Journal of Experimental Psychology:
Learning, Memory, and Cognition
2002, Vol. 28, No. 1, 34 – 45
Copyright 2002 by the American Psychological Association, Inc.
0278-7393/02/$5.00 DOI: 10.1037//0278-7393.28.1.34
In Search of Perceptual Influences of Sentence Context
on Word Recognition
Timothy R. Jordan and Sharon M. Thomas
University of Nottingham
A major issue in the study of word perception concerns the nature (perceptual or nonperceptual) of
sentence context effects. The authors compared effects of legal, word replacement, nonword replacement,
and transposed contexts on target word performance using the Reicher–Wheeler task to suppress
nonperceptual influences of contextual and lexical constraint. Experiment 1 showed superior target word
performance for legal (e.g., “it began to flap/flop”) over all other contexts and for transposed over word
replacement and nonword replacement contexts. Experiment 2 replicated these findings with higher
constraint contexts (e.g., “the cellar is dark/dank”) and Experiment 3 showed that strong constraint
contexts improved performance for congruent (e.g., “born to be wild ”) but not incongruent (e.g., mild )
target words. These findings support the view that the very perception of words can be enhanced when
words are presented in legal sentence contexts.
In contrast, from a modular perspective, sentence contexts do
not affect processes involved in word perception (e.g., Fodor,
1983, 1998; Forster, 1979, 1981, 1985, 1994; Seidenberg, 1985).
According to this view, sentence effects arise from knowledgebased representations that affect nonperceptual decisions regarding
the identity of words and which, therefore, affect speed of reading
and comprehension in ways that do not reflect processes underlying actual word perception (Fodor, 1983; Foss & Ross, 1983; Foss
& Speer, 1991; Forster, 1979, 1994; Hess, Foss, & Carroll, 1995;
N. E. Sharkey & Mitchell, 1985; A. J. C. Sharkey & Sharkey,
1992; N. E. Sharkey & Sharkey, 1987). For example, report of
people in the sentence “the auto accident drew a large crowd of
people” may be facilitated because the appropriateness of this
word to the sentence context is determined faster, nonperceptually,
when sentences are legal and meaningful (i.e., plausibility checking mechanism; Norris, 1986; compound cue theory; Ratcliff &
McKoon, 1988; sophisticated buffer hypothesis; A. J. C. Sharkey
& Sharkey, 1989; lexical distance model; A. J. C. Sharkey &
Sharkey, 1992; global concept priming; Hess et al., 1995).
A well-documented finding in the word recognition literature is
that responses to words are facilitated when they are preceded by
a legal sentence context (e.g., Duffy, Henderson, & Morris, 1989;
Ehrlich & Rayner, 1981; Forster, 1981; Foss, 1982; Morris, 1994;
Ratcliff, 1987; A. J. C. Sharkey & Sharkey, 1992; N. E. Sharkey
& Sharkey, 1987; Simpson, Peterson, Casteel, & Burgess, 1989;
Stanovich & West, 1983; West & Stanovich, 1982). For example,
the word people is identified faster in the sentence “the auto
accident drew a large crowd of people” than in the sentence
“accident of large the drew auto crowd a people” (Simpson et al.,
1989). However, although this advantage is well known, the actual
influence exerted on word recognition remains to be determined.
Of particular importance for theories of sentence processing is
whether the influence of sentence context on word recognition is
perceptual or nonperceptual. For example, from an interactive
perspective, word recognition receives direct facilitation from
spreading activation produced by sentence content and structure,
producing preactivation of likely word candidates (e.g., Duffy et
al., 1989; McClelland & O’Regan, 1981; Morris, 1994; Simpson et
al., 1989). Accordingly, sentence effects arise from direct facilitation of processes underlying access to lexical entries such that
the process of visual word perception itself is facilitated. In line
with this view, the context “the auto accident drew a large crowd
of” preactivates the lexical entry for people and, when required to
report this target, participants’ responses are facilitated because
perception of the stimulus benefits from the existence of a preactivated match.
Identifying Perceptual Influences
The ready availability of nonperceptual influences is a major
obstacle to identifying the influence of sentence context on target
word perception. In particular, participants may use explicit
knowledge of sentence content and structure (i.e., semantic and
syntactic knowledge to deduce which target words can complete a
sentence; sentential constraint) and explicit knowledge of how
words are spelled (i.e., lexical knowledge to deduce what letters
can complete partially encoded target words; lexical constraint) to
enhance strategically their performance with target words in legal
sentence contexts. Thus, when tested in experiments, target word
performance may differ between different sentence contexts even
when the actual perception of target words remains unchanged
across conditions. This presents problems for many paradigms in
word recognition research, including those in which influences of
sentence context on target word performance are measured using
Timothy R. Jordan and Sharon M. Thomas, School of Psychology,
University of Nottingham, Nottingham, England.
This research was supported by Biotechnology and Biological Sciences
Research Council Grants S12111 and S02532 awarded to Timothy R.
Jordan. We are grateful to Ailis Murphy for running pilot experiments.
