Bilingual memory¶
The interacting organization of lexical forms, conceptual knowledge, and episodic retrieval across two languages, shaped by proficiency, acquisition history, context, and language of encoding and recall.
Core Idea¶
Bilingual memory is the interacting organization of lexical forms, conceptual knowledge, and episodic retrieval across a person’s two languages. It concerns how words connect across languages and with meaning, how those links are used during translation and recall, and how proficiency, acquisition history, usage context, and encoding language alter access.
Influential models distinguish language-specific lexical representations from conceptual representations. The Revised Hierarchical Model proposes asymmetric access for many late bilinguals: early second-language words may reach concepts partly through strong links to first-language words, while proficiency strengthens direct conceptual access. This is a testable model, not one final architecture for every bilingual.
How would you explain it like I'm…
Two Words, One Idea
Two-Language Word Web
Word-to-Meaning Links Across Languages
Scope of Application¶
The abstraction is used in psycholinguistics, cognitive psychology, neuropsychology, language learning, translation, and autobiographical-memory research. It covers lexical–conceptual access and language-dependent episodic cues.
It is not bilingualism as a whole, the bilingual lexicon alone, or a brain-imaging subtraction method. Shared activation does not prove identical representations, and different activation does not prove separate stores. Simultaneous and sequential bilinguals, scripts, modalities, dominance, and switching contexts require separate consideration.
Clarity¶
The abstraction separates word-form association, conceptual mediation, and episodic context. Translation direction can recruit different access routes; semantic interference can reveal conceptual involvement; and reinstating the language of an event can aid recall without implying that the event exists only in that language. These distinctions turn “language affects memory” into testable claims.
Concrete and abstract translation equivalents may also overlap differently in meaning and cultural association. A model should therefore predict patterns across carefully matched items, not treat dictionary equivalence as psychological identity. Reaction time, error, priming, and recall each constrain a different part of the proposed access structure.
Manages Complexity¶
Two languages create many mappings among forms, meanings, contexts, and experiences. Models compress these into stores, links, strengths, and pathways. This supports prediction but can mislead when unlike bilingual histories are averaged into a fictitious architecture. Participant profiles and task demands must remain part of the explanatory model.
Abstract Reasoning¶
Specify languages, acquisition age, proficiency, dominance, and use. Choose tasks that discriminate lexical association, conceptual mediation, or episodic cueing. Compare translation direction, semantic context, reaction time, error, and recall while controlling frequency, cognate status, and task switching. Use converging behavioral and neural evidence, and ask whether the proposed architecture predicted the asymmetry before rather than after observing it.
Knowledge Transfer¶
The role map transfers among bilingual populations only after link strengths and histories are re-estimated. It can extend to multilingual memory, but additional languages introduce new mappings rather than simple copies. General principles such as encoding specificity transfer broadly; language-specific cues and proficiency effects remain domain accents. A finding transfers literally only when the receiving group and task preserve the relevant access conditions. Otherwise it remains analogy.
Neighborhood in Abstraction Space¶
Bilingual memory sits in a sparse region of the domain-specific corpus (94th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.
Family — Unclustered & Miscellaneous (2551 abstractions)
Nearest neighbors
- Cohort Model — 0.80
- Fan Effect — 0.80
- Underlying Representation — 0.78
- Lexical Definition — 0.78
- Levels-of-Processing Effect — 0.77
Computed from structural-signature embeddings · 2026-10-08