,blcmon & CoYnrtrpn

1994 22 r2l~ 133-144

Word superiority over isolated letters :

The neglected case of forward masking
TIMOTHY R. J0RDA1 and KIWI M . BEVA\
Gnmersrty of St Andrews, St Andrews . Scotland

Previous research shows that when briefly presented alphabetic stimuli are followed by pat-
tern mask,, letters in words are reported more accurately than are isolated letters (the "Word-
Letter Phenomenon," or NVLP) ; however, when these masks are replaced by blank fields, the w'LP
disappears . These findings have led to the popular notion that the WLP reflects selective mask-
ing of ongoing stimulus processing and so criticallv depends on the use of poststimulus masks .
Here we report three experiments which re-examine the role of masking in the WLP by contrast-
ing the effects of postmarked displays with the effects of premasked displays in which words and
isolated letters were preceded by a pattern mask and followed by a completely blank field . Despite
the critical role generally assigned to poststimulus pattern masks, similar WLPs were obtained
with pre- and postmarked displays . Implications for theories of word and letter recognition are
discussed.

By limiting the amount of time for which a word, non-
word, or senile letter is presented, information can be
gained to determine which physical characteristics of
words are encoded by the reader and aver what time scale
this encoding takes place . One particular finding in this
area has maintained a considerable impact on contem-
porary views of word recognition since it was first re-
ported more than twenty years ago . Using a two-alternative
forced-choice procedure (generally known as the Reicher-
Wheeler paradigm), a number of studies have shown that
whet alphabetic stimuli are followed by a pattern mask
(composed of irregular arrangements of letter fragments
or similar contours), letters in words are reported more
accurately than are the same letters presented in isolation
(c . a ., Carr, Davidson, & !-Iawkins, 1978 ; Carr, Lehm-
kuhle, Kottas, Astor-Stetson, & Arnold, 1976 ; Ester,
1975a ; Hawkins, Reecho, Rogcrs, & Peterson, 1976,
Holezader, 1979 : Johnston, 1978, 1981 ; Johnston &
McClelIand, 1973, 1974. 1980 . Jordan & de Bruijn, 1993 ;
Massaro Rz Klitzke, 1979 ; Purcell & StanovECh, 1982 ;
Reichcr, 1969 ; RumeIhart & McCleliand . 1982 ; Taylor
& Chabcrt, 1978 ; Wheeler, 197Q) . Moreover, this advan-
tage for words appears to be due to the perceptual pro-
cessing of these items, and not to the use of some form
of postperceptual guessing strategy based on orthographic
knowledge . Far example, when investigating the word ad-
vantage aver isolated letters, Johnston (1978) argued that
This research was supported by Grant SPG 8931914 from the Joint
Council Initiative m Cognitive Science and was reported at the September
1992 meeting of the European Society for Cogm[i,, e Psycholn,7 m Pans .
We are grateful to Hannah Buchanan-Smith, Veronica Dark, Albrecht
Inhoff, GcoTfrev Loftus . and an anvmmous reviewer for help and in-
aghtful comments concerning this reSCarch Requetits for repents should
!~e sent to T R Jordan, Psychological Laboratory . University of'St An-
drews . St Andrews, Fife KY'16 97U . Scotland
if redundancy were being used at any stage in the pro-
cessing of word targets . performance with words would
be related to the amount by which the identity of the crit-
ical letter is constrained by the identities of the remain-
ing letters in each word . Johnston (1978) found no evi-
dence of such a relationship under display and testing
conditions similar to those used by Reicher (1969) . Thus,
when viewing tame is limited by the use of a poststimulus
pattern mask, letters an words appear to be more percept-
ible than letters presented in isolation .
Before this advantage for letters in words over isolated
letters [the "Word-Letter Phenomenon" (WLP) : John-
ston & McCIelland, 1973] can be fully explained, the pre-
cise role played by pattern masks in this phenomenon must
be revealed (e .g ., see Eriksen, 1980) . More specifically,
while the word advantage over single letters presented in
arrays of meaningless characters (e .g ., number signs,
#}t6 ; dollar signs, Th$$ ; ampersands, &h&&) appears
to be insensitive to the type of poststimuius field employed
(patterned or blank ; Johnston & McClelland, 1973 ; Mar-
chetti & Mewhort, 1986 ; Mewhort & Johns, 1988), the
word advantage over isolated letters (e .g ., h ) has so
far been observed only when stimuli are followed by pat-
tern masks and disappears when blank poststimulus fields
are used (e .g ., Johnston & McClelIand, 1973 ; Juola,
Leavitt . & Chce . 1974, '.vfarchem & :vlewhort, 1986,
:VIassaro & Klitzke, 1979 ; see also Taylor & Chabot .
!97$) . For example, Johnston & McClelland (1973) pre-
sented four-letter words (e_Q_, coin), letters in number
signs (c###) . and isolated letters (c ) in two backward-
masked conditions . In one condition, pattern masks (com-
posed of an irregular pattern of contours resembling letter
fragments) were used ; m the other condition, each mask
was simply a blank white field . Johnston & McClelland
(1973) found that letters in words were reported more ac-
curately [hale were letters in number signs under both blank-

1., Copyright 1994 Psychonomic Society. Inc.