Correspondence concerning this article should be addressed to Timothy
R. Jordan, School of Psychology, University of Nottingham, University
Park, Nottingham NG7 2RD, England. E-mail: [email protected].
ac.uk
34
SENTENCE CONTEXT AND WORD RECOGNITION
naming, lexical decision, or fixation durations (e.g., Balota, Pollatsek, & Rayner, 1985; Balota & Rayner, 1983; Duffy et al., 1989;
Ehrlich & Rayner, 1981; Morris, 1994; O’Seaghdha, 1989, 1997;
Rayner & Well, 1996; Simpson et al., 1989). For example, in
naming studies, responses to targets may be faster and more
accurate in legal contexts (e.g., “the auto accident drew a large
crowd of people”) because target words can be predicted from
context and guessed from partial target information (e.g., peo—)
better than in control contexts (e.g., “accident of large the drew
auto crowd a people”) where no supporting context is present but
the same partial target information is perceived. Similar nonperceptual influences can also reduce lexical decision latencies and
target fixation durations, where targets may be identified using
minimal perceptual information enhanced strategically by explicit
knowledge of, in this example, the predictability of the word
people in the legal context and how this word is spelled. In
contrast, the same amount of target information may be insufficient to identify target words in control contexts that provided far
fewer constraints. In this case, participants would need more
information from the target, resulting in the longer lexical decision
latencies and longer fixation durations that are found. A crucial
step towards assessing the nature of sentential influences on word
perception, therefore, is to examine sentential influences under
conditions that suppress the ability of sentential and lexical constraints to enhance target performance nonperceptually.
One technique that satisfies these requirements is the twoalternative forced-choice procedure originally developed to investigate the relative perceptibility of individual words, nonwords,
and single letters under brief viewing conditions (generally referred to as the Reicher–Wheeler task, after Reicher, 1969, and
Wheeler, 1970). Using this procedure, each presentation of a
briefly presented target word is followed immediately by a forced
choice between alternative words that differ by one (“critical”)
letter whose serial position and probability of occurrence cannot be
predicted from any other letters in the target. For example, if the
word dark were presented, recognition of the third letter may be
tested by asking participants to choose between the alternatives
dark and dank, where (a) both critical letters, r and n, are consistent with the remainder of the word, (b) dark and dank are equally
likely to be presented in the experiment, and (c) any serial position
may be critical on a particular trial. Under these conditions, accurate performance hinges on perception of the critical part of the
target that cannot be determined nonperceptually from any other
part of the target or forced-choice alternative. Despite this stringent
control, the Reicher–Wheeler task has consistently revealed advantages for legal stimuli (e.g., words, pseudowords) over other
types of alphabetic display (e.g., nonwords, single letters), indicating differences in processes of perception rather than differences in nonperceptual enhancement (e.g., Baron & Thurston,
1973; Jordan, 1990, 1995; Jordan & Bevan, 1994, 1996; Jordan &
de Bruijn, 1993; Jordan, Patching, & Milner, 1998, 2000; Reicher,
1969; Wheeler, 1970). Indeed, the Reicher–Wheeler task appears
to rule out nonperceptual influences not just during the selection of
an overt response but at any stage of processing (see Bjork &
Estes, 1973; Johnston, 1978; Rumelhart & Siple, 1974; Thompson
& Massaro, 1973). For example, Johnston (1978) argued that if
nonperceptual constraint was being used at any stage in word
perception, performance should be more accurate for letters that
are highly constrained by the remainder of the word (such as the
35
final position of four, where only r and l are legitimate) than for
letters that are not (such as the final position of hear, which can be
completed by at least five alternatives—r, l, d, p, and t). Johnston
found no such trend using the Reicher–Wheeler task.
When applied to sentence research, the Reicher–Wheeler task
offers an assessment of the perceptibility of target words that is not
confounded by nonperceptual influences derived from sentential or
lexical constraints. For example, if the legal context “the cellar
was” were presented, followed by a brief presentation of a target
word (e.g., dark) in its appropriate location at the end of the
sentence, perception of the target word may be tested by asking
participants to choose between the alternatives dark and dank. Of
particular importance here is that both words are congruent with
the remainder of the sentence, dark and dank are equally likely to
be presented in the experiment, and any serial position of a target
word may be tested on a particular trial (in this case, Serial
Position 3). Under these conditions, participants are prevented
from using information from other words in the sentence or from
other positions in the target word to determine the identity of the
correct response; overall accuracy of target performance relies on
perception of the critical information in each target and this information cannot be determined by sentential or lexical constraint.
Consequently, if performance with target words in this context
were more accurate than in control conditions (e.g., when context
words are transposed or replaced by random words), this advantage could be attributed reliably to enhanced perceptual processing
of target words in legal sentence contexts rather than to nonperceptual enhancement based on sentential or lexical constraints.
