Attention, Perception, Learning, Memory and Forgetting

Overview

Information processing begins with selection. The world offers far more stimulation than can be processed deeply, so attention selects, sustains and shifts processing toward relevant input. Perception then organises selected input into objects, depth, movement and meaning, using both sensory data and prior knowledge. Learning is a relatively enduring change in behaviour or knowledge resulting from experience, distinct from temporary changes due to fatigue, maturation or drugs. Memory is the system that encodes, stores and retrieves experience, and forgetting is failure at one of those stages or distortion during reconstruction. Study strategies that respect these stages, such as spaced practice, retrieval practice, elaboration and varied context, work because they strengthen encoding and retrieval paths rather than merely increasing exposure time.

Attention: forms and models

Attention takes several forms. Selective attention focuses on one input while ignoring others, as in following one conversation at a noisy gathering. Sustained attention maintains focus over time, as in monitoring a display for rare signals, and it typically declines with time on task, a vigilance decrement. Divided attention splits processing between tasks, succeeding when tasks are well practised or use different resources and failing when both demand the same limited capacity. Shifting attention moves focus between locations or tasks, with costs in time and errors on switch trials. Overt attention involves eye and head movement, while covert attention shifts processing without moving the eyes. These forms are not interchangeable: a person may sustain attention well yet divide it poorly, and a model must explain that pattern.

Early filter theory proposed that unattended input is blocked early on physical features, explaining why shadowing one ear in dichotic listening leaves little memory for the other ear's content. Attenuation theory softened the filter: unattended input is weakened, not blocked, so highly relevant material such as one's own name can break through. Late selection theory argued that all input is processed to meaning and selection occurs later, explaining semantic effects in unattended channels. Resource models treat attention as limited capacity that can be allocated flexibly, with performance falling when demand exceeds supply. Modern network views distinguish alerting, orienting and executive control networks in the brain, linking attention to arousal, spatial orienting and conflict resolution. The cocktail party effect, the Stroop effect and attentional blink each test these models differently: breakthrough of meaningful unattended material challenges early filters, while dual-task costs challenge unlimited-capacity views.

Perception: approaches, organisation, constancy

Perception is the organisation and interpretation of sensory input into experience of objects and events. Bottom-up approaches emphasise data from the senses: features are detected and combined into wholes. Top-down approaches emphasise prior knowledge, expectation and context: the same sensory pattern is perceived differently when context changes, as in ambiguous figures or context-dependent word recognition. Gestalt principles describe bottom-up organisation: proximity groups near elements, similarity groups like elements, continuity favours smooth paths, closure completes incomplete figures, and common fate groups elements moving together. Figure-ground organisation separates an object from its background, sometimes reversibly, showing that organisation is an active achievement, not a passive copy of the retinal image.

Depth perception combines binocular cues, such as retinal disparity and convergence, with monocular cues, such as relative size, overlap, linear perspective, texture gradient and motion parallax. Perceptual constancy preserves stable experience despite changing sensory input: size constancy keeps an object perceived as the same size as it moves away, shape constancy preserves shape across viewing angles, and colour constancy preserves colour across illumination changes. Illusions reveal the rules of this system by showing where it misapplies them, for example when depth cues suggest a size difference that is not real. Perception is also shaped by culture, experience and motivation: experts perceive patterns novices miss, and needs and expectations bias ambiguous perception, though strong claims that perception is wholly constructed overstate the case. The balanced view is interaction: sensory data constrain perception, and knowledge organises it, with neither alone sufficient.

Signal detection and information processing

Signal detection theory separates two factors that ordinary accuracy confuses: sensitivity, how well a person can distinguish signal from noise, and response bias or criterion, how willing the person is to say yes. In a detection task, a hit is saying yes when a signal is present, a miss is saying no when it is present, a false alarm is saying yes when only noise is present, and a correct rejection is saying no to noise alone. Sensitivity is indexed by d-prime, which grows as hit rate rises relative to false alarm rate. Criterion is indexed by beta or by the criterion location: a liberal criterion produces many hits and many false alarms, a conservative criterion produces few of both. Payoffs, base rates and instructions shift criterion without changing sensitivity: a radiologist who must not miss disease adopts a liberal criterion, accepting more false alarms, while a judge who must avoid false convictions adopts a conservative one.

Information processing models extend this logic to cognition as a whole. The classic stage view treats input as passing through sensory registers, attention, short-term working memory and long-term memory, with encoding, storage and retrieval as distinct operations. Working memory refines short-term memory into a system with a phonological loop for verbal rehearsal, a visuospatial sketchpad for images and locations, a central executive for control and attention, and an episodic buffer integrating information. Capacity limits, chunking and automaticity explain why practice frees resources: well-learned skills need less working memory and therefore interfere less with other tasks. Processing depth also matters: deeper semantic processing produces stronger memory than shallow physical processing, because it creates richer, more distinctive traces and more retrieval routes.