X34 JORDAN AND BEVAN
field and pattern-masked conditions . However . letters in
cards mere reported more accurately than were isolated
letters only when pattern masks were utied : when stimuli
were followed b~, a blank field . performance with isolated
letter, was actuallti more accurate than ii was with vkords .
The contrasting cftccts of pattern masks and blank fields
on the relaU~e pt'_rcepuhilitN of cords and isolated letters
have inspired the copular notion that the WLP reflects
the replacement of ongoing target procesJn? by a subse-
quentl-v presented pattern mask and, therefore, that the
WLP requires the use of poetstimuLus (hereafter termed
backward) pattern masks (e .g ., Golden, 1986 : Johnston .
1981 . Johnston & 'vfcClelland . 1980 . '.v1cCIetland, 1985 .
1986 . 1991 ; McClelland & Rumelhart, 1981, i4&$ ; Pang,
\ewaomc . McDonald. & Sctavancvc[dt, 1982 ; Rumelhart
& McCletland . 1982) For example . consider the accounts
of the R'LP provided b~ the interactive-activation model
(TAM) of McClelland and Rumelhart (1981 : see also
'.~cClelIand & Rumclhart, 1988 . Rumeihart & McClel-
1and . 1982) and the hierarchical model (HM) of Johnston
(I9$ I ) . Each model postulates a hierarchical network of
detector units forming three levels of representation : one
for letter fragments, one for letters, and one for words .
Both models contain direct connections between nodes in
adjacent levels (i .e ., between letter-fragment nodes and
letter nodes and between letter nodes and word mass) ;
however, nodes for letter fragments are not connected
directly to noes for words . When an isolated letter is pre-
sented, fragments of the letter arc analyzed first, and this
information feeds forward into tie fetter bevel and acti-
vates the appropriate letter node . When a word is pre-
sented . similar processes take plus at the letter-fragment
and letter levels for each letter Ln the ward . but informa-
tion from activated letter nodes now feeds forward into
the word level and activates the appropriate word node .
Although the IAM and HM differ somewhat in tetras of
their functioning (most notably because the HM as a purely
feed-forward system, whereas the IAM permits feedback
from the word level to the letter level), bath models pro-
pose that backward pattern masks are necessary for the
WLP because the phenomenon itself reflects the selec-
tive replacement of ongoing processes of target percep-
tion by masking noise . More specifically, when a word
or isolated letter is presented, accurate tamer identifica-
tion cannot take place until activation in tie appropriate
representation for the target has built up This buildup
of activation takes time and if, in the meantime, a pattern
mask is presented to the system, fetter fragments in the
mask will add noise to the network and quickly replace
the pattern of activation currently existing at the letter
level, However . because fetter-fragment nodes arc not
connected directly to word nodes, the pattern of activa-
tion currently existing at the word level will be far less
susceptible to replacement by masking poise . This dis-
parity in the degree [o which ongoing processes of word
and letter perception are replaced by the subsequent wave
of acti,.ation created by a backward pattern mask is, ac-
cording to these accaantc, fundamental to the WLP . In-
deed . this interpretation of the influence of backward pat-
tern masks on target perception is also used to explain
the absence of the w'LP when targets are followed by a
blank field because blank fields do not contain the letter
fragments necessary to selectively replace the processing
of a preceding alphabetic stimulus (see Johnston, 1981 ;
Johnston & 'vicClelland . 1980 : 'McClelland & Rumelhart .
1981) .
Althou,h the role of mashing in the WLP proposed by
the IAVI and HM may be attractive, two concerns (one
emptricat, one theoretical) sesg~est that the importance as-
st,ned to backward pattern masks is premature .
First, the empirical concern . Despite the popular in-
tcrprctatton placed an the findings of earlier research into
the effects of pattern masks and blank fields on ward-letter
performance, this research actually provides only equivo-
cal support for the nation that backward pattern masks
are a necessary component of the WLP . Specifically,
while Johnston and McClelland (1473), Juola et al .
(1474) . -Marchetti and Mewhort (1986), Massaro and
Klitzke (1979), and Taylor and Chabot (1978) found that
the WLP disappeared when blank fields were used instead
of backward pattern masks, these findings were actually
obtained using displays in which a blank field both fol-
lowed and preceded each stimulus . Specifically, while
each study used blank poststimulus fields, 7uo3a et al .
(1974), Marchetti and Mewhort (1986), Massaro and
Klitake (1979) . and Taylor and Chabot (1978) preceded
each word and isolated letter with a plain dark field, and
Johnston and McClelland (1973) used a plain white field
for both poststirrtulus mask and prestimulus fixation field .
Therefore, rather than demonstrating the critical impor-
tance of backward pattern masks m the WLP, ftse studies
actually show that the WLP disappears when pattern
masks are absent from each trial entirely, not jest from
the poststimalus field- Consequently, rather than indicat-
ing that backward pattern masks are necessary for the
WLP, such studies demonstrate that the WI,P may require
merely that a pattern mask be presented on each trial, is-
respective of whether this is before or after each ward
or isolated letter is shown . ` Thus, a more informative way
to investigate the role of backward pattern masks in the
WLP is to contrast performance under backward-masked
conditions (where words and isolated letters are preceded
by a blank field and followed by a pattern mask) with per-
formance under forward-masked conditions (where words
and isolated letters are preceded by a pattern mask and
followed by a blank field) . In this way, not only would
backward pattern masks be replaced by a blank poststirnu-
lus field (the critical comparison for backward-masking
theorists) . but the overall content of forward pattern-
masked displays would also match that of backward pattern-
masked displays . Where the WLP has traditionally been
observed In so doing . problems of interpretation asso-
ciated with contrasting backward-masked performances
witty those obtained using "control" displays in which
stimuli arc presented without any pattern mask at all would
be overcome .
 WORD SUPERIORITY OVER ISOLATED LETTERS
We turn now to the second . theoretical, concern In-
terruption mask= is not the only prevailing theory that
can explain why pattern masks ditirupt the perception of
alphabetic tar,ets For example, it has been known for
some time that the sensory qualities of target and mask
can become integrated to form a composite gercept in-
corporatin; aspects of both stimulus fields (e . see Breit-
.g
meyer, 1984 ; Felsten & Wasserman, 19$0 : Ganz, 1975 ;
Jordan, 1990 . in press) . Thus, although backward pattern
masks have the potential for interrupting target process-
inQ, it is far from clear that interruption, rather than in-
tegration, plays a critical role in the WLP . Indeed, Jor-
dan and de Bruin {1993) have recently argued that a
possible role for backward pattern masks in the WLP is
that of selectively impairing the discriminabiIity of iso-
tated letter targets through target-mask integration . More
specifically, where words and isolated letters are presented
under backward pattern-masked conditions, both types of
target may become formed into composite percepts of tar-
Qet and mask . However, isolated letters (because of their
diminutive size) maa be less easily discriminated than
words from each pattern mask, which may inspire an ad-
vantaQe for word targets (see also Estes, 1975a, 1975b) .
This account can also explain the absence of the WLP
when blank fields are used . Quite simply, the word ad-
vantage over isolated letters disappears when blank fields
are used because blank fields do not contain the contours
necessary to selectively impair target discrimination .
Thus, when pattern masks are used, the structural simi-
larity between the contours of each target and mask may
disgwse the precise location and extent of each taract, but
the absence of these contours when blank fields are used
allows both words and isolated letters to be readily dis-
criminated in each display . Consequently, while blank
fields may impair performance in other ways (e .g . . by
lowering target contrast when a flank white flash is used,
as Johnston & McClelland . 1973, suggest), the dis-
criminabiiity of isolated letters may not be selectively im-
paired in these displays, and so the possibility of obtain-
ing a word advantage is removed .
The accuracy of Jordan and de Bruijn's (1993) inte-
Qration-discrimination hypothesis has yet to be fully de-
termined . At the very least, however, the hypothesis
proposes that the role of masking in the WLP is not inter-
ruptive and that the WLP is produced because words and
isolated letters differ with respect to the ease with which
they can be discriminated from the masIun, stimulus . If
this were the case, there seems little reason to suppose
that a WLP should not also be produced under forward-
masked conditions- a comprehensive examination of the
literature reveals that this critical test of the role of mask-
inu in the WLP has never been reported .
The picture being developed, therefore, is of a WLP
that may be intipired under both backward- and forward-
masked conditions . But, as we have already shown . this
N icw contrasts sharply with current accounts of the role
of masking in the WLP . Indeed, for the HM, a word ad-
%antage "depends on the use of a patterned mask follow-
ing target exposure" (Johnston, 1981, p . 76) while, for
135
the IAVi, pattern masks inspire a word advantaQc by
"drivma the system towards a new steady state and wip-
ing out the remaining traces of the previous stimulus"