Using Serial Position Performance to
Assess Context Effects
The Reicher–Wheeler task also provides a means of revealing
priorities attached to different positions in targets and so offers a
fine grained analysis of the effects of sentence contexts. For
example, many studies of single word perception report “outside–
in” patterns of identification performance, indicating that the exterior letters of words are unusually important in processing words
(e.g., Butler, Mewhort, & Tramer, 1987; Campbell & Mewhort,
1980; Humphreys, Evett, & Quinlan, 1990; Jordan & Bevan, 1994,
1996; Jordan & de Bruijn, 1993; Jordan et al., 2000; McCusker,
Gough, & Bias, 1981; Merikle, 1974; Merikle & Coltheart, 1972;
Merikle, Coltheart, & Lowe, 1971; Merikle & Glick, 1976; Mewhort & Campbell, 1978). A similar pattern of performance for
target words in sentence contexts would indicate that similar
processes of orthographic analysis were present in more general
reading situations. Moreover, if patterns of performance were
influenced by sentence context (i.e., legal vs. controls), these
differences should be revealed even if overall levels of target word
performance remain similar across context conditions. One previous study has provided information about target letter positions in
sentence processing, although the Reicher–Wheeler task was not
used. Ehrlich and Rayner (1981) recorded eye-fixation duration
and fixation probability for targets in high and low constraining
texts. When the original word was substituted for a visually similar
word differing by one letter, fixation duration and fixation probability were more likely to increase when the substitution involved
first or last letter positions (e.g., when right was substituted for
night). This led Ehrlich and Rayner to conclude that the linguistic
JORDAN AND THOMAS
36
content of sentences can reduce word identification thresholds and
suggests that exterior letter positions take priority in word processing during reading (see also Humphreys et al., 1990; Jordan, 1990,
1995; Jordan et al., 2000). It remains to be seen if similar findings
are produced by the Reicher–Wheeler task, which overcomes
nonperceptual enhancement and provides a more explicit test of
target perception in sentence contexts.
General Procedure
In this study, we investigated the advantage for words in legal
sentence contexts by adapting the Reicher–Wheeler task to provide
a new paradigm in sentence research that suppresses nonperceptual
influences of sentence context on target word performance. Under
these conditions, influences of legal sentence contexts on word
perception (relative to controls) should be revealed without contamination from nonperceptual enhancement. Legal sentences
were constructed using three words followed by one member of a
pair of matched target words that differed by just one letter (e.g.,
“the cellar was dark” vs. “the cellar was dank”) and where both
members of each word pair were syntactically legal and semantically viable. The three context words in each display were presented simultaneously as a complete string, followed by a brief
presentation of one of the two equally likely target words in its
appropriate location at the end of the (now complete) sentence (see
Method section, Experiment 1). Participants were then required to
select which of the two target words had been shown. Following
the arguments of Forster (1981; see also Duffy, et al., 1989),
control conditions were chosen to provide appropriate baselines
for assessing the influence of legal contexts on target word performance. According to Forster (1981), evidence of facilitative
effects for legal contexts are often due to comparisons with neutral
baselines that actually inhibit word recognition. For example,
neutral sentences such as “the word was dark” may create a greater
demand on integration or checking processes than “the cellar was
dark” and so impair performance. Thus, an advantage for target
words in the latter condition may mean that the influence of these
contexts is merely less inhibitory than the influence of neutral
contexts against which they are measured. From Forster’s (1981)
findings, random sequences of words provide a better baseline for
assessing facilitation provided by legal sentence contexts.
Four conditions were used in each of the three experiments
reported in this article. In legal contexts, target words were presented in a context that was syntactically legal and semantically
viable (e.g., “the cellar was dark”). Three control conditions were
then constructed to match the legal contexts. In word replacement
contexts, the words in legal contexts were replaced by words
selected randomly (Forster, 1981) with the restriction that replacements matched the originals for number and shape of letters
(ascender, descender, neutral) to minimize effects of simple visual
differences between conditions (e.g., “flu rubber won dark”). Nonword replacement contexts that were visually matched to legal and
word replacement contexts were used to investigate lexical influences on performance caused by using words as replacements
(e.g., “bhx sokkac mev dark”). Finally, transposed contexts were
formed by transposing the words in legal contexts so that they
contained the same lexical content as legal contexts but disrupted
legal sentence structure (e.g., “was cellar the dark”).
Experiment 1
Our first aim was to establish whether being in simple legal
sentences was sufficient to confer an advantage on target word
perception (as measured by our adaptation of the Reicher–Wheeler
task) when sentence contexts provide minimal cues (perceptual or
nonperceptual) to target word identity. Accordingly, target words
were attached to low-constraining legal sentence contexts that
minimized semantic and associative links between context words
and targets. For example, “it began to” was followed by either flap
or flop and participants had to decide which of these equally likely
target words was shown on each occasion. If legal sentence contexts exert only nonperceptual influences on target word perception (e.g., the modular perspective), accuracy of target word report
should not be higher in the legal contexts of this experiment,
relative to controls. Indeed, from the findings of Forster (1981), the
“unfocused” legal sentences used in this experiment may actually
produce poorer performance than word replacement controls. In
contrast, if legal sentence contexts influence the processes underlying the actual perception of words (e.g., the interactive perspective), accuracy of target word report should be higher in legal
contexts, relative to controls, despite the suppression of nonperceptual enhancement.
Pilot testing confirmed that 2 s was long enough for contexts to
be processed effectively. However, longer presentations may increase contextual influence. For example, Duffy et al. (1989)
found that increasing the duration between context and target
improved naming latencies for neutral and incongruent sentences.
Moreover, Keefe and Neely (1990; see also Seidenberg, Waters,
Sanders, & Langer, 1984) provided evidence that effects of context
take time to build to a facilitative level of activation. Therefore,
contexts were presented for 2 and 5 s in Experiment 1.
