target \\?as randomly enlbcdded ncar the middle of a stream of 25 numerals. At the cnd of stinlulus presentation, subjccts typed .. the earlicst occurring numeral they could remember as wcll as the thrcc subsequent oncs. T h e four rccalled numerals comprised the rcsponsc scqucncc. This rccall task enablcd us to probe thc distribution of attention over a longer tmmc perlod [and at no additional cost (I.?)] than would be possible with just onc recalled numeral. Over 10,000 trials were conducted with threc subjects. Rcprese~~tative data from all thrcc are shown (Fig. 2, a-i). Kccall data wcre analyzd in terms o f the estimated probabilities p(tj,k) of reporting a numeral from frame j (timc t j ) in position k of thc rcsponsc. Thc target occurred at time t,, = 0 . We first considcrcdp(t,), thc probability of reporting a stimulus item from time t, anywhcrc in thc response (the cnvelope curves o f Fig. 2):

Dynamics of Automatic and Controlled Visual Attention

The time course o f attention was experimentally observed using two kinds o f stimuli: a cue to begin attending o r to shift attention, and a stimulus to be attended. Precise measurements o f thc time course o f attention show that it consists o f two p a r t i d y concurrent processes: a fast, effortless, automatic process t h a t records t h e cue a n d its neighboring events; and a slower, effortfd, controlled process that records the stimulus to be attended a n d its neighboring events.

A

mscc after the targct. The onset time of a al attcntion is attcntional gating ( I ) , rcmernbcrcd nu~ncraldcfined the attentionthe proccsswhcrcby some incoming rcaction timc. Whcn targct difliculty o r tarinfor~nntionis sclccted for further analysis o r get probabiliy was altcrcd, thc attcntionfor memorization, while othcr infornmation rcaction time varied much the same \\lay as is ignored o r attcnuatcd and lost (2-8). the motor-reaction timc (13). In this invcstigation we studied an addiNormally, cyc movements and attcntional shifi-rs arc tightly coupled. In this report wc tional attentional process that had complcte- Thc basic finding from the control espcridiscuss the attentional processes that can ly different properties, using two proce- mcnt is shown in Fig. 2a. The numcral occur \vhilc the cycs arc stationary (9, 10). dures. In both procedures, the subjcct main- simulta~lcouswith the targct and the one Early in our i~wcstigatiot~s of thc dynam- taincd visual fixation on a numeral strcam following it werc almost nevcr r e p r t c d . ics of attcntional gating, we noticed curious throughout thc trial. Proccdurc 1 (Fig. l a ) , Rcpottcd nurncrals occurrcd 200 t o 700 himodal distributions of attention shift an adaptation of thc RSVP attcntion shift msec aftcr the targct. The trace of the first times that suggcstcd that \vc wcrc obscnling paradigm, scrved as a control condition. reported numcral showed that 50% of the not mcrcly a single act of attention but nvo Subjects began a trial by attending t o a letter time it was the numcral occurring 300 nmscc consccutivc, partially overlapping acts. Dur- stream t o thc left of thc numeral strcanl. aftcr the target; thc remaindcr o f the timc it ing the investigation of thcsc phcnomcna They wcrc instructed to shift attcntion from was thc nurncral occurring 200 or 400 mscc ( I I), we learned how to attain scparatc, and the letter strcaln to the numcral strcam as later. In all ~ignific~mt respects, thesc data almost indepcndcnt, control of thc time quickly as possiblc o n detection of thc target rcplicatcd the data o f Rcevcs and Spcrling coursc of cach attcntional proccss. The first lcttcr C and to retnembcr t l ~ ccarlicst four (1). They indicated that detecting thc target proccss is a quick, effortless, automatic pro- numerals they could (simdtaneous with o r C in thc lettcr stream to thc left of visual ccss triggered by targct dctcctioi1;thc sccond subscqucnt to targct detection). In proce- fixation and shifting attcntion to thc numcris a slower, cffbrtful, controlled process (12) dure 2 (Fig. lb), thc expcrimcntal condi- al stream at visual fixation takes 300 ? 100 whose latcncv dcpcnds on practicc and task tion, therc was only onc stream. Subjccts nlscc. fixatcd a numcral stream in which the target difficulty. Typical data from proccdure 2 (targct The procedure wc uscd to nmcasutc atten- was embedded. T w o kinds of targets were 0 s ) are sho\vn in Fig. 2b. The distribution tion \\?as based o n the rapid scrial visual uscd to open the attcntional gate: (i) outline of thep(tj) clearly is bimodal. Most numerprcscntation (RSVP) attcntion shiti para- square condition (OS), an outline square als wcre recalled from nvo separated periods digm of Sperlirmg and Rccvcs (13). RSVP around onc of the numcrals, and (ii) high- in time, thc first mode bcnvcen 0 and 100 (14, 15) p r d u c c s manv regularly spaccd lighted numcral condition (HN), a bright- mscc aftcr target presentation, the sccond cvcnts that can bc used to resolve the tempo- ening of onc of the nurncrals. modc bcnvccn 300 and 400 tnscc after For cach trial of both procedures, the target presentation. For all subjects, the ral propertics of attentive processes. In the Spcrling and Reevcs procedure, a subjcct monitored a strcanm of alphabetic characters Hashcd one on top of the other to thc left of Fig. 1. (a) Illustration of proccdure a visual fixation for thc presence of a targct 1 (control cspcrimcnt) and (b) proIctter. 011dctection. the subject i~~mediatcly ccdurc 2 (main cxncrimcnr). 011 ,,. . 5 ' I ' . , . . . 0 . 6.600 6 . 600 shificd attention to a numcral strcam at thc face of a carhodc-ray rube, 25 8J.--1.i-.' frames, composcd of lctters and nu! G . '40~ visual fixation and attempted to remember mcrals in prMcdure 1 or numerals T a m . I 7 - . .0 -. 4.400$ $ , cmc31 letter'? c fi thc earlicst o c c ~ ~ r r i nnumcral g from thc I N (, ., . . . - 200 2 . ;ilcmc in proccdurc 2, appcar in numeral *