(Rumelhan & McGelland . 1982, p . 61) . However, while
the accuracy of these accounts has not vet been challenged
by the report of a WLP under forward-masked conditions,
the role of mashing that they propose has already been
challenged by a recent study by Prinzmetal (1992) .
Prinzmetal was also concerned with the notion that back-
ward pattern masks are necessary for the W LP, and ex-
amined its accuracy by presenting a pattern mask simul-
taneously with each stimulus . Despite the fact that no
backward mask was presented on any trial, an advantage
for words over isolated letters was observed (Pnrumetal,
1992, Experiments 4 and 5) . However, Prinzrrietal was
concerned with cstabiish= a WLP without brief presen-
tations and so limited performance by using unusually
small stimuli (for example, four-letter words subtended
only 0 .426°) which were presented until subjects re-
sponded . This deviation from the conditions under which
the WLP has traditionally been reported makes it diffi-
cult to compare Pnnzmetal's findings with those of pre-
vious research in this area . Moreover, Prinzmetal did not
employ a backward-masked condition in his experiments
and so provided no benchmark with which to compare
the WLP obtained without backward masks . This raises
the possibility that a stronger word advantage might be
obtained when backward pattern masks axe used and,
therefore, that backward pattern masks still play a spe-
cial role in the WLP . Therefore, while the findings of
Prinzmetal offer encouragement for the view that back-
ward pattern masks are not necessary for the WLP, the
precise extent to which backward pattern masks inspire
the WLP traditionally observed under abbreviated view-
ing conditions remains unclear . From the arguments pre-
sented earlier, a comparison of the effects of forward and
backward pattern masks on the relative perceptibility of
briefly presented, "normal-sized" words and isolated let-
ters is needed to resolve the role of masking in the WLP .
In sum, therefore, the three experiments reported in this
article were motivated by the view that the WLP tradi-
tionally observed with backward pattern masks may also
be observed with forward pattern masks even though,
under forward-masked conditions, words and isolated let-
ters are followed by a completely blank field . Moreover,
by directly comparing the effects of forward and back-
ward pattern masks on the relative perceptibility of words
and isolated letters . information will be gained about the
precise involvement of pattern masks in the WLP; if sim-
ilar word advantages are observed with backward- and
forward-masked displays, the popular notion that back-
ward pattern masks play a special role in the WLP (e .g . .
as proposed by the IAM and HM) will require substan-
tial revision .
EXPERIMENT I
Experiment 1 way conducted as a straightforward test
of the hypothesis that the WLP can be produced even
1116 JORDAN AND BEVAN
tthen wordy and isolated letters are preceded by a pat-
tern mask and followed by a completely blank field
Method
Subjects Sixteen paid sublects participated m the experiment
All cubtects lead normal or corrected-to-normsi +, isfon and v.ere na-
t.%e speakers of English
Stimuli Ninety-six purti of four-letter words were selected ac
experimental stimuli, with a mean frequency of written occurrence
of 123 ner million (Carroll . Davies, & Richman. 1971) The mem-
hers of each pair differed by dust one "critical" letter
snow) with critical letters occurring equally often at each of the
four letter positions Target stimuli were constructed from a propor-
tionaEly spaced character set which provided natural variation in
the width of words and letters. This variation avoided presenting
the maniacal letters of target stimuli m bus[ four absolute screen
positions. since this may have provided abnormal cues to the post-
uons of letters m [eructs; m addition, the two critical letters of each
stimul s pair shared the same width and spacing to atioid response
strategies based on the width of each display . Ninety-tiix pairs of
isolated-letter stimuli were funned by deleting the three noncriti-
ca1 letters from each word pair, leaving each isolated letter m ex-
actly the same scrcen position as a appeared in the word. Two groups
of 96 stimuli were torrned from these word and isolated-letter stim-
uli. with each group containing one member of each word pair and
its matched isolated letter An additional 40 word pairs and 40 letter
pigs were constructed to provide 80 practice stimuli at the begin-
rung of each session
For each trial, a different pattern mask was constructed from
p5cudorandomly arranged fragments of the letters used in the charac-
ter se[. with the built-in constraint that no letters were formed by
these fragments A prel¢rfvnary detection task showed that these
masks greedy impaired perceptibility relative to a no-mask condi-
tion, and rendered words and isolated letters undetectable when stim-
ulus exposures were sufficiently brief. The character set and ex-
ample mask and word stimuli are shown m Figure Z.
Visual conditions . Masks, words. and isolated letter were pre-
sented m white on a dark gray oscilloscope screen . Words and let-
ters were presented m a proportionally spaced, lowercase font, based
on the Letraset "Pin Sill" typeface (see Figure Z) Background
illumination of the oscilloscope screen was approximately 1 cdlm',
and the luminance of stimuli and masks was approximately
25 cdlml.
A single letter x subtended hon7ontal and vertical visual angles
of approximately 0.20° . Four proportionally spaced letter xs sub-
tended a visual angle of approximately 1 .1Q° horizontally . Same-
Figure 1. The character set and example mask and word stimuli
used in Experiment 1.
sized masks were used `or cords and isolated letters Each mask
eubtended a %nua1 ,=lc of apprnrimatO, 0,500 ~er?i[ally and a
horizontal visual ansle mditiiduall~ matched to the width of cacti
%tord ?arLet
Design . Subiccts took part m No tiessions, one on each of 2 dif-
ferent days Half of the sub!iectti were show R 5timuluc Group I and
half were shown Stimulus Group Z ; for each su`?iect, the Name stim-
ulus group was Shf7AS1 7R both sessions . Each ,eevian was, dnided
into tree sections (practice . A, and B), u ith no obvious transition
from one section to the nest In each session, each word and yoked
isolate~f-letter stimulus were presented in a different section of the
show .
experiment (A or B), this ensured that the members of each word/
,cotated-letter pair were not presented in dose succession The a!-
location of stimuli to sections A and B was re-randomized for each
suhicct, with the constraint that each section contained an equal num-
bcr of each stimulus type (4$ words and 4$ isolated letters) . Stimuli
were shown m cycles of 16 trials, counterbalanced across stimulus
type and critical-letter position
Apparatus. The experiment was controlled by s Cambridge Elec-
tronic Design 1401 Intelligent interface slaved to a computer . Stimuli
were plotted en a Hewlett-Packard f?32A oscilloscope equipped
wrth rapid-decay Pd phosphor with a spot persistence time of 10
ycsec to 1Q%. The screen of the oscilloscope was completely coi-
crcd with matte black card except for an area at rte center rrzcasur-
mg approximateh S ° vertically and 2° honronta?ly 1n addition,
the oscilloscope had been modified to enable precise control over
the visual angle of stimuli and to provide a higher resolution dis-
play (Jordan & Martin . 1987) The experiment was conducted in
a darkened booth, and the subjects entered their responses via two
illuminated keys interfaced with the computer
Procedure At the beginning of their first session, each subject
was familiarized with all 26 Letters of the character set used m the
experiment At the start of each trial, a small fixation point appeared
a2 the center of the screen The subjects were instructed to look
at this point when initiating a display When they pressed a key,
the fixation paint disappeared and the following display sequence
was initiated : 500 msec blank: mask ; stimulus ; 500 msec blank.
Exposure durations for stimuli were determined individually for
each subject (see below) ; masks were presented for 50 msec longer
titan was each stGmulus as an aid to the effectiveness of mashing.
Four dashes were then shown, corresponding to the four letter po-
sitions m a four-letter word At one of these dashes, two letter
were shown, one above the dash and one below, and the subjects
had to decide which of these two letters had been shown at the po-
sition indicated by the dash . To make their choke, the subjects
pressed one of two keys to select either the upper or the lower al-
ternative. For isolated letters, the dashes were somewhat redun-
dant, but the same procedure was used far both types of stimulus
Throughout the practice and experimental sections, exposure du-
rations were reassessed for each subject aftereach cycle of 16 trials .
Exposure duration was increased (by 2 msec) if the number of cor-
rect responses m a cycle was below 11 (6R 75 `7 ) and was decreased
(by 2 msec) if the number of corrca responses in a cycle was above
13 (81 .25`7) Within each cycle, words and isolated letters were
shown for the same exposure duration ; when adjustments to expo-
sure duration were made at the end of a cycle, the same adjustment
was made for both types of target . This adjustment procedure car
surcd that overall performance fell in the midrange of the perfor-
mance scale and that each condition was represented at the same
exposure duration an equal number of tomes. Average exposure du-
ration for words and isolated letters was 48 msec
Results and Discussion
The raw data were submitted to an ana[vs~,, of variance
(A\OVA) with one between-sub lects factor (stimulus
(Iroup) and one within-subjects .factor (stimulus type) . The
findings were clear : mean correct performance was 73 27c,
 WORD SUPERIORITY OVER ISOLATED LETTERS 137