Method
Participants. Sixteen paid participants between the ages of 18 and 35
years were recruited from the University of Nottingham. Each participant
took part in two 40-min sessions and all had normal or corrected-to-normal
vision and were native speakers of British English.
Stimuli. Forty-four matched pairs (20 practice, 24 experimental) of
four-letter words were selected as target stimuli, with a mean frequency of
written occurrence of 71.88 per million and a mean difference within each
pair of 17.77 (SD ⫽ 13.66). Frequencies were calculated by averaging the
reported occurrences in Francis and Kučera (1982) and Carroll, Davies,
and Richman (1971). Each pair differed by only one letter and all four letter
positions were represented equally (e.g., walk/talk; cave/cove; flop/flap;
soul/soup).
Forty-four contexts, each consisting of three words, were constructed to
accompany each word pair in such a way that a legal phrase was formed
when either of the pair members was added at the end of the context. The
congruency of the final word in each sentence was assessed by a Cloze task
in which 32 judges from the same population as the participants in the
experiment were asked to generate a four-letter word most likely to
complete each three-word context (e.g., “it began to
”). No target
word was selected by any judge. Control contexts were constructed as
described previously. For example, “it began to flap/flop” (legal), “of taper
he flap/flop” (word replacement), “ek hojem bu flap/flop” (nonword replacement), and “to began it flap/flop” (transposed). Further examples of
these contexts are given in Table 1.
Visual conditions. Sentence contexts and target words were presented
in lowercase Courier font. Sentence contexts and target words were presented on one continuous line of text. Each complete sentence was shown
SENTENCE CONTEXT AND WORD RECOGNITION
Table 1
Sample Sentences, Separated by Context Type, Used In
Experiment 1
Alternative
target words
Context
Legal
he wanted to
look into the
it began to
what an odd
walk/talk
cave/cove
flap/flop
soul/soup
Word replacement
do armful be
bead aide flu
of taper he
sled is ill
walk/talk
cave/cove
flap/flop
soul/soup
Nonword replacement
ba sevlot hu
huet evda fbo
ek hojem bu
zloh uc eff
walk/talk
cave/cove
flap/flop
soul/soup
Transposed
to wanted he
look the into
to began it
odd what an
walk/talk
cave/cove
flap/flop
soul/soup
centered around a central fixation point. A single four-letter word subtended 1.2 degrees horizontally. Words were presented in black on a white
background. Background luminance was 23 cd/m2, foreground luminance
was 0.5 cd/m2.
Design. Participants took part in two 40-min sessions, one on each of
two different days. Each session consisted of 80 practice trials and two
target groups (A and B), each containing 96 experimental trials (24 target
words ⫻ 4 contexts). Target Group A consisted of all four contexts
completed by one randomly assigned member of each target pair. Target
Group B consisted of all four contexts completed by the other member of
each target pair. Both target groups were shown randomly intermingled in
each session. Both members of each target pair were presented as forcedchoice alternatives on each occasion. Each session was divided into cycles
of 16 pseudorandomly chosen stimuli counterbalanced across context
conditions and critical letter position. Each participant took part in both
context duration conditions (2 s and 5 s); to avoid disruption produced by
changing context duration within a session, the same duration was used
throughout each session. The order in which context duration was assigned
was counterbalanced across participants.
Apparatus. The experiment was controlled by a Cambridge VSG2/3
display controller (Cambridge Research Systems, Cambridge, England)
attached to a Pentium computer. The stimuli were presented on a highresolution display equipped with rapid-decay phosphor with a spot persistence time of 1 ms to 10%, at a viewing distance of 65 cm. The screen had
been modified to enable precise control of the visual angle subtended by
stimuli and to maximize the resolution of the display (Jordan & Martin,
1987). The experiment was conducted in a darkened booth and participants
entered their responses through an illuminated keyboard interfaced with the
computer.
Procedure. At the start of each trial, a fixation point appeared at the
center of the screen. Participants were instructed to fixate this point when
initiating a display. When participants pressed the return key on the
37
keyboard, the fixation point was replaced with the sentence context (for 2
or 5 s) before onset of the target word in its appropriate position at the end
of the sentence. The context display remained on the screen while the target
word was presented briefly (for a predetermined duration), completing the
sentence display. The whole screen went blank at the offset of the target
word. The forced-choice alternatives were shown 750 ms after target
offset, one above and one below (randomly determined) the screen position
previously occupied by the target. Alternatives were presented in uppercase
to prevent responses based on simple visual matching. Participants selected
either the upper or lower alternative by pressing the up or down arrow key
on the keyboard. After responding, participants were required to say “yes”
if they thought the complete sentence was legal (i.e., legal condition) and
“no” if they thought it was not (i.e., word replacement, nonword replacement, or transposed conditions). These responses were recorded by the
experimenter.
