Attention Issues in Attention Research • Process of focusing • What happens to unattended information? mental effort on • What factors affect our ability to divide our particular task or stimulus attention? –Selective – Shiftable – Divisable –Limited IP Psychology Selective Attention Models Studies of selective attention (e.g. Selective Attention Models Cherry, 1953) • subjects are presented with 2 or more stimuli at the same time – this is called dual-task performance • Early Selection Models – Attention determines nature of info that goes into •for selective attention tasks, subjects are working memory (short-term store) instructed to attend to one input only – Selection occurs “early”, before info enters working memory 1 Dichotic listening task So what do we notice? • Listeners are good at detecting physical info • 2 simultaneous messages – could tell whether unattended channel was… – one to each ear • Male vs. Female • each message has different meaning • Speech vs. Buzzing – like listening to TV & radio at same time • Listeners are bad at detecting meaning info • Subjects must shadow one ear only – couldn’t identify any word/phrase – couldn’t tell whether unattended channel was… – repeat the message from one ear only • Forward vs. Backward • English vs. German • Experimenter then asks about unattended ear! – Listeners extract only certain info from unattended • So, attention is ear – sensitive to physical properties – insensitive to meaning Broadbent’s Filter Model of Broadbent’s Filter Model of selective attention selective attention • sensory information (speech) is received in channels Other channel no longer meaning extracted – channel = ear! processed after filtering… here on after • Info goes first to buffer (=temporary store) • Then filter selects attention (chooses 1 of the channels) I based on physical characteristics (e.g. voice tone) N • This single channel is now the focus of attention P U • Ignored (=unattended) info can be recalled from the buffer T but this decays rapidly! sensory filtering short term – We can only attend to one channel at a time buffer (based on memory – Switching attention between channels requires effort physical characteristics) Broadbent’s Filter Model of (early) Evidence for the model selective attention • The split-span • Subjects must report what • Why ‘early’ model? procedure Broadbent they heard, either: – Imagine: (1954). A) pair-by-pair 48, 95, 62 – 6 digits presented in 3 OR pairs B) ear-by-ear 496, 852 – 1 member of each pair Physical Properties simultaneously meaning presented to each ear: • Participants perform better ear-by-ear L 4 9 6 attention is not here (late) selected here (early) R 8 5 2 2 Why? Evidence Against requires switching back and • Cocktail Party Effect pair-by-pair 48, 95, forth between ears. Model (Moray) 62 says this requires lots of effort – Shadow 1 Channel – Report Salient Words vs. Only need 1 switch… from Un-shadowed therefore less effort… Channel therefore we’re better! •e.g. Own Name •Fire! ear-by-ear 496, 852 While we’re reporting the left ear (496) we hold the right ear info (852) in the temporary buffer Switching Ears (Moray, 1959) Errors in Shadowing • Shadow left, until “switch ears” signal • Treisman (1964) • Attended (yes) Unattended (no) • Subjects sometimes report mixtures of • “Bill switch ears” things from both ears – Attended (yes) – Unattended (yes) Galvanic Skin Response Attention Models • GSR -- indicates increased arousal • Training: shock for certain words • Later: present to unattended ear during shadowing • Late Selection Models • Emotional response to – All info gets into working memory (short-term) word (altho not aware of – Unattended info rapidly forgotten hearing it!) – Selection doesn’t occur until after recognition 3 Norman’s Model Norman’s model & the data • Late Selection • Cocktail Party Effect – All stimuli processed to recognition – Name gets into STM just long enough to be – Bottleneck after recognized recognition just before – Because important, gets extra processing the formulation of response • Contextual Errors in Shadowing • Selection (pertinence) – Info from both ears gets into STM based on importance – Contextually relevant info gets extra • Memory devoted to processing because pertinent selected inputs Deutsch & Deutsch Model of Deutsch & Deutsch Model of (late) selective attention (late) selective attention No filter/No attenuation meaning fully • ALL stimuli are analysed for meaning but only SOME extracted from both are selected for response channels – unattended message may be recognised but not pass I into awareness N • allows for subliminal perception; P U – processing inputs with no conscious awareness T sensory short term • Non-attended inputs are noticed only if their buffer memory relevance exceeds the relevance of the attended Subliminally aware of meaning channel • allows for breakthrough Deutsch & Deutsch’s Model of Evidence for the model (late) selective attention • Why ‘late’ model? • Sophisticated meaning analysis of unattended – Imagine: channel – Mackay (1973). subject shadows this They threw stones towards the bank Physical Properties meaning ……… … …money or not here channel is ……… … …river QUESTION: what does the sentence mean? (early) selected here (late) Subjects favoured the interpretation of the sentence suggested by the unattended word. 4 Treisman & Geffen (1967) Contra Late Selection • Method • Results – Task 1: shadow – Tapping to green in • Treisman & Riley (1969) message in one ear shadowed ear: 87% – Although meaning analysis is good on unattended channel, it’s BETTER on attended channel. – Task 2: tap pencil – Tapping to green in when you hear the the other ear: 8% 7 9 4 6 1 4 … Shadows this. Stop when target word “green” you hear a letter. 3 8 6 3 1 B … • We’re much better at detecting the letter on the attended channel (7 9 4 6 1 B …)! – Therefore…the unattended channel is NOT fully processed. Attenuator Model Treisman’s attenuation theory • a filter attenuates unattended input rather than “turning it • Attenuator affects off” “subjective loudness” – Pay attention – increase • so non-attended meaning does pass on, but in a weaker loudness form. – Shadowed msg “louder” than non-shadowed • because it’s weaker (attenuated) we can ignore it • Concepts in Dictionary differ in loudness • stimuli that are emotionally important (our name) or threshold biologically relevant (“sex”) are pre-set and so not – Name always low attenuated – Thresholds change w/goals – allows for breakthrough Treisman’s Attenuation Model of Problems with the model selective attention Other channel gets meaning extracted • What is an attenuated processing of attenuated … here on after meaning??? • What does it meanRABBIT to ‘partially’ work out meaning?? I N partial meaning P full meaning U ???? T sensory Attenuator short term buffer memory 5 Johnston & Heinz’s Hybrid Early or late selection? Model • We understand at least some meaning on the • We filter out channels either... unattended channel. SO Early selection models are not EASY satisfactory. • But we still get more meaningful info from the attended • EARLY (based on physical (vs. unattended) channel. SO Late selection models are characteristics) OR not satisfactory • LATE (based on meaning) • Perhaps an attenuated theory… But what is attenuated meaning processing??? HARD Perhaps we need an alternative approach? Johnston & Heinz’s Hybrid Model Johnston & Heinz’s Hybrid Model • Evidence: Dual • Results Task – Task 1. – Task 1. Blah blah blah … Attend to this. Blah blah blah … Attend to this. [different voice] … Blah blah blah … – Task 2. Watch screen. Say when a light appears. – Task 2. “NOW!” “NOW!” FAST Summary: Selective Attention Johnston & Heinz’s Hybrid Model Summary: Selective Attention Unattended channel: • Results we don’t know – Task 1. • Early models meaning – select attention based on physical characteristics Blah blah blah … Attend to this. • Late models [different meaning] … – select attention based on meaning Unattended channel: – Task 2. we do know • Attenuation models meaning “NOW!” SLOW – have ‘attenuated’ meaning on unattended channel • Hybrid model seems most satisfactory – Can do both early and late selection - Early selection is easier 6.