Human Attention: From William James to Modern Cognitive Models
E N D
Presentation Transcript
Attention • Attention is about choice • Choice can be governed by what’s out there in the world • Choice can be governed by internal mental states • For humans, which is it?
Attention is about choice • What should we attend to? What should we ignore? • Some things are more relevant to our survival – threats, food sources, mates. • But how are we to know which things these are? • On the basis of physical qualities such as colour or direction of movement? • Or on the basis of abstract qualities such as name or category?
William James got there first • James (1890) • distinguished between two different systems: • Focal attention – controlled internally, by intentions and goals • Automatic attention – controlled externally by sensory events that demand attention • We’ll come back to this idea at the end
Choices governed by external world • This is the basic behaviourist view – our behaviour is controlled by things in the world. • The earliest cognitive models of attention adopted largely this approach • First important model was Broadbent’s (1958) Early Filter model.
Broadbent’s Early Filter model • Very large capacity sensory store feeds information through a filter into small capacity short term memory (STM) • The filter responds to physical characteristics such as colour or movement • Filter can be directed to take output of a given sensory channel (e.g., one ear or the other)
Evidence for Broadbent’s model • Dichotic listening studies in ’50s • Shadowing tasks: subjects listen to one track and ignore other track. Later can report on physical aspects (e.g., sex of voice) of ignored track but not on message content. • Broadbent (1954) – Split-span task • Subjects listen to headphones; separate series of 3 digits in each ear. Better at reporting by ear (65%) than by simultaneous pair (20%)
Sensory Register Selective Filter Short-Term Memory Broadbent’s Model • The filter mechanism selects on the basis of physical characteristics only
Broadbent’s Model • Notice the simplicity of the internal mechanism (the filter and STM) • Very little in the way of ‘moving parts.’ • There’s not much more complexity than in standard Behaviourist S-R models. • This model was in trouble pretty soon…
Problems with Broadbent’s Model • Treisman (1960) – meaning of messages can promote selection in dichotic listening task. • Subjects shadow one ear, ignore other. Story being shadowed switches to ignored ear. Most subjects then continue shadowing it for a few words before catching themselves and switching to ‘correct’ ear. • Continuity of story makes words in ignored ear relevant after switch.
Treisman’s (1964) Attenuation Model • Filtering is not all-or-none. Stimuli in SM can be more-or-less filtered out – and can be attended if they are very salient or very familiar. • Classic example – you’re talking to someone at a party and behind you, a person says your name – and you notice. • Notice that what happens inside head now starting to be important.
Problems with Broadbent’s Model • Moray (1960) – subjects detect their own name in ignored ear. • Gray & Wedderburn (1960) – 2 Oxford undergraduates; alternated words in a phrase from ear to ear. Subjects reported phrases intact. • Attended Ear Poor 2 Jane • Unattended Ear 5 Aunt 9 • Response “Poor Aunt Jane 5 2 9”
Corteen & Wood (1972) • First paired 3 city names with electric shock • Checked that conditioned response existed • Presented city names in unattended channel, during standard shadowing task • Stimulus GSR Frequency Shock-paired city names 38% Old words, not shock-paired 12% New city names 23% New words not in learning list 9%
Problems with Broadbent’s Model • Stroop Effect • Task: name colour of ink a word is printed in • Effect: RT is longer if word spells a competing colour. • REDYELLOWGREEN • PINKBLUE BROWN • According to Broadbent, should be able to attend to colour and ignore word.
Late Filter Theory – Deutsch & Deutsch • Deutsch & Deutsch (1963) responded to the many problems with Broadbent’s model by arguing that all stimuli are analyzed for meaning. • They suggested that selection happens only after semantic analysis, and means selection for awareness. • Problem is that while selection fails sometimes, it also succeeds sometimes. • But like Broadbent, D & D believed in role for Central Processor.
Central Processor • Broadbent’s version of STM was revised in Atkinson & Shiffrin’s (1971) Modal Model. CP made more complex. • CP now has many control processes: selection, encoding, rehearsal, and such. • All tasks require access to this central processor • Implication – if several tasks want access to CP at same time, should be cost
Problems with CP idea • Allport et al. (1972) – two complex tasks can be done simultaneously if they don’t share input and output modalities. • Shadowing task – aural input, verbal output • Playing piano (sight reading) – visual input, manual output • No cost to dual-task performance compared to when each task performed alone. • Cannot be only one “central processor”
Problems with CP idea • Spelke et al. (1976) – two complex tasks interfered originally, but after 85 hours of practice, no interference • Write to dictation – aural input, manual output • Read out loud – visual input, verbal output
More complex approaches • By end of 1970s, Johnston & Heinz were advocating a model in which person can choose either early or late selection, depending upon their purpose. • Johnston & Heinz (1978) • Dual task study, varying instructions • Showed subjects could choose either earlier or later selection, but later selection interfered with other processes.
Norman & Shallice (1986) • Supervisory Attentional System • Part of a model of control of thought and action • SAS involved: • in novel situations & poorly learned tasks • in dangerous situations • in troubleshooting • in tasks involving planning & decision making • when habitual responses must be stopped
Norman & Shallice (1986) • Model consists of: • Special purpose cognitive modules (e.g., for object recognition) • Schemas (programs that run on the modules) • Contention scheduling (resolves conflict when several schemas want to run at the same time) • Supervisory Attentional System (which activate or inhibit schemas to bias their selection)
Supervisory Attentional System Control schema Schema trigger database Effector System Contention scheduling N&S 1986 - Supervisory Attentional System Perceptual System
Norman & Shallice (1986) SAS Model • Multiple, autonomous, mutually competing schemas • Schemas are influenced by supervisory attentional system • SAS enhances one schema over others, allowing it to attain dominance and control behaviour • Compare to original Broadbent model for complexity of processes inside the head
Norman & Shallice (1986) • SAS model accounts for many facts about actual behaviour – i.e., action in the world, including slips and errors. • Capture errors – when familiar stimuli are responded to in wrong context • Crosstalk errors – interference between two tasks • Similar to errors caused by damage to frontal lobes
Current state of theory • Since 1986, idea of a single executive controller has become less popular. • Many people now say, idea was just a metaphor taken from politics or management, but brain doesn’t work that way. • Brain seems to work with many subsystems operating in parallel, but no single controller • Homunculus problem
Fractionation of executive control Current view inspired by much research in Neuropsychology and connectionism: control emerges from complex inter-actions among lots of special-purpose modules. • Control is not administered by some agent. • Allport (1989) • Logan (1985)