Throughout the practice and experimental section, target exposure duration was reassessed after each cycle of 16 trials. Within each cycle, all
targets were shown for the same exposure duration, irrespective of condition. Target exposure duration was increased by 6 ms if accuracy fell
below 11 (68.75%) correct and decreased by 6 ms if it rose above 13
(81.25%) correct within each cycle. When adjustments to exposure duration were made at the end of a cycle, the same adjustments were made for
each condition. This adjustment procedure ensured that overall performance fell in the midrange of the performance scale (theoretical midpoint ⫽ 75%) and that each condition was represented in the experiment at
the same exposure duration an equal number of times. Average exposure
duration of targets was 13 ms.
Results
The results of Experiment 1 are shown in Figure 1. Legality
responses were practically error free (⬍ 1%), indicating that participants processed sentences appropriately.
A within-subjects analysis of variance (ANOVA) was conducted with context duration (2 or 5 s), target group, context (legal,
word replacement, nonword replacement, and transposed), and
critical letter position as variables. Main effects were found for
context, F(3, 45) ⫽ 6.87, p ⬍ .01, and critical letter position, F(3,
45) ⫽ 3.92, p ⬍ .02. No other main effect or interaction approached significance. Newman–Keuls tests revealed that legal
contexts produced higher response accuracy than all other contexts
(all ps ⬍ .01) and transposed contexts produced higher response
accuracy than word replacement and nonword replacement contexts (all ps ⬍ .01). Response accuracy was higher for Positions 1
and 4 (79% and 78%, respectively) than for Positions 2 and 3 (66%
and 65% respectively; ps ⬍ .01). Analysis by items revealed a
similar pattern of effects: a main effect of context, F(3,
138) ⫽ 3.98, p ⬍ .01, critical letter position, F(3, 138) ⫽ 2.72,
p ⬍ .05, and no other main effect or interaction. Newman–Keuls
tests revealed the same pattern of differences as in the by-subjects
analysis.
Discussion
Despite using low-constraining contexts for legal sentences and
the Reicher–Wheeler paradigm to suppress nonperceptual influence, the findings of Experiment 1 show that accuracy of target
word report was greater for legal contexts than for word replacement, nonword replacement, and transposed controls. Moreover,
this advantage for words in legal contexts was consistent across
target groups and letter positions, indicating that although legal
38
JORDAN AND THOMAS
participants to process each context. However, although transposed sentences were reported reliably as “illegal” at both exposures, these contexts produced greater identification accuracy relative to word replacement and nonword replacement contexts
(although less than legal contexts). This suggests that sentence
coherence cannot account for all of the facilitation afforded by
legal contexts over controls and that absence of sentence coherence cannot account for the lack of facilitation found for
transposed contexts in previous studies (e.g., Masson, 1986;
O’Seaghdha, 1989, 1997; Simpson et al., 1989). Notably, these
previous studies used RSVP, where contexts unfold one word at a
time and which may accentuate the illegal order of words in
transposed sentences. In contrast, the advantage observed for
transposed contexts in Experiment 1 when each entire context was
presented simultaneously before and during target presentation (as
in natural reading situations) may be a more relevant indication
of the ability of transposed legal contexts to enhance word
perception.
Experiment 2
Figure 1. Mean percent correct report (% Correct) of target words presented in each context in Experiment 1. Contexts preceded presentation of
target words for 2 s (Panel A) and 5 s (Panel B).
contexts enhanced perception of targets, this enhancement was
achieved without changing the relative perceptibility of target
alternatives or the pattern with which orthographic information
was encoded in other context conditions. Indeed, the U-shaped
serial position curves we obtained in each context closely match
those previously obtained with single-string displays (e.g., Butler
et al., 1987; Campbell & Mewhort, 1980; Carr, Lehmkuhle, Kottas, Astor-Stetson, & Arnold, 1976; Jordan & Bevan, 1994; Jordan
& de Bruijn, 1993; Jordan, Smith, & Phillips, 1995; Merikle, 1974;
Merikle & Coltheart, 1972; Merikle et al., 1971; Merikle & Glick,
1976; Mewhort & Campbell, 1978; Prinzmetal, 1992; Prinzmetal
& Silvers, 1994; Rumelhart & McClelland, 1982; see also Jordan
1990, 1995; Jordan et al., 2000; Jordan, Thomas, & Scott-Brown,
1999), suggesting that these earlier findings reflect processes of
orthographic analysis that occur with sentential material. Although
not the focus of the present study, this similarity suggests that
models of word recognition developed to account for data obtained
in single word recognition tasks are relevant to word recognition
processes used during reading and mirrors the finding of Ehrlich
and Rayner (1981) that exterior letters take priority (although see
Briihl & Inhoff, 1995).
Context exposures of 2 and 5 s produced similar effects of
context and both produced virtually error-free identification of the
legality of sentences, suggesting that 2-s exposures were ample for
To provide a relatively straightforward measure of the effects of
legal sentence contexts in the Reicher–Wheeler paradigm, the legal
contexts used in Experiment 1 were low constraining to minimize
semantic and associative links between contexts and targets. However, sentence contexts in everyday experience often provide cues
to the identity of words that are greater than those provided by the
legal contexts of Experiment 1. Therefore, in Experiment 2, legal
contexts were now more predictive of target words while both
members of each target pair still matched the context in which they
were presented; for example, “the horse seemed lame/tame,” “the
stone was opal/oval,” “the cellar was dark/dank,” and “the house is
near/neat.”