for words and 65 .1 % for isolated letters ; the 8 .1 % (SE _ stantially from 75 `1 correct. Reassessment occurred after every 16
11 .341) advantage for words was significant lF(1,14) _ trials for practice stimuli and after every 32 trims for experimental
i .74, MS, = 13 .24, p < .O1] and showed no tiiun of stimuli All remamino aspects of this experiment were the same
as for Experiment I
Interacting with stimulus orpup [F(1 .14) _ .039 . MSS _
6 .48, p > .S0] . Thus, the WLP does not require the use
Results
of backward pattern masks even when stimuli are briefly
Two types of data were collected in this experiment,
presented and of a normal size (cf. Prirvmetal, 1992) and
namely, accuracy of report and stimulus exposure dura-
may be observed even when targets are preceded by a pat-
tions . These two data sets were each subjected to an
tern mask and followed by a completely blank field (cf .
A10VA with one between-subjects factor (stimulus
the TAM and HM) .
croup) and two within-subjects factors [mask type (for-
ward, backward) and stimulus type , (word, isolated let-
EXPERIMENT 2
ter)1 . The individual assessment of performance for each
mask condition produced the desired effect ; similar levek
Experiment 1 established the existence of a WLP under
of overall accuracy were obtained with forward and back-
forward pattern-masked conditions, however, this find-
ward masks [69 .29 for forward-masked displays . 69 .6%
ing is novel and requires replication In addition, the ex-
for backward-masked displays ; F(1 .22) = 2_0b, MS, =
tent to which the WLP may differ under forward and back-
11 . 16, p > .10] . This matching was achieved by present-
ward pattern-masked conditions has yet to be determined .
ing stimuli in forward-masked displays for x mean of
Therefore, Experiment 2 was conducted to provide a
47 .25 rnsec, compared with a mean of 12 .48 msec for
direct comparison between the size of the WLP observed
stimuli in backward-masked displays [F(I,23) = 1190 .57 .
with forward and backward pattern-masked displays .
MSe -- 1 .35, P C .0001] .
Previous rcticarch suggests that pattern masks are gener-
Accuracy data for Experiment 2 are shown in Figure 2 .
ally more disruptive m forward-masked displays (e .g .,
Overall performance for words was 8 .8% more accurate
Kietzman, Bayle, & Lindsiey, 1971 ; Michaels & Turvey,
(SE = a-0 .66) than it was for isolated letters [F(1,22) _
1979 : Schiller, 1966 ; Schiller & Smith, 1965 : Smith &
9237, MS, = 3 .26, p C .0001] . No other main effects
Schilier, 1966 : Sperling, 1960, 1965 ; Turvcy, 1973) ; a
or interactions were found. The advantage for wards
preliminary investigation of the forward- and backward-
across forward- and backward-masked conditions was ex-
masked displays planned for Experiment 2 revealed a sim-
amined more closetv by a subsidiary A10VA for each
ilar difference between masking conditions . Specifically,
mask condition . A WLP was observed with each type of
when the same exposure durations were used for words
mask : for forward-masked displays, the word advantage
and isolated letters in forward- and backward-masked dis-
was 8 .3% [SE _ ±0 .38, F(1,23) = 19 .11, MS, = 4 .12,
plays, overall performance with forward masks was barely
p G .0(}11 ; for backward-masked displays, the word ad-
better than chance (55 %), while performance with back-
vantage was 9 .3% [SE _ ±0 .73, F'(d,23) = 61 .61,
ward masks came close to perfection (96%) . Thus . in
M& = 3 .9I, p C .00 11, Neither WLP showed any sign
order to obtain an appropriate comparison between the
of interacting with stimulus group (both Fs C !) .
effects of forward and back-ward pattern masks on the size
of the WLP, overall levels of performance for the
Discussion
forward- and backward-masked displays used in Experi-
ment 2 were matched by adjusting exposure durations sep- The findings of Experiment 2 add further support to the
view that the WLP does not require the presentation of a
arately for each mask condition .
backward pattern mask . Moreover, the word advantage ob-
served with backward masks decreased only slightly (by
Method
Subjects Twenty-four subjects from the same population as that 1 % ) when forward masks were used, suggesting that much
used in Experiment t participated in Experiment 2 .
Stimuli . The stimuli used in Experiment I were used in Experi-
EXPERISiEtT 2
ment 2 Pattern masks were constructed as before, but were now
used in two mask conditions Forward-masked display,., were iden-
tical to those of Experiment 1, comprising die sequence fixation FORWARD MASKS BACKWARD VASKS
point, 500 msec blank; mask ; sttmufus : 500 msec blank. Backward-
masked displays comprised file same components . but m the fol- C
O e°
lowing sequence fixation point: 500 msec blank . stimulus ; mask ;
500 msec blank. All masks were presented for 100 msec R
x m
Design As in Experiment 1, subjects took part m two sessions,
E
one on each of 2 different days . In Session 1, half the words and C
then matched isolated letter,, were shown m the forward-masked
condition and half in the backward-masked condition. In Session 2. WORD LETTER SiURV LETTER
masking conditions were reversed Stimuli were shown in cycles
of 16 trials, counterbalanced across mask npe (forward . backward)
.
sTINfUM", TYPE
stimulus ttipe, and critical-letter position Figure 2 . Mean percentage of critical letters correcth reported
Exposure duration-, we e reatiseued indn-iduaih for forward- and for wnrdsand isolated letters in the force ard- and backward-masked
backward-masked displavc and ad lusted if performance differed sub- displa-is of Experiment 2 . Error bars represent standard error .
138 JORDAN ,AND BEVAti
of the influence or backward pattern masks on tne WLP
remained under forward-rnaskecl ccmd~tions 7'h~~ ~imilar-
it% between forward- and backward-masked displays un-
derscores the problems associated with centertiporan ac-
counts of the VVLP in which special powers arc attributed
to backward pattern marks ; from the findings of Experi-
ment 2, the word advantage over isolated letters shows little
sensitrvih to the temporal ordering of stimulus and mask .
While a similar-siccd WLP was obserYCd with forward-
and backward-masked displays in Experiment '21 . overall
Jevef~ of accUracv were equated by presenting stimuli for
longer duration,, m forward-masked di%plavs, which suo-
oests that pattern masks were generally more disruptive
when presented before a stimulus was shown A5 dis-
cussed earlier . thi4 finding is consistent with previous
comparisons of the effects of forward and backward pat-
tern ma-,k,, (e .,- , Kietzman et at _, 1971 ; Michaeis cec Tur-
vey, 1914 : 5chilIcr, 1966 : SChiIler & smith, 1965 : Smith
& Schilicr, (466 ; Sperling . 1960, 1965 : Turves . 1473,
which indicates that the different degrees of masking ob-
served in Experiment 2 were not a quirk of that particu-
lar experiment . Moreover, as contemporary accounts of
the WLP assign no special importance to particular ex-
; m backward-masked displays, there is
posure duration-
no principled reason why the longer exposure duration5
necessitated bv the use of forward pattern masks should
detract from the finding that similar forms of the Wi.P
were obtained under backward- and forward-masked con-
ditions . Nevertheless, for the purposes of comparing the
two forms of the WLP, greater reliance could be placed
on their apparent similarity by knowing more precisely
how exposure duration relater to tie relative perceptibility
of words and isolated letters in each mask condition . For
example, longer exposure durations with forward pattern
masks suggest that stimulus perception (words and let-
ters) in forward-masked displays occurred over a longer
dine scale than it did in backward-masked displays . Con-
sequently . aid importantly for comparisons between the
two forms of WLP, these different time scales may reflect
different temporal relationships between word and letter
perception for each type of mask ; that is, the perception
of words and isolated letters under forward-masked con-
ditions may have been separated by a greater delay than
under backward-matiked conditions, even though similar
overt word advanta.Qes were observed with each mask con-
dition an Experiment 2 . On the other hand, a great deal
of evidence indicates that masking can occur at different
titages of perceptual pi-messing : peripherally (occurring
from the retina up to, and perhaps including . primary pro-
jeetiQn areas of the visual cortex) and centrally, where
the output of peripheral analyzers is integrated into rela-
tional or categorical features b%, corneal analysis (for re-
views . see Breitmever, 1984 : Breitmcycr & Ganz, 197( :
Kahneman . 1968 ; Kolers, 1968 : Turves, i973} . In addc-
tian, many researchers argue that fmAard masks create
much of their interference at a peripheral stage of per-
ception and that peripheral masking is greater )tiith for-
ward masks than it is with backward masks (c-o-, Grecn-
,.pon & Eriksen . 1968, Kictzman et al ., 1971, Michaelti
& Tur-,e-, . I979 : Moscowch, 1986, SchilSer, 1966 ;
Schiller & Smith . 1965 : Smith & SchilIer, 1966 : Sper-
ling . 1960 . 2965 : Turves . 1973) . Applying this logic to
the findings of Experiment 2, the longer exposure dura-
tions required for forward masks may reflect merely an
overall disadvantage for forward-masked stimuli at a pe-
ripheral stake, outside the word-fetter-rcco~mtion mod-
ule (e .g . outside the modules represented by the IAVI
and HM) . and word perception and letter perception mad
be separated b% a similar temporal delay under forward-
and backward-masked conditions . Thin, while forward
and backward pattern masks in Experiment 2 produced
a similar W[,P, is remained to be seen whether these overt
word advantages were produced by the same dynamics
of processing m each mask condition (see McClelland .
1991, for further discussion of the importance of estab-
lishing the temporary dynamics of processing) ."
One way to examine the dynamics of processing under-
lying floc WLP in forward- and backward-masked displays
is to establish the amount of increase m exposure duration
for isolated letters that is required to remove the word ad-
vantage m each mask condition .' More specifically . if the
word advantages observed with forward and backward
masks in Experiment 2 were each produced by the same
temporal difference between perceiving letters in words
and letters in isolation . the ward advantage should be re-
moved m each mask condition by a similar increase in ex-
posure duration for isolated letters . If, however, the longer
exposure durations used with forward masks concealed a
greater temporal difference between perceiving letters in
words gad letters 9n isolation, a greater increase in expo-
sure duration for isolated letters should be required to re-
move the word advantage from forward-masked displays .
If different increases :n exposure duration are necessary .
it will 5c difficult to argue that the same processing dy-
namics are responsible for both forward- and backward-
masked forms of the WLP .
EXk'ERIN'fENT 3
The procedure used in Experiment 3 was adapted from
Tusvey {1973 ; Experiment 10) to make a ptacedure that
was more applicable to the Keicher-Wheeler paradiCm .
Specifically . exposure durations were adjusted indepen-
dently for words and isolated letters in forward- and
backward-marked displays so that forced-choice perfor-
mance was aimed at 75% correct for each titimusuc con-
dition . In this way, the relative perceptibility of each stim-
ulus type (word, isolated letter) X mask type (forward,
backv.~ard) combination way revealed hv differences in ex-
posure duration, using testing conditions and a level of
performance matched to those empfaNLd when the WLP
has pre,, touslt been ob~crred .
