.31), independent of context group.
• Interference reduction: Dual-context participants produced 38% fewer intrusions than single-context participants (4.09 ± 4.82 vs 6.57 ± 4.69 intrusions out of 80 items; RM-ANOVA p = 0.014, ηp² = 0.13).
• One-week retention (T5 relative to T4), moderated by presence: Among high-presence participants (mean-split), dual-context yielded 92% ± 7% retention vs 76% ± 12% for single-context (interaction p = 0.03, n² = 0.11; simple main effect p = 0.002). Low-presence participants showed no difference (p = 0.47).
• fMRI neural reinstatement (dual-context): Trials with high representational fidelity (greater similarity between recall-time brain patterns and the learned-context template) had higher recall than low-fidelity trials across T4 and T5 (main effect F(1,21) = 13.712, p = 0.001, ηp² = 0.395; mean recall proportion High 0.50 ± 0.17 vs Low 0.45 ± 0.18). Effects held at T4 (F(1,21) = 8.60, p = 0.008, ηp² = 0.29; High 0.56 ± 0.19 vs Low 0.51 ± 0.20) and T5 (F(1,21) = 8.53, p = 0.008, ηp² = 0.29; High 0.44 ± 0.19 vs Low 0.39 ± 0.20).
• Interaction of reinstatement prompt × fidelity: Across T4/T5, F(1,21) = 6.59, p = 0.02, ηp² = 0.24; driven by T5 (simple interaction p = 0.006). After incongruent reinstatement prompts, high-fidelity trials showed a 10.1% T5 advantage (0.45 ± 0.19) over low-fidelity trials (0.35 ± 0.20); this effect was absent for congruent prompts (both 0.43 ± 0.20).

The findings demonstrate that distinctive, immersive VR contexts can be leveraged to optimize learning of interference-prone materials and to support transfer outside the original learning environment through mental context reinstatement. Learning each language in its own unique context reduced cross-language interference and protected against one-week forgetting, but critically, this advantage depended on participants experiencing a strong sense of presence within the VR environments. Mental reinstatement of the original learning context enhanced short-delay recall in non-VR settings, indicating that guided imagery can counteract context change-induced forgetting. Neuroimaging provided mechanistic evidence: greater reinstatement of encoding-context neural patterns during retrieval was associated with better recall immediately and one week later, especially when overcoming incongruent pre-recall prompts—suggesting that successful, high-fidelity reinstatement under challenge acts as a desirable difficulty that strengthens memory. These results highlight the central role of spatial context as a scaffold for episodic retrieval and suggest practical strategies for education and training: pairing confusable materials with distinctive contexts and training learners to reinstate those contexts during recall can enhance retention and reduce interference.

This work shows that combining distinctive VR learning environments with guided mental context reinstatement can enhance acquisition, reduce interference, and substantially improve delayed retention of challenging foreign vocabulary, achieving up to 92% one-week retention among high-presence learners in dual-context conditions. fMRI evidence links trial-wise neural reinstatement fidelity of the learned context to successful recall, illuminating a neural mechanism for contextually supported retrieval. Future research should: (1) test more immersive HMD-based VR (and locomotion) to increase presence; (2) examine time course and durability of benefits across multiple delays; (3) evaluate learning without explicit reinstatement cues to assess spontaneous reinstatement; (4) incorporate ecologically relevant contexts (e.g., domain-relevant environments) to potentially amplify effects; (5) measure and manipulate engagement, motivation, interest, and agency alongside presence; and (6) increase sample sizes to study individual differences and generalizability.

• Mental reinstatement was explicitly cued before each recall trial, limiting inferences about spontaneous reinstatement and ecological generalizability without cues.
• fMRI analyses used whole-brain multivariate patterns with participant-specific voxel selection; this approach increases sensitivity but limits conclusions about specific brain regions’ causal roles.
• Presence moderated behavioural benefits; desktop VR may have limited presence compared with HMD-based systems, potentially underestimating effects.
• Different language backgrounds across experiments (monolinguals in behavioural vs bilinguals in fMRI) may affect generalizability of neural findings to monolingual populations.
• Sample sizes, while adequate for observed effects, limit fine-grained individual differences analyses.
• The VR contexts were arbitrary with respect to language content; effects might differ with semantically or practically relevant contexts.
• Engagement, intrinsic motivation, and agency were not quantitatively measured, yet may contribute to learning outcomes.