Method
Participants. Sixteen paid participants were recruited from the same
population as in Experiment 1 and took part in a 40-min session. All had
normal or corrected-to-normal vision and were native speakers of British
English. None had taken part in Experiment 1.
Stimuli. Forty-four pairs (20 practice, 24 experimental) of matched
target words and associated legal contexts were constructed in the same
way as Experiment 1. Targets had a mean frequency of written occurrence
of 77.43 per million and a mean difference within each pair of 22.13
(SD ⫽ 17.49). Legal sentences now reflected concrete concepts; each
context contained a highly imageable subject noun (e.g., horse, stone,
cellar, house) followed by a descriptive four-letter target word (e.g.,
lame/tame, opal/oval, dark/dank, near/neat). The congruency of the final
word in each sentence was assessed by a Cloze task in which 32 judges
from the same population as participants were asked to generate the
four-letter word most likely to complete each three-letter context (e.g., “the
horse was
”). All target words were selected by between 6 and 16
judges. Control contexts were constructed as in Experiment 1. However, to
control for intralexical priming between context words and targets, the
subject noun in each phrase was kept the same distance (i.e., separated by
the same number of words) from the target as in legal contexts (e.g., “was
cellar the dark/dank;” for discussion, see Duffy et al., 1989; Gough, Alford,
& Holley-Wilcox, 1981; A. J. C. Sharkey & Sharkey, 1992; Simpson et al.,
1989).
Design. No effect of context duration was observed in Experiment 1.
Consequently, because 2 s was the more natural reading time for such short
SENTENCE CONTEXT AND WORD RECOGNITION
phrases, all contexts were presented for 2 s. All other aspects of Experiment 2 were the same as Experiment 1.
Results
The results of Experiment 2 are shown in Figure 2. Legality
responses were practically error free (⬍1%), indicating that participants processed sentences appropriately.
A within-subjects ANOVA, with target group, context, and
critical letter position as variables, showed main effects of context,
F(3, 45) ⫽ 8.34, p ⬍ .01, and critical letter position, F(3,
45) ⫽ 4.42, p ⬍ .01. No other main effect or interaction approached significance. Newman–Keuls tests revealed that legal
contexts produced higher response accuracy than all other contexts
(all ps ⬍ .01) and that transposed contexts produced higher response accuracy than word replacement and nonword replacement
contexts (all ps ⬍ .01). Response accuracy was higher for Positions 1 and 4 (80% and 79%, respectively) than for Positions 2
and 3 (66% and 65%, respectively; ps ⬍ .01). Analysis by items
revealed a similar pattern of effects: a main effect of context, F(3,
138) ⫽ 5.39, p ⬍ .01, and no other main effect or interaction.
Newman–Keuls tests revealed the same pattern of differences as in
the by-subjects analysis.
Discussion
The findings of Experiment 2 are in close agreement with those
of Experiment 1. Despite the use of the Reicher–Wheeler task to
suppress nonperceptual influence, target identification accuracy
was greater for legal contexts than for word replacement, nonword
replacement, or transposed controls. Moreover, the enhancement
produced by legal contexts was again consistent across target
groups and letter positions, indicating that legal contexts did not
change the relative perceptibility of target alternatives or the
pattern with which orthographic information was encoded in other
context conditions. The advantage produced by transposed contexts relative to word replacement and nonword replacement contexts was also replicated in Experiment 2. Consequently, this
advantage was not a quirk of the stimuli used in Experiment 1 and
Figure 2. Mean percent correct report (% Correct) of target words presented in each context in Experiment 2. Contexts preceded presentation of
target words for 2 s.
39
supports the notion that when contexts are presented simultaneously before and during target presentation (as in natural reading
situations), transposed legal contexts can provide some enhancement of target word perception, despite loss of sentence structure.
Experiment 3
So far, the advantage for words presented in legal sentence
contexts has been obtained using contextual constraints selected to
match both members of each target word pair with equal congruency. However, if sentence context truly affects the perceptibility
of words, legal contexts that constrain strongly the identity of
target words should produce different levels of perceptual enhancement for target words of different contextual congruency.
For example, when presented with the context “born to be
,”
the target word wild would be a highly congruent completion. In
contrast, the matched target word mild would be a very incongruent completion. From previous research, highly constraining contexts should improve performance more for congruent targets than
for incongruent targets, relative to baseline context conditions
(e.g., Duffy et al., 1989; Fischler & Bloom, 1979; O’Seaghdha,
1989; Stanovich & West, 1983; West & Stanovich, 1982). However, it remains to be seen whether a similar distinction between
congruent and incongruent targets is observed when nonperceptual
influences on target word performance are suppressed by using the
Reicher–Wheeler task.