 WORD SUPERIORITY OVER ISOLATED LETTERS
WORD LETTER WORD LETTER
FORWARD FORWARD BACKWARD BACKWARD
Method
Subjects . Sixteen paid subjects from the same population as that
used in Experiments 1 and 2 participated in the experiment .
Design . Throughout the practice and experimental sections, stim-
uli were shown in cycles of 16 items, representing the S stimulus
type x critical-letter position combinations for each mask condi-
tion . Within each cycle, stimuli were shown in a random order,
and exposure durations for words and isolated letters in each mask
condition were independently assessed after x118 items for that mask
condition had been shown. All remaining aspects of this experi-
ment were the same as for Experiment 2 .
Results
The results of Experiment 3 are shown in Figure 3. As
in Experiment 2, two types of data were collected : ac-
curacy of report and stimulus exposure durations . These
two data sets were each subjected to an ANOVA with one
between-subjects factor (stimulus group) and two within-
subjects factors [mask type (forward, backward) and stim-
ulus type (word, isolated letter)] . Turning first to the data
for accuracy of report, no main effects or interactions
were obtained (all ps > .15) . Indeed, virtually identical
levels of performance were obtained for words and iso-
lated letters in forward- and backward-masked displays :
for forward masks, 73 .1 % (words) and 72 .2 % (isolated
letters) ; for backward masks, 73 .2 % (words) and 72 .3
(isolated letters) .
Turning now to the data for exposure durations, a main
effect was found for mask type : stimuli were presented
in backward-masked displays for an average of
13 .37 msec (12 .21 msec for words, 14 .52 msec for iso-
lated letters), while the same stimuli were presented in
forward-masked displays for an average of 47 .60 msec
[46 .53 msec for words, 48 .66 msec for isolated letters ;
F(1,14) = 258 .12, MS, = 1 .65, p < .0001] . A main ef-
fect was also found for stimulus type : mean exposure du-
rations were 2 .22 msec longer (SE _ ±0 .25) for isolated
letters than they were for words [F(1,14) = 16 .47, MS, =
0 .52, p < .041] . However, this effect showed no sign
of interacting with mask type [F(1,14) = 0 .28, MS, =
139
1 .64, p = .60] ; indeed, the word advantage was removed
to the same extent in each mask condition (see accuracy
data above) by virtually identical increases in exposure
duration for isolated letters [2 .13 msec (SE _ ±0 .85) for
forward-masked displays and 2 .31 msec (SE = 10 .28)
for backward-masked displays] . Moreover, the negligi-
ble difference (0 .18 msec) between the two increases was
in the opposite direction to that predicted by the hypothe-
sis that the longer exposure durations observed with for-
ward pattern masks in Experiment 2 concealed a greater
delay between the perception of letters in words and iso-
lated letters relative to that produced by backward pattern
masks . Consequently, while the findings of Experiment 2
show that stimulus perception in general occurred over a
longer time scale with forward pattern masks, the findings
of Experiment 3 indicate that this difference does not ex-
tend to the temporal relationship between word and letter
perception in each mask condition because forward- and
backward-masked forms of the WLP were both removed
by essentially the same increase in exposure duration .
Discussion
In the light of these findings, it seems likely that the
overall increase in masking observed with forward-
masked displays in Experiments 2 and 3 was due to in-
creased interference at a peripheral level of analysis, far
removed from representations for words and letters. Con-
sequently, while obtaining an advantage for words over
isolated letters with forward pattern masks provides prob-
lems for contemporary accounts of the WLP, the longer
overall exposure durations observed with forward-masked
displays in Experiments 2 and 3 and the absence of a con-
comitant influence on the relative perceptibility of words
and isolated letters are both consistent with contemporary
theories concerning the different loci of masking effects .
We shall return to this point in the General Discussion
section .
At first sight, the shift in exposure duration for isolated
letters (a little over 2 msec) required to remove each form
of the WLP observed in Experiment 2 may appear to be
implausibly small . However, the finding that such a small
duration shift can remove a substantial word advantage
does have a precedence, at least for backward-masked dis-
plays . Taylor and Chabot (1978) examined the influence
of exposure duration on the relative perceptibility of words
and isolated letters in backward pattern-masked displays .
Stimuli were presented over a range of exposure dura-
tions, and their relative perceptibility was determined by
the interval between target offset and mask onset that was
necessary to achieve a particular performance criterion
(accurately reporting four consecutive targets) . When the
duration of isolated letters was increased by just 2 msec
(e .g ., from 8 msec to 10 msec), the perceptual advantage
observed for words when both words and isolated letters
were presented at the same exposure duration (for this
example, 8 cosec) was completely removed . It should be
said that some interobserver variation in performance was
apparent in Taylor and Chabot's experiment (only two
subjects were used) and that no data were obtained for
140 JORDAN AND BEVAN
forward-ma5kcd displays Nevertheless, in terms of the
effects of exposure duration on word superiorrt~ over iso-
lated letters, it appears that a shin of 2 mscc may indeed
be sufficient to remove an advanta4c for words .
Finally . despite selectively reducing the exposure du-
rations for wards in Experiment 3, an identical neoligi-
btc word advanta2c (0 .97) remained in each mask con-
dition . Thus, both forms of the wLP were equally
resistant to attempts at removal and, while forward and
backward pattern masks produced very similar effects on
the relative accuracy with which letters in words and iso-
lated letters were reported, the absolute shift in exposure
duration required to remove these word advantages com-
pletely as likely to be fractionally greater than the mean
of 2 .2 cosec obtained in Experiment _3 .
GENERAL DISCUSSION
The goal of the experiments reported in this article was
to examine the role of pattern masks in the WLP . More
specifically, contemporary interpretations of previous
findings in this area propose that the WLP observ ed under
abbreviated viewing conditions critically depends on the
utie of backward pattern mask,,_ This, in turn, has inspired
modes of word and letter perception (e .g . . the IAVI and
the HM) to account for the VvLP in terms of the ability
of pattern masks to disrupt ongoing processes of word
and letter analysis and to overlook the possibility that the
WLP may be produced under forward-masked conditions .
Contrary to these accounts, however, the experiments re-
ported in this article show that tie WLP can be produced
when stimuli are preceded by a pattern mask and followed
by a completely blank field .
One immediate implication of the word advantage ob-
served with forward pattern masks is that it cannot be ac-
commodated by existing explanations of the WLP ob-
served with backward pattern masks The IAM and HM
propose that the role of pattern masking in the WLP is
explained by the behavior of a system in which stimulus
processing gets under way before masking disruption takes
place . However, under forward-masked conditions, stim-
ulus presentation will impinge on a system already affected
by mask activations . Thus, while accurate stimulus rec-
ognition under backward-masked conditions can be char-
acterized as the resistance of ongoing stimulus process-
in- to subsequent masking effects, accurate stimulus
recognit on under forward-masked conditions can be char-
acterized as the overcoming o£ ongoing masking effects
by subsequent stimulus processing .
The implications of the differences between these two
processing scenarios become clearer when attempts are
made to simulate the w'LP observed with forward-masked
displays using the software implementation of the IA:vI
pro,, ided by 'vlcClelland and Rumelhart (1988) .4 This
software simulates the hierarchical IA'vi arransement of
units at each of three processing levels- feature, letter .
and word . At the feature bevel, there is a set of unity for
features in each of four letter positions . Within each set .
units respond to the prcscncc of each of the line se;cnents
composing the simple font adopted by Rumelhart and Si-
ple (1970 . At the letter level, there are four sets of utter
unity (each set contains units for each of the 26 letters
in the English alphabet), one set for each position in a
four-letter word . At the word level, there is a single set
of detectors for each of 1,179 four-letter words .
The software receive, simulated visual input from a key-
board or text file (i .e ., the simulation does not have eves)
and so describes only the central aspects of the percep-
tual processes involved . Notwithstanding this restriction,
the model can be used to simulate trials from psychologi-
cal experiments . For example, tieveral successive displays
within a single "experimental trial" may be specified,
with each display characterized by an onset time and an
arxav of "visual" information . In this way, trials can be
made up of displays of any sequence of words, isolated
letters, and pattern masks (the latter composed of indi-
vidual characters simulating an ovcr5truck X-0 combi-
nation), although the software normalizes each input into
a consistent font br redefining each letter and mask
character in terms of the presence or absence of 14 letter
features (after Rumelhart & Siple, 1974) .
The simulation operateti m discrete time slices (or ticks),
updating the activations of all of the units in the system
once in each cycle, and assumes that the information- used
to respond in backward-masked displays is that present
in the system dust before the onset of the poststirnulus mask
('_vlcCleIland & Rutrtelhart . 1988) . However, McCIclland
and Itumelhart (1988, p . 214) assert that when stimuli are
not followed by a pattern mask, the point in processing
at which a response is chosen occurs "after activation and
response strengths reach their asymptotic values ." The
model',, performance under forward-masked conditions
was examined using this assumption .
The basic parameters of the simulation, such as decay
rate, maximum activation . and minimum activation, were
not altered and words and isolated letters were processed
for as long as it took the activations produced by these
stimuli to reach asymptote . A forward mask was "pre-
sented" to the model for 16 cycles, and was followed by
a word or isolated letter until activations reached asymp-
tote . The outcome of this simulation revealed that, at
asymptote, words and isolated letters dad a similar cor-
rect response probability (_8Q for wards . _83 for isolated
letters ; see Figure 4) and so simulate the findings of
previous research from which the model was derived,
namely, that the WLP disappears when no backward pat-
tern mask is used . In this form, however, the simulation
clearly fails to accommodate the forward-masked WLP
reported in this article .
However, while ;VIcClelland and Rumelhart', (1988)
assumption that readout takes place at a,;vmptote when
no backyard pattern mask is presented may he appropri-
ate for displays in which nn pattern mask at all (post- or
prestimulus) is used, it may be unreasonable to make the
same assumption about readout under forward-masked
conditions for example, under forward-mucked condi-
 FORD SUPERIORITY OVER ISOLATED LETTERS '41

