Selective attention determines which information we consciously process from the environment. Broadbent's filter theory proposes that attention acts as a bottleneck early in processing, while later theories suggest the bottleneck can occur at semantic levels depending on task demands.
The central puzzle of selective attention is that the brain receives far more sensory information than it can fully process — you are surrounded by dozens of sound sources, visual objects, and bodily sensations at any moment, yet experience a coherent, focused perceptual world. Selective attention is the mechanism that resolves this: it selects a subset of incoming information for full conscious processing while filtering or attenuating the rest. Filter models attempt to specify *where* in the processing hierarchy this selection happens, with major theoretical disagreements about whether filtering is early (before meaning is extracted) or late (after meaning is processed).
Broadbent's Filter Theory (1958), the founding model, was motivated by the dichotic listening task: participants wear headphones with different auditory streams in each ear and are asked to "shadow" (repeat aloud) one ear while ignoring the other. Broadbent proposed that the filter operates on physical features — channel identity, pitch, location — *before* semantic content is processed. Only the attended channel passes through the filter; the other is blocked at a pre-semantic bottleneck and never reaches the level of meaning. This early selection view is computationally elegant: filtering early reduces the processing load most efficiently. But it was immediately challenged by the cocktail party effect: people reliably notice their own name spoken in the supposedly unattended channel — a finding that requires the unattended channel to have been processed to at least the level of semantic identity.
Treisman's Attenuation Theory (1964) modified the model to handle this finding: rather than blocking the unattended channel completely, the filter *attenuates* it — reduces its signal strength without eliminating it. Words with high personal relevance (your name, words like "fire" or "danger") have low thresholds for conscious access, so they break through even attenuated. This preserves the efficiency of early filtering while accounting for the cocktail party effect. Late selection theories (Deutsch & Deutsch, Norman) went further: all inputs are processed to the semantic level automatically, and selection happens only at the stage of action or conscious response. On this view, the bottleneck is not perceptual but post-perceptual — the question is not what gets recognized but what gets *responded to*. Evidence from the implicit processing of unattended information (semantic priming from ignored words) supports this position.
The contemporary resolution is that the locus of selection is flexible rather than fixed. The perceptual load hypothesis (Lavie) proposes that when the attended task has high perceptual load — consuming most available perceptual processing capacity — selection happens early and distractors are not processed. When perceptual load is low, spare capacity spills over to process distractors, producing late selection. This explains discrepant findings across paradigms: studies using simple, low-load tasks find late selection (distractor interference); studies using demanding high-load tasks find early selection. The filter is not a fixed gate at one location but a dynamic allocation of resources that adapts to task demands.
The broader significance of filter models is that they forced cognitive psychology to grapple with the distinction between automatic and controlled processing, between what the brain does *to you* (automatic perceptual registration) and what you do *with your brain* (effortful selective allocation). Every filter theory is implicitly a theory about the architecture of consciousness: it locates where the transformation from unconscious registration to conscious experience occurs, and what determines which representations make the crossing. The debate between early and late selection, still not fully resolved, is at its core a debate about the relationship between perception and awareness — one of the deepest questions in cognitive science.