Language and Memory: Word Identification, Perception, and Acquisition

Vocabulary flashcards covering memory systems, lexical access, word frequency, contextual diversity, interactive activation, speech perception, and connectionist models of language acquisition based on the lecture.

Episodic Memory

Memory for personally experienced events, arbitrary associations, and specific study episodes (e.g., answering 'When did you meet this person?'), often characterized by the subjective experience of mental time travel.

Semantic Memory

Memory for general world knowledge and word meanings; its retrieval is typically automatic, lacks mental time travel, and exhibits greater resistance to brain damage and forgetting than episodic memory.

Lexical Decision Task

An experimental procedure in which participants are shown letter strings and must rapidly decide whether each string is a word or a nonword, with response time (RT) serving as the primary dependent variable.

Repetition Priming

The finding that lexical decision response times are faster for repeated words than for non-repeated words, even when intervening items separate the repetitions.

Semantic Priming

Faster response times and facilitated access for a target word when it is immediately preceded by a semantically related prime word (e.g., responding faster to 'doctor' after seeing 'nurse').

Spreading Activation

A network retrieval mechanism where accessing a concept node increases its activation level, which then diffuses along links to related words in the mental lexicon and decays over time.

Word Frequency Effect

The phenomenon where words occurring with high frequency in natural language are accessed, read, and decided upon faster than low-frequency words.

Corpus Analysis

A method of establishing word frequency by tallying word occurrences across large text collections, historically printed books (Kučera & Francis, 1967) and currently film subtitles (SUBTLEX) or social media feeds.

Mixed-List Paradox

The empirical observation in free recall where high-frequency words show a strong recall advantage on pure lists (100% HF or 100% LF), but this frequency advantage is markedly reduced or eliminated on mixed lists containing both HF and LF words.

Mirror Effect

A recognition memory pattern where low-frequency words yield both a higher hit rate ('yes' responses to studied targets) and a lower false alarm rate ('yes' responses to unstudied foils) compared to high-frequency words.

Context Variability

A linguistic measure defined as the number of distinct documents or contexts in which a word appears; it accounts for lexical decision latencies better than overall word frequency.

Rational Analysis

John Anderson's theoretical framework proposing that memory and cognition are optimized according to environmental need probability, explaining why words useful across many contexts (high context variability) receive processing advantages.

Spacing Effect

The memory benefit observed when repetitions of an item are distributed across time rather than presented massed consecutively.

Graceful Degradation

A characteristic of neural networks and the brain where partial damage or severed connections causes gradual, localized loss of specific items rather than the catastrophic failure of entire rule systems.

Interactive Activation Model

A computational neural network model developed by McClelland and Rumelhart (1981) that accounts for word perception via bidirectional flow of activation and inhibition across feature, letter, and word layers.

Lateral Inhibition

Mutual competitive inhibition among units within the same layer (such as words competing in the interactive activation model), ensuring that only the most active item dominates perception.

Bottom-Up Processing

Sensory-driven processing wherein raw physical features (e.g., visual strokes or acoustic frequencies) directly activate corresponding letter or phoneme representations, which subsequently activate word units.

Top-Down Processing

Expectation- or knowledge-driven processing in which higher-order representations (such as words in the mental lexicon) feed activation back down to support and constrain lower-level perceptual units (such as letters).

TRACE Model

An interactive activation model of speech perception created by McClelland and Elman (1986) consisting of feature, phoneme, and word layers where phonemic input is processed sequentially across time.

Phoneme

The basic distinctive sound unit of a spoken language, which maps onto acoustic features rather than 1-to-1 onto orthographic letters.

Right Context Effect

A phenomenon in speech perception where acoustic information heard subsequent to an ambiguous or omitted sound resolves and alters the perception of what came before (e.g., hearing '-ift' resolved as 'gift').

Speech Segmentation

The cognitive process of parsing continuous acoustic speech streams—which lack physical acoustic pauses between words—into distinct, recognizable lexical items.

Poverty of the Stimulus

Noam Chomsky's argument that the language input children receive is too limited and impoverished to account for their rapid, rule-consistent linguistic output without an innate language mechanism.

Universal Grammar

Chomsky's proposed innate, biologically determined structural blueprint for language to which the specific grammar and syntax of all natural human languages map.

Syntax

The formal linguistic rules governing word order and grammatical structure, which Chomsky posited are processed prior to indexing lexical semantics during sentence comprehension.

Parallel Distributed Processing (PDP)

A connectionist computational approach where knowledge is distributed across network connection weights, and learning occurs through iterative error correction between model predictions and target inputs without explicit rules.

Wickelfeatures

Representational letter-trigram units (e.g., breaking 'Foster' into FOS, OST, STE, and TER) utilized in Rumelhart and McClelland's (1986) model to encode root and past tense verbs.

Double Dissociation

A neurological pattern where damage to one brain system selectively impairs function A while sparing function B, and damage to another system selectively impairs function B while sparing function A (e.g., Alzheimer's impairing irregular past tense and Parkinson's impairing regular past tense).

Deep Learning Models

Multi-layered artificial neural network architectures (typically containing 10 to 20 or more layers) trained on large datasets to perform complex perceptual and language production tasks.