,,

45- `
a

V QC

a
t 02 - r.
o
e - - -
a
_02- ~=-==--

0 "a 20 30 40 50 5o
7me (cycles)

M ;urc 4 . The time course of aMivatinns produced by the 14X1 software (tilcC3e1-
land & Kumerhart, 1998) for words and isolated ietterc under fnraard-masked
conditioms . :Masks were replaced by targets after 16 c~clec . dote the advantaEe
for wards, Khich be;ins to appear at 22 cycles (i .e ., 6 crcle> after target once[)
and which peaks at 32 c6clec (i .e . . 15 cycles after tar2et onset) .

tions . performance mav'le limited by the eFflcicncti with
which the idcnti!v of a target can he extracted from a
rapidly fading composite ima,e of target and mask . In
other words . a severe temporal limitation may be impotied
on the analyse of target information in forward-masked
displays even Though target proccssina is not interrupted
by mask activation This, in turn, may mean that the prob-
abilitv of responding correctly ~~i!! he higher before
asymptote is reached (cf. the JAM) and, therefore, that
suh,ccts may adopt this strategy' in`ozward-masked cx-
peraments . The simulation of forward masking provided
by the [AM softw are doe,, not reflect this sequence o` ac-
tnation . However, an inspection of the Ic~els of actt,,a-
tion produccC b~, words and isolated letters in Figure A
reveals chat . prior to the acvmptovc 1e~cl~ produced by
the simulation, an ad,,antaee `or words does exist . Spe-
cifically, after tar ;ct inFor^iat!on has been processed '0!-
6 cycles, a cap begins to form ',ctwccn -he levels of act+-
~ation !'or words and isolated letters !hat peaks at 16 c%--
cles to form a substantial word advantaRe (a difference
m activation of .1-1) . Therefore, if we repIace the assump-
tion (made hv VIcCle1!and & Rumelhart, 1988) that w hen
no backward pattern mark !,4 presented . responses w0'
be read out when actuation reachc4 as~°~r?p,ote "s!th the
ss~ufr?ptior. that fer%tiard patter-masked .stimul ; hake on_-,
a finite time in NNYch !o he Prc+cessec' (ifl Our simulation .
'6 crc[c0 . tie IA_'vI _s able to s .^iu'ate the f<»«ard-
tnasked fore o' *he ~;'LP obta .ned w c ".h human suti1ect4 .
Indeed . while t~e absc'.u+.e !eye's c,`rcrfnrr;iarce ror «ores
3T'~ '.C?CEFti in the Sji?)U'w~'0`1 «EtC S1-&~~ ~'ICrCT1L from
'.~OSC Obf31reC' '7 r`-1C CCC'Cr1mC^lti reDC'f'(Cd here . 3 c  "I'a~-
,,!zed «'LP %tias 01+?amed «,t'~ '-ac ',ware- and for%;Ord-
maskeC~~n~s'3t~t~~tir ' . 'cIr~aC'.%sar~anC 1 1fo_
ferNtard masks . see F~2ure s' . which concurs ni?h t'le
'~~c?in~ n` Erre " ,r,enrc 2 arc' ? T'-u4, c'_c~*,te !`ie sre-
crai role that the IAA assigns to backward pattern rnatiks,
the model has some success m accounting for the forward-
masked form of the WLP reported in this article
It is north nottnc, at this paint that when a W-LP was
observed with human subjects under forward-masked con-
ditions . tarRct exposure durations were considerably
longer than they were under backward-masked condition,;
(Experiments 2 and ~} . Thuti, it may seem strange that
in the forward- and backward-rrsa,~ked simulations pro-
vided by the TAM, a similar-sized WLP was produced
in each masking condition after target stimuli had been
processed for a similar number of cycles (for forward
masks . .16 cN clev ; for backward masks, 14 cycles) . How-
ever . it remains to he ~ :en exactly how exposure dura-
tion relates to processing time . Far example, one expla-
nation of the Ion= exposure dnratians observed with
forward-masked displays in Experiments 2 and 3 is that
this difference reflects increased disruption at peripheral
1er-elc c1£ proccticlnv and that the d-% namics of procession
within the word -recognition module are similarly dis-
rupted by forward and backward mask,, (sec Discussion
.section °ollowinR Expenment 3) ; indeed . this explanation
is supported b% the findings of Experiment 3 . Thin, the
apparent temporal discrepancy hetu eon the IAVI's stmu-
lation arc' tie '-fuman data ti got data?
But rrec,sel% wh:, should similar rormti ol' the 'NLP ~e
produced %tiit'i forward- and backward-masked c'_isnla~ ~"
Fo'!o,~; in- ±~e ar;umeiis o` Tordan and do Sruijn (199) .
one peccibi'_in is that the identity of a target can be ex-
[ K'.ed `rum mi5lcac'.m- ^rask m`ormatio^ mere efficient:N
`or " ords t~19 It can `ca-co'zred letters and that this situ-
ation occurs cqua? :,, fog '?0''" backward and fo-narc_
Specifica . :,, . 'n %ie  of t'ie
d,r^mut :~-e size of isolated 'erers, integration bet~,.ccn
!ar2e! a^c'. ^nark ti+c~u[c'_ p?ace iso!atec' ,eners in a diseropor-
;4Z TQRDAX ADD SEVAN
COPWARD MASKS
p =- wo .Ce " - ieo'e "ee ~e "" e~n
pd
7P
02
~s 2e 2s
',me rcycles)
Figure 5 . A comparison )f the levels of actuation produced b" the IAti4 coftHare +McClelland & Rvme2hart .
TM) under for-ard " and backward-masked eonditionc for cords and isolated letters . 1n each mark condition .
activation levels are shown for '_5 clcles of mask presentation and !6 t%Cles of tamer presentation . For forward
masks. the word adi, antage peaked 1 .14) at ?2 cycles fi .e ., 16 cticleti after target onset) . f,±r back- ard masks, the
word advantage peaked LI31 at 14 c%cles 6 .e . . 14 ciclec after target onset) .
?ionate'.% 'arse tiackgrounC of ^?ask contours_ Conse-
auent?v .'~e two forms of the WLP ~~+eer~cc' vi(h hac ;-
ttiard and forward pattern masks mav refTec, dr`ference,,
m the dncnmInahil* (if words aid isolated letters from
,he background provided ~% their respective masks . such
that a word ad,,antagc :,, inspired because differences 7n
relative mask size mean [hat the vocation any extent of
each l;ard are mare easi2-, discriminated than arc the '.o-
cation and extent of each isolated letter in a comnetsrte
percept of target and mask .
Support for the v iew t!~at the WLP may be expired bN
the ease with i;xhich targets arc discr,.mmated `mm pat-
tern masks comes from Pnrizmeta!'4 (1992) study in which
a word advantas!e ever isolated 'otters %vati found .~ hen
+aroet,~ were nhvsical :y embedded in a pattern-m`~ ;k back-
RrounC . Given that the present article is the first to re-
Port the forward-ma :.kec'`orm of the WLP, the role of
,ar2et-mask d,scrimrnahility in this form of *.he nhenoni-
enon retrains to be examined : however . lordan and de
3ruijn ('993) prov!de evidence o .` [,`le rnPuence of tarL~et-
mask discrim,natii'it% nn the 'nac!:war8-masked form of
'.he WLP As Jordan and de Srunr pent out . a num'Ner
of studies have ,tiol,%n that tar4ets are difficult ±o c'iscrim-
inate when 'hebackaround in which the target is embed-
ded contains features that are also Lresent~in the tame'
item {e L . Beck, 1966, 1967 : Beck & An-!h'er t972,
'971 : Tre!sman & GorriicLR . 19RR?, and fiat the time
taken to dscr :mina!e a tamer from ~t~, background !n-
creases with fhe size of t~e hack-round array . pantcular'_v
NN hcn rar ;e[s contain no one feature that di5tm~,ui%'iec them
from t~ei- ',ack,rourid fc , . Duncan, 1479 . 1987 . Dun-
can & Hurnphrers, 19Q9 : Hump~rc%> . Qu!n_an . (.S, Rid-
doc'i . 1989 : Tre7sman & Ge'aCe . '94Q1 . T'-iu4_ 7c-c'ar and
cue Brurr Teasoncd that ~'' tic Ni'LP ref ects tie ciryeratio^
CAF S!TY71'?-l C'`C~Ini~'13'.ll~r ^rCCt55e~ '.111ol~~ 111 ?'.!'~~C'\ d!?C
masks . r'-ie NVLp -rse" tic sensrice to t'~e cne ff
r'iC ^!".?s~~inc C~Ir~'1`'aS O\ l'r ? series o~ CKpe-rnen`s . 1or-
BACKWARD VASKS
= worea ~- ieo~e "ec ! e~ie "s
ng-
0 a ~ f I ~` I
sp 0 5 '0 15 29 25 30
T-e ~cycles)
day and de Bnztm examined the effects of varying the size
of each pattern mask re!ati%e to that ofthe target being
.racked They found that the WLP way produced . re-