Learning: theories and conditioning

Classical conditioning, studied through Pavlovian paradigms, pairs a neutral stimulus with an unconditioned stimulus that naturally elicits a response, until the neutral stimulus alone elicits a conditioned response. Acquisition is the growth of that response across pairings; extinction is its decline when the conditioned stimulus appears without the unconditioned stimulus; spontaneous recovery is its return after rest; generalisation is responding to similar stimuli; discrimination is responding only to the trained stimulus. Higher-order conditioning pairs a new stimulus with an already conditioned stimulus. Classical conditioning explains much emotional learning, including fear conditioning and conditioned taste aversion, where a single pairing can produce lasting avoidance because the learning is biologically prepared.

Operant conditioning, developed in Skinnerian paradigms, concerns behaviour controlled by its consequences. Reinforcement strengthens behaviour: positive reinforcement adds a pleasant consequence, negative reinforcement removes an unpleasant consequence, and both increase the behaviour, a distinction students must not confuse with punishment. Positive punishment adds an unpleasant consequence and negative punishment removes a pleasant consequence, and both decrease behaviour. Schedules matter profoundly: continuous reinforcement produces fast learning but fast extinction, while partial reinforcement produces slower learning but greater resistance to extinction. Fixed-ratio schedules produce high rates with pauses after reinforcement, variable-ratio schedules produce high steady rates, fixed-interval schedules produce pauses followed by responding near the interval end, and variable-interval schedules produce steady low rates. Shaping builds complex behaviour by reinforcing successive approximations, and chaining links steps into sequences. Cognitive and social learning theories add that learning can occur without direct reinforcement, through observation, modelling and expectation, as shown by latent learning and by children imitating modelled aggression without immediate reward.

Verbal and discrimination learning

Verbal learning studies how words, pairs and lists are acquired, using paradigms such as free recall, serial recall, paired-associate learning and recognition. Serial position effects are central: items at the start of a list benefit from primacy through rehearsal and long-term encoding, items at the end benefit from recency through short-term availability, and middle items suffer most. Recall is generally harder than recognition because recognition supplies cues, while recall requires self-generated retrieval. Meaningfulness, imagery, organisation and elaboration improve verbal learning: concrete, imageable words are remembered better than abstract ones, organised lists better than random ones, and self-generated elaborations better than passive reading. Transfer of learning can be positive when prior learning aids new learning, or negative when it interferes, and proactive and retroactive interference describe old learning disrupting new and new learning disrupting old respectively.

Discrimination learning studies how organisms learn to respond differently to different stimuli. In simple discrimination, one stimulus is reinforced and another is not, until responding tracks the difference. In reversal learning, the reinforced stimulus is switched, testing flexibility. Concept learning is a form of discrimination extended to categories: learners abstract the features that define a category and ignore irrelevant variation, through hypothesis testing, exemplar comparison or prototype abstraction. Error patterns reveal strategy: a learner who persists with an old rule after reversal shows perseveration, while a learner who attends to an irrelevant dimension shows a failure of selective attention rather than of memory. Verbal and discrimination paradigms together show that learning is not mere repetition: attention to relevant dimensions, organisation of material and retrieval practice determine what is actually acquired.

Memory and forgetting

Memory is commonly divided into sensory memory, which holds brief, modality-specific traces; short-term or working memory, which holds a small amount of actively processed information; and long-term memory, which holds vast knowledge over long periods. Long-term memory divides into explicit memory, consciously recalled facts and events, and implicit memory, influence of past experience without conscious recall, including procedural skill, priming and classical conditioning effects. Explicit memory further divides into episodic memory for personal events and semantic memory for general knowledge. Encoding specificity states that retrieval succeeds best when cues at test match cues at encoding, which explains context-dependent memory and the value of studying in conditions similar to testing. Retrieval is reconstructive: memories are rebuilt from traces and schemas, so they can be distorted by suggestion, leading questions and post-event information, as shown by false memory research.

Forgetting has several mechanisms, and naming the right one matters. Decay or trace fading proposes that unused traces weaken with time, which fits some sensory and short-term loss but poorly explains long-term forgetting of well-learned material. Interference proposes that other learning disrupts retrieval: proactive interference from old learning and retroactive interference from new learning, both prominent in verbal learning. Retrieval failure proposes that the trace remains but cues are inadequate or blocked, as in tip-of-the-tongue states and context change, where a better cue restores recall. Motivated forgetting and repression are more contested: people can suppress retrieval strategically, but claims of wholesale unconscious repression of trauma remain controversial and require careful evidence. Amnesia from brain damage distinguishes retrograde loss of past memories from anterograde failure to form new ones. Effective study follows from this science: spaced retrieval practice, interleaving, elaboration and varied cues strengthen traces and retrieval paths, while mere rereading creates familiarity without retrievability.

Conditioning compared

Reinforcement and punishment