From a perceptual perspective, strong contextual constraints
should preactivate probable word candidates, which then facilitates
the perception of these items. Thus, when presented with the
context “born to be
,” perception of the congruent target
word will be facilitated because the visual input from the target
word will correspond to the preactivated candidate. In contrast,
preactivation by this context should not facilitate perception of the
incongruent target mild. Consequently, when testing perception of
these words in the Reicher–Wheeler task, a greater advantage
should be observed for wild than for mild, relative to word replacement and nonword replacement baselines where no contextual advantage for congruent over incongruent targets should be
produced. However, it could be argued that the frequent cooccurrence of words in familiar phrases creates associations between their lexical entries (e.g., Fodor, 1983; Forster, 1981). As a
result, a context word may activate not only its own entry but also
that of the target, producing spurious “context effects;” for example, born may prime wild. Transposed contexts (e.g., “be born to
wild”) should reveal these effects of lexical priming between
contexts and targets, if they occur. Indeed, in light of the advantages already observed for transposed contexts with apparently
much less capacity for lexical priming (Experiments 1 and 2),
transposed highly constraining contexts may produce substantial
advantages for congruent targets, if lexical priming is present.
If the arguments concerning the ability of the Reicher–Wheeler
task to suppress nonperceptual influences are valid (see the introduction), an advantage for congruent target words (e.g., wild) over
their incongruent counterparts (e.g., mild) should not be contaminated by nonperceptual enhancement. Indeed, if nonperceptual
influences were present in the task, strong contextual constraints
should bias responses towards congruent alternatives, and so produce more errors for incongruent targets relative to baseline conditions. If this disadvantage for incongruent targets were found, the
40
JORDAN AND THOMAS
suitability of the Reicher–Wheeler task for studying contextual
influences on word perception would be brought into doubt. However, if an advantage for congruent targets were observed with no
disadvantage for incongruent targets relative to baselines, the view
that performance in the Reicher–Wheeler task reflects perceptual rather than nonperceptual contextual influences would be
consolidated.
Accordingly, Experiment 3 used strongly constraining legal
contexts derived from well-known sayings, slogans, and song
titles. These target words were then each paired with a visually
matched alternative to produce a legal but incongruent ending
(e.g., “born to be wild/mild,” “singing in the rain/ruin,” “go with
the flow/flaw,” and “lie of the land/lane”).
Method
Participants. Sixteen paid participants, from the same population as in
Experiments 1 and 2, took part in a 60-min session. All had normal or
corrected-to-normal vision and were native speakers of British English.
None had taken part in Experiments 1 or 2.
Stimuli. Familiar phrases were chosen from well known sayings, slogans, and song titles that ended in four-letter words (e.g., “born to be
wild”). The congruency of the final word in each sentence was assessed by
a Cloze task in which 32 judges from the same population as participants
were asked to generate the four-letter word most likely to complete each
three-letter context (e.g., “born to be
”). Fifty-six (24 practice, 32
experimental) phrases that generated the same target word for all judges
were chosen as congruent targets in the experiment. Each congruent target
word was then partnered with a visually matched incongruent alternative
that differed from the target by only one letter and produced a legal but
incongruent ending (e.g., “born to be wild” vs. “born to be mild”); clearly,
no incongruent targets had been selected by judges in the Cloze task.
Targets had a mean frequency of written occurrence of 54.83 per million
and a mean difference within each pair of 24.54 (SD ⫽ 19.68). All other
aspects of this experiment were the same as Experiment 2.
Results
The results of Experiment 3 are shown in Figure 3. Legality
responses were again essentially error free (⬍ 1%), indicating that
participants processed sentences appropriately.
Figure 3. Mean percent correct report (% Correct) of target words presented in each context in Experiment 3. Contexts preceded presentation of
target words for 2 s.
A within-subjects ANOVA, with target group (congruent, incongruent), context, and critical letter position as variables,
showed main effects of target group, F(1, 15) ⫽ 22.43, p ⬍ .01,
context, F(3, 45) ⫽ 3.64, p ⬍ .02, critical letter position, F(3,
45) ⫽ 3.56, p ⬍ .05, and an interaction between target group and
context, F(3, 45) ⫽ 5.78, p ⬍ .01. Newman–Keuls tests revealed
that response accuracy for congruent targets was greater in legal
contexts than in any other context (all ps ⬍ .01), although response
accuracy for incongruent targets did not differ across contexts (all
ps ⬎.10). In addition, response accuracy was greater for congruent
targets than incongruent targets in legal contexts ( p ⬍ .01) and
in transposed contexts ( p ⬍ .05) but in no other contexts (all ps ⬎
.10). As before, response accuracy was higher for Positions 1 and 4
(82% and 80%, respectively) than for Positions 2 and 3 (69% and
67%, respectively; ps ⬍ .01).
Analysis by items revealed similar findings: main effects of
target group, F(1, 62) ⫽ 12.13, p ⬍ .01, context, F(3, 186) ⫽ 3.89,
p ⬍ .01, critical letter position, F(3, 186) ⫽ 3.01, p ⬍ .05, and an
interaction between target group and context, F(3, 186) ⫽ 6.82,
p ⬍ .01. Newman–Keuls tests revealed the same pattern of differences as in the by-subjects analysis.
Discussion
When presented in highly constraining legal contexts, performance accuracy for congruent targets was substantially greater
than for matched incongruent controls. However, this advantage
for congruent target words was not present in baseline contexts
(word replacement, nonword replacement) and performance accuracy for incongruent targets remained unchanged across all context
conditions. Thus, the highly constraining legal contexts used in
Experiment 3 enhanced performance with congruent targets without affecting performance with incongruent controls (relative to
other context conditions). This finding, in particular, suggests that
legal contexts increased the perceptibility of congruent targets
rather than merely inspired a bias in participants to select the more
congruent of the alternative responses on any trial. Moreover, as in
Experiments 1 and 2, the influence of legal contexts on target
performance (congruent and incongruent) was consistent across
all letter positions, indicating that legal contexts did not alter
the pattern of orthographic encoding present in other context
conditions.