moved . and even reversed simpzti by altering the width
of each mask relatne to the vidth of the target it ~,Nac
matikin cNcn though, at all rimes, masks covered tar ;cts .
For examnie, when pattern masks matched the width of
a girds and exceeded the width of isolated letters . isolated
letters were presented in a disproportionately larce mask-
in- field and the V6'LP was observed . Howe,, er, when the
width of each mask matched the width of each target (word
or stiotatcd letter) or when mask,, exceeded the width of
words and isolated setters by the same proportional
amount, the WLP was reversed (i e, . performance was
1)eEter for isolated letters than it was for words) .
It may, be premature to dismiss [5e idea that mask inter-
ruption contributes to t~e WLP obserieC with backward-
masked di~,p'ay5 \cicrthcicss, one parsimonious incrr-
pretation of similar effects produced )% Forward and
backward masks on the re2atne perceptihi2ity of words
and isolated letters reported in this article is that irzter-
rup?zen )lays no parc ,n either foal of the ~'4'LP Indeed .
the similar effects ohser,ed ui+.h°orward and backward
ma,,k,, ha~.c %mdctiprcac implications "or contcmporar%
views on maskin ; . In Etarncu!ar, nlam researchers argue
?`:a? `ors+are masks exert most of their ir.'7vence at pe-
nnhcra? tirasc~ of precc~~,rt~, while orl,, backward masks
exert effects on central (cortical) stakes o` target percep-
tic,n ~c , ., Grccntip(in &T Eriksen, '968, Kietzman et al, ,
197!, VTictiae'.s & TttrNc% . 1979 : VIv4covirch . 19Q6 :
Schi .'er_ 1966 : S&i!ler & Smith, 1965, Smith & Sctiiller,
'966 . 5TLr'in,' . 1960 . :9b5= TuY~ c% . 197_ : ,cc Discussion
sec'.ion, Experiment 2) For efa-rrilc . %t'ic^ c'.~,,cuccm ; the
c"ccts c " ,-=t,rnard arc: ~uc'{w arc'_ nattcrR nias ;> . TJrN cl
:97',, n . concudcc tl :at ._Fo,narC anC '+acknard
7laski^c ce^ `,()?h c+ccp, ~cr~n~e~a~'~ , but on'% hackn and
ma"kir- oc~_urs ?o an% Ce~,-ee CC-rra"% . . \toreoie, . , "
 WORD SUPERIORITY OVER ISOLATED LETTERS
is avoarcnt from the accounts of the TAM and HM, n~'icre
the .V4'LP ~,, produced hveelect:~,e disruption of represen-
tauons for letters and words . that 'both models assume that
backward pattern masks exert their critical influence on
the relative verccptibiiitv of wordti and isolated letters at
a central (cortical) cta,e of processing tpresumabl,, for
roost indniduals . in ?he left ~ettusphere ; e - . . Geschwmd,
1991, although see Bishop, 19881 . Nevertheless . the ef-
fects orserFed for forward and backward pattern mask,,
in the experiments reported in this article sug ;est either
that the critical role of pattern masks in the WLP occurs
at a peripheral stage (where both forward and backward
pattern masks mar be exerting an effect) tar that forward
pattern mask,, have more of an influence on central stages
of tamer perception than previously thought . Further com-
parisons of the effects of backward and forward pattern
masks on the refati),e perceptibilitN of words and letters .
u,;m, monopuc and dichoptic presentations o`tarRcts and
masks, are required to help resolve this issue y
In conclusion . while contemporary accounts of the WLP
propo,;e a critical role far backward pattern masks, a sim-
ilar V4'LP is produced when targets are preceded by a pat-
tern mask and followed b\ a blank field . Moreover . the
s,imilar effects produced by forward and backward pat-
tern masks on the relative perceptihilitN of isolated let-
ter, and fetters in word-, suavest that the role of pattern
masks in both forms of the WLP may be best explained
by considering aspects of masking common to both types
of mask ; target-mask diticrirnination provides one such
explanation . ~
REFERENCES
PecK, 1 (1456f Effect of orien[aum and of kheee stRiIlarity na per-
CepYUa' grouping Perccpirnrr & Pstithoph %srcs, 1, ?DQ- :02
EFCx, J ~19Fi1 Perceptua_ ~_Trooptng Producec:'~.N line figures Per-
ceprrwr R Pnchoph.~srcs, 2 . 49l-~95
BFC K, t s, ANtA!eR, 8 t19721 Discnr-natielu~ ofCif'crenceti in '~nc
slope anc! .ii line arrangement ati s function p' mask ce'av Pvrcep-
tron d Psi c7iophlscs,12, 1 3-7,S
? & 4titat!-R, 8 ('97. ~_Sect, The e':ects e~ concentrated one' dza-
tnhutec', attention nn penrheral ac?j :t% Perraprron & PsNchnphlcrc+ .
14 . 225-2 :0
BrsHt>-, D t' ti7 r !99O Can tie -_ht hcmisnhere medtatc language
as well as !'le loft' A c-itical rc`teH of revert research Cogninir
Y~uropsycltolag~, g, ?g;_ ;b7.
RRFtt10FVFx . B G . (1944i v,sual masking An mrrqrarrtie approach
Otiford 0-.`b-,2 L'mve!'si`), P,-re, . C!arendor Pros,;
AerlTuEiER .3 G,xGNNz .t,(L9",6) Tmplicauonco`sasta-iec`.and
rransicnl c`+anne_t fo- i:heores nF ti ~5u._' pat'ern ,+ask!ng >acc3c'~c eup-
rre5slon . acid information *,-(.xetsmg Pnrlinfo,ewal Re+,e~t . 33 . '.-36
CaRR,T N 1) N% iDso` . B ? .& H%itxIN s, H i rig'$i Percertual
'lev?,i' ;t% ,~ ~,.orc[ recocrniuo~~ Sr-a`egieq wrest orthogr2nti-c cori-
o
T?utallcln -it rot ;coca' acce<c ]nurna! ~Ezpeartenrn! Ps,tl:ielngi
Hwrmr Perc,~prmn R Prr(r "rrtQnce, -1 . 5"4-690
CaRR, T f! . !eITuK, PT F, 5 'N . KOTTAS . 8 . ~srou-S'~rso~ . E C .
x AR,,nr_~ . D ~ :976} T_, " ~.et rncire,~are ora.trce n the ~e'e-ir,rcn_
" innn"et".,-n%an~rgennteNts A %torc. sui+ei-jortt, e"ect Pe,cep-
~rw~ & Psrcharh~,x : I9,
C~14ROC. T R . T? .1%1-c P R ~fC4~1A1 . ~ ~~9"~1 1~1f ~'"'(n(art
Her,~n4~' +~r "d rcyuc~wt nn~ ; !3o~o-i Hova'~~nn-1~,^,in
D! \c%\ 1 ~'g7U~ Dig ~c'.eds'.tcrnor The %~~hd1: 11 "no-c .~a~ ttie sun,
~`~^cna~~ Jnu~nUlnfErPrrir~~r~ " ~~rP.ticiml~ ;~ Hu»iar'Pc~rrpluvn
f>: ~c ~~ ! ~'9~-} Atten,- .t and -eod~ria \6"ho'e, .!-i~~ ea*,c .r c5aze
143
rccn2-i,'.iOin ',i VT Cc>>t5eart IEC!_1_ Arrcr-n- and pr-fnrrrmnre X11
',I I "(" e . L- K E~'.ha~~r
Dr Nca, .? . ~ H[+-Rrrs, G W ('.9R9) ~-scz' sea"c~ and shriu-
' ::s s, ",~~'d.ia Ps~rf+or~ ;rCa! RP~re~t . 96 . G ; ;_d{R
E:;L,KSi:x C « (iq$n) Tie use e .` a % Tcua! ma« r,~% certeuti'ti con-
°oun~~ tiour experiment Percennnn cE PsYl"nplncrcs . 28, R9-92
EST~S .1~' K ~!~'Sa~ T`te'c~c~sefi~`e~c~ha'a :~e're" ceDneaI~rcKecsec
~n'et!ert~e^e .ica'ion .Iour.;nlrfFeperrtentnlP~~ci+olo~l General,
104 . !~~-'1s
EsTFN, W K anZ dec,s :ur !n'ettcr idcn-
~~tic2'.1on In R Snlcn MEd ) . Infnrmatron prnr"eCCr- and cognition.
7I~_- Lntinla Cimpv, rum 'Dr, :-301. Po,omac MD E~'haurr.
Fe,.~7r% . G . A W,kreFx-vnv . G S (1980) Visua' maslon_ VfecSa-
ni~ms anew t~eeres Pslcholnorcai Bu'letr~. 89. 329-~5~~
GaN/, L ([9-5) TcmpC)-,j! `ac!orc 17 %?,u21 ne"ceot+o-1 In E C Car-
?erettc k V Y F-iedman eEds i Handhnn~ o(rerrPptrorr (bbl 5
r? 169-231 } Ncu York Academic Press
GFa HuiND, N~ !I991 ~ SFecial2ationc o~ Lhe~uman `-1*aui In w' $ -X
%Vane !Fd i, 777e emer¢rnce q(Ianguaqe : Detielvpment and eiolu-
nnrr rpp. 72-471 Neu York. w H, Freeman
Goz oEN , R M (1986) A c?cvelermental neural moc'el of ~~,SUa', « ord
perc=,LOr_ Cognrnie Sc+rnce 10, <4]_1"F
GReFNsroN . T S . & FxiKCeN, C w~ ~ :95Rj .'nterncular non-
independence Yercepnon & Pctichophtisrc;s, 3, 93-96.
HA RK2N S, H L , RE7CTILA . G. VI , ROGERS, U ., Pc PETERSON, [.
t976~~ Flexible coding m word recognition, Jvurnal of Fxpersmerr-
1tr1 Ps%chnln,'I Human FCrrrpnor7 & Pc" r;FOrmancc . 2 . .$0-=85
Hoi rNo=e. D (1079) . Identification of fetters in %+o" ds and of smg'.e
letters with pre- end eiostknea'edge %s nostknouiedge of the al.er-
na?1l.e3 Percepnnn & PsNc'ItuphAcrrc, Z4z . 3j3-3 .R
Ht strttQei 5, G W , QUNT,a,, P T . & Rmnoc -, M l . r1989) Group-
»g qroces5eti in xisu,1l search Effects um?h tingle-any' combined-
zature targctc Journal ufE.cprrlmenral Ruhn,'ogl- General. 118 .
258-279
Jn!t"sroN J C (197 8~ A test o` the sovhisncatzd oucs ;mg theory of
cord percenuon CognItitie PsIcleoloQ . 10 :2?-'S3
1ottNsroN, 1 C ,2981? . L'nderstarzdm<_ wore! cerccphon~ Clues from
study trig the ward suT+cr,orrty effect In O Tzeaj : & H . Sintrer tFds .)
Perception of pnnr Readtng rercarof1 rn Fx~enmc~nraf psicholng~
rpp 65-R¢) Hi! ;sdale . \!, Erlhaum
IOli+sruN T C & McCLsLL,»:D . ] L (i97?ti V±sual factors m anrd
oercepdon Perception & Pcvhoplnsrcs, 14, X65-3,7Q
kO"rNsroI; . ! C , Rc VICCLLLL4Ai3, I L, (3974) Perception o" letters
m wards . Seek not and }e shall end Science, 7124 1192-119d
!nt±NsroN . J C . ~ '.bScCLLL4ND, J ! (lqRO) Experimental tests
of a tiserarchica' model of word recogmt5nn Jou~nul of L'erbalLearr-
me &- Verbal Beha~inr, 79, SQ3-524
?oRa%ti . T R (149fl~ Presentjna words without mtenor letters- Su-
penorin mer single ictter5 and influence of qottmatk boundaries
JattrrrUl of F_aper!mental Psic7,nlogi Humar Pcr~eptrnn & Yerfor-
mQnre, 1b R9?-9p9
JORDAN . T R . (in Press)- Contrasting e`:ects of fetter fragment masks
and non-Letter fraomcnt fields on alphalictic targets Examining the
-ole of mask con5~,uration .lnrcmaf of Gzpenmer nl Ps«Anln~,~ ~ Nu-
r+wn Perceptror R Performance
?ow)AN . T R . x ne Sxt i  . 0 (199; Wo,d ~cperora, over isa
'ate 'et'crc~ T`te neglec?ec' roc Oil' fankirg vac
; rorteurs Jounnai
of Erperrmc"rrral Ps,rhalqol Numnr Prrrrprnr, cC Perrormarrce .
19 . 5~9-56;
?oeOaN, T ft . .c MARTIN . C n_ (191,C) Tie orporancc of ~isua!
arg!e nr reeogmtinn A "chr :nkmC screen" mcxiificatmn for
s ~cva.l d1cn'at c jjc`fac~irrrr Re~~rar0h t"e,hcus Jn,e~~~~roenrr, X Cnm-
purerc 19 307-7,14
rr o_A ! F T .-A\ TIT . !) D . A Cwo,, (7 ti 1"9"=) Lette- idcnrfi-
tat,oa ii  oTd noTmnrd, e-d C single-'e'.'.e- c'icr~a~s 8rrllc~rrr vrrhr
PSlChI111nrrnc ~~QC1Prl . 4. -- C_,4O
!<~~t~ES~a~ D (1UW Slet~od .`nd~nqc nnc'c5eors ~s+.edieso' ~,anat
Rxl'r,r: . 74 1rL-G~.6
h~~z~" +~ t1 L Sm r !- .R C .cL~NnsFf U 8 r19"I) Fcrccp-
~!i,t' -nafk,n, Pertphera' cc contra' :acts- . Percepr=nr+ & PsNc)ro-
9rzs-
Kn-Rs. P A ('96g) Sonic r~ ;chol~ " ~,x~' a~rec~s ~,~ rafters reeo_~n~-
14-4 JORDAN AND BFVAN
"nr .-I u 4 Ko'e ", c(~ \1 Fd2^ '5~4 I Rr", nyrrn,e 7111'e-fl Sc~~" n^
\",~Pr~t~
~", sec ~r7T~ I' \? . & ~?t~tUr,v , f7 i !C r i94r,i Qn 'tie
c "~pen~~-"~ c-eC : Pcultoroo,ruf Recea,t'1; 4~Q _~5
ti"N5sAao D W . & I<, -,7,,F . ll ~ ly'y, 71,e -u'e ol macl~,iu
and ~rth~o_rn~~it ~t-uc'ure - 'et'e, 3n~ A(,rd ~ecn~n,r~~,~ lfcta P~i-
cHnlneuu . 4? 4 --12r,
11~ C, , r >\,) I L i ;9k5i pu*-g ',nvuled ;='+r r, d~xr 4 ~Lticmc
'n~ n~nr_~~^~^~~s_ para!'e! er .uc',~sres e~ .he n~ t',~Krrrr~e Scrence . 9 .
113
~t~-Cr F~ ~n~~ ] L r 1901 Ttie ore__-ammah' e 1,1 ick,oa,C -ode I("
In ? ' . VcC1e'land& !) F ftUr~e!'~an weds ~, Yaralir!
,iiburrc p~ou~3snrF Erpforarion, in the micro crrrcn~re n"co ;nrrron
rVo' - -- i--"1F9~ Carr'_+,,dcc . MA ti4'.T Pret~
M( (7, I L ('991) Stoc~asuc mteracr,UC nroceccec jnd ttie
cl"cc, o` contex'_ on Percentiog Cokrrarse PcirhoInQ1, Z?, '-4-:
ti4cCU_tnNo. J I . h RU+ttL[tnRC . D E OW) AR in!er2cti~c ac-
+n ahnn r-odel n'concert effcets in fet'.er percent~k- Pay' 1 A~ ~~.-
coant of tizac'inc'm ;c Pc%chofor«al Rrvicu . SR 3-5_1p-
McC' FT-La,D, J L . & RuaFrxaQ-r, U_ E ri9M Frploranorrs in
parnllcl d,rrrGrrrrd procey>rnr Camtimdoc, MA MIT Yreca
L?F%s-nar fl ! K . N low,s . E E '19W Some rats a' :~e
mteract«e-actitidVonmo+e'forworc'Identification Pnr)rnlnR;culRe-
searclr, 50. 13{-147
M ~cicaaLv . C F . & TI- N . T ff479T Central source- of visual
matkmn Indexing ~ .ruCtvres evpni±rtin~ wine at a an,le, "nef gl3ncc
P~,cholo2+ca1 Research . 41 '-~I ~ _
%t0GCn%'TC+i M (1986, Hemetnhenc specia'ttztio~, ipterhemi%phenc
ccxie~ . and transm,w(,q times irfcrcncec f,,-,m N etiual mask:n= ctu~~eti
,n nn,ma' ~~eoVe In F Lcnp,e . M Ptitn, & H H Tj~oer ~Ed; ~~ .
T,tn hcnttcPhcrrr nrrc hratn Frmcnunt ref dre rnrruc rullukunr
(OP .a.Q'-S](I) ~C\\ t'OfiL LESS
PnAP . K R \f«tiavE, 5 L , ti1cDo"M n, I E . n S(,3\ nE~ri.v ".
g ~ti r19821 An ,~ch~duon-~cnftcat~on ;T'ode . for lc'tcr and Mod
rcco_nu!or, 1'he wart'-cuoerio,i« effect Rc,rr,~ . 89 .
5^3_sy4
Yxe%niFr+i, ti4 (1992? The word-supenorttv effect dneE not requi~~e
a T-scc .u Perception cg Psiclrophirrrs, Vii, d7?-4$4
PtecrI t n G & STnNovxN,K F (19K) Some Soundany ccnc'i
tens .`or a unrd ~upenontv effect Quarterh Journal of EtpenmFrr-
ral Pstirlinlna% 34;A_ [!7-[~a
RFrcHF2 G M r 1969)_ Percevtual reco,Mition at a Function b' rnear[in~7-
`~~Inecs ~f-sf!~n~s'us matenal Journal ojErpc"nmenrQl Pc) rholnR~ 31
^_i$_yR0
Ri_ METH 1Q? . J ti , K VfcCzEi[.A,D, J L l19KZ} An ~n .e~acu~e sc-
~natit~n mode' of content ePfeci~ ".n letter nercep~ior Part'- Tic con-
textual enhance-nent effect anC scare tes?c and evenstons of the rria :el .
Pc%chnlrrQrcel Retiex, &4, 60-4=
Rttire!1tNxr U E . .c 5'rT-F P 1_9?=1 'Iyc rroccs, of reco ;mam_-
tachianECOpicauh :~rc~en,eZv.ords P,tcTvolr~;i<alReireA- 81 `3V'4
Sc . . . . . P i? i'965, Fc±r~s~~~ aid ~+ac'tuirc' n, ;±c'ti~~~_ a~ a function
o' " e~a " ~e otic~'ar, and ~~~enti "" ~ nF tzcr ane' ~^e~',r~~ tir~mu'~ P~rcen-
7rrnr ~ Pcirhop{nrrcc, 1, :6'- ;5=
Sc-ii- ee . P H . ~ Cvi~v . ~S C r1g65) h Ur-oanson nr'oruar~?
tmd tiackuard riackna Psjclrorrom!c ScrPnrC, 3, '7-i4
Sv-~, b1 C . R Sc,`L__a . P P 1966, Fonaard any F,ickudrc' matil-
n^ A corrparicnn C~rr~rdrun louTrra' qrP~irirnlngti . 2U, 337-~u~
SvE x' ~c,, G 1196n, The ~r~nr-ndt~on a~aila`~'e in'+~~ef t ~cua' rrrccn-
,ahors Ps ~cholnhicui `dnr,n,rapl±s, 74 (Wide No A9R}
5oFk--t,, G- '19F5i TcmFora' and spa'.~a' %i5ua' maskirn ! \laek-
n~'~} uo~u'se °ati`~ P"urnal nfrhe UpirrQ! Socrrn ofrlmerrcu GS
Sd!-5$9
TNN,-ort G A RC-!nnnT,k 1 (1979) DiFferen![a,~ack~Aar~mr.0;-
~, e` u~~~'e a~~ 'cttcTs `~~ rucks c` ~ a-c c~~ onhn~ra~ti~c ctru~r~re
,demon & Cngrrr+rn% 6, 629-6"s
Ta-isu,~,, N , & G[=aD[ . G x'9301 A feature integra'ioR the(,,rt of
attention Cognrnia Fnrhnlogl, ;2 . 97-1?5
TREIS'fA,,, A . & GoRvicNN, S 11989) Fe ;iture aRZlyszs in eer'v ~i~i-
sica Evidence from cea-eh asymmetries Prtcltnl~nrealRPtre,, 95,
ig-aK
TL kyE7 . y1 T 1 19-31 On nerrohera~ and ce~t " a' Lroce,see in Nision
P~\clri)log+cul Reties, 80, I-52
WhrFi_r-R,p D , L9i0} Prowestiesmwordd '-eco?nivon CakrrarteP.,i-
cF,olvx~, L 59-85
`DOTES
It is ;nie-es,m!_ to note " ha' ~n the seminal cn "nnarison o` *'ie e`-
fects of v,Iitrrn masks and hfank feldc on 'he WLP reDarted bv John-
stcm and McC'1eIland r 1973 1 . CtL h p3ttem-masked'arRe? wss followed
and preceded b) , pa!te'rr "na,4 Tn hinds3eht . ~t n cunuu~ that the 5nd-
in,5 nl t'~~~ stud v led to co much :rnoonancc hcino attached to the use
of bnchsiurd pd!tern ma~kc in the WLF
I! ~~rnuld'~c cmnhasized !hat we are not dicnvnng tFat the olcrult
Time course o( target prcxe ;qng Ts ionger,n forward-citasked ditiplays,
but ;ire c'ebatinL, whether the tame course , c°puruting the perception of
worc'c and 1Sn'ateC iMer1 i"hx`! . pre~umab<<, inspires the WLP~ in
cacti riaA condi'.iun !s 'oneer to forward-maskeC d3spla%s
? kcachnn-timc stuc'iec, in which the dependent va,tahlcNsas Lhe time
taken to identify the criuca' letterc m each condition . were considered
f3oweNer . 2 n0M stud-, »d,cated that the tcmine of overt rc~pnnces }sac
not a suficientl~, sensitive technique !o rcvcai Tempera' differenceti be-
twecn tyre eercegtual nrecessmo of anrds and isolated '.etters under. ah-
?+rektatet ~ieuiaR cnndrhnns Coneequent[% the reaction-time paradigm
uas rc~ccrcc'
The HVI was not a%a+'ab'e .o u, m suftNtiare form and so cou'd
not be tested in this wJN
Nanuscr,ot received .ianuary 7-C, 194,
~e~«io~ accepted °n~ 7uh1icahnn Tune ?0 I49? 1