However, the finding that legal contexts produced no benefit for
incongruent target words over all other contexts is a departure
from the benefits observed for all target words in the legal contexts
of Experiments 1 and 2. This suggests that although low constraining legal contexts can improve perception of several candidate
words, high constraining legal contexts produce benefits that are so
focused that perception of only the word “defined” by each context
is improved. Indeed, performance in transposed contexts suggests
that similar word-specific benefits were taking place, presumably
through lexical association, indicating that transposed contexts
produced a subsidiary enhancement in target performance that
resembled a weaker version of that observed with legal contexts.
The advantage observed for congruent words in transposed contexts indicates again that transposed legal contexts can enhance
target word perception despite loss of sentence structure. However,
despite the greater opportunity for stronger lexical associations
between context and (congruent) target words in Experiment 3
SENTENCE CONTEXT AND WORD RECOGNITION
than in Experiments 1 and 2, the difference in the amount of
improvement produced by legal and transposed contexts did not
decrease in Experiment 3 and has remained remarkably similar
across all three experiments (about 10%). We return to these points
in the General Discussion.
General Discussion
The three experiments reported in this article reexamined the
well-established advantage for words presented in legal sentence
contexts using an adaptation of the Reicher–Wheeler task to suppress nonperceptual influences of contextual and lexical constraint.
Specifically, the ability of participants to identify briefly presented
target words in different context conditions was tested using a
forced-choice procedure in which accurate target identification
relied on perception of specific target information that could not be
deduced nonperceptually from the context in which a target word
was presented or from information present in other (noncritical)
parts of each target word. Despite this stringent testing procedure,
each experiment showed an advantage for words presented in legal
contexts over all control contexts (word replacement, nonword
replacement, and transposed). Indeed, this advantage was observed
not only when legal contexts provided strong contextual constraints (Experiment 3) but also when legal contexts provided
much weaker indications of target identity (Experiments 1 and 2).
Moreover, in contrast to previous findings (e.g., Masson, 1986;
O’Seaghdha, 1989, 1997; Simpson et al., 1989), an advantage for
target words was also produced when legal contexts were transposed, even when the content of these contexts seemed to provide
very little opportunity for lexical priming (Experiment 1).
The first indication to emerge from these findings is that legal
sentence contexts can facilitate the actual perception of words
rather than merely the ability of nonperceptual processes to assist
in the selection of the most likely candidate. For example, in a
“strong” modular account of sentence processing, legal sentence
contexts facilitate word identification by guiding nonperceptual
decisions about a word’s identity made on the basis of output from
the perceptual module (e.g., Fodor, 1983, 1998; Forster, 1979,
1981, 1985, 1994; Seidenberg, 1985). Under normal viewing conditions, output from the perceptual module would usually correspond to the correct identity of a target word that would then be
integrated into the remaining words in the sentence. However, if
the perceptual module is unable to identify a word (e.g., as in the
brief presentation conditions of the experiments reported in this
article), nonperceptual influences based on sentential constraint
provided by legal contexts should increase the likelihood of accurate target identification, especially when lexical constraint provided by partial target word cues can facilitate guessing. In the
present study, the potential benefit of these influences was nullified by the nature of the task, where accurate identification of
targets required the perception of critical information that could
not be determined by any other information in the display, in either
the context or the target. Indeed, even when contexts were strongly
constraining, and so provided ample opportunity for nonperceptual
calculation of the likely identity of each target word (Experiment
3), performance with incongruent targets showed no indication that
participants were using context in this way; that is, relative to the
levels of performance observed in baseline contexts, strongly
41
constraining contexts inspired more frequent report of congruent
targets only when congruent targets were shown.
A modification of the strong modular account (e.g., Fodor,
1983; Forster, 1979), which Potter, Moryadas, Abrams, and Noel
(1993) termed the lexical-priming modular model, does little to
improve the ability of this general class of theory to explain the
findings obtained. According to the modified account, frequent
co-occurrence of words in discourse creates associations between
their lexical entries that lead to intralexical priming when these
words are encountered in legal sentence contexts where cooccurrence is more likely to occur. However, target identification
was better in legal contexts than in transposed contexts (which
contained the same words as legal contexts) in all three experiments, suggesting that target identification was enhanced by more
than intralexical priming. Moreover, if this modified modular
account of our findings were to succeed, intralexical priming
would have had to occur between target words and words in the
weakly constraining contexts (legal and transposed) used in Experiment 1 (e.g., “it began to flap/flop”). This does not seem likely.
The evidence also appears to be inconsistent with less strong
modular/interactive models (i.e., the expectancy model; Becker,
1980; Becker & Killion, 1977; and the modular/interactive model;
Potter, et al. 1993; Potter, Stiefbold, & Moryadas, 1998) where,
despite including interactivity between contextual and perceptual
levels of processing, the locus of contextual influence remains at
the final word selection stage. In these models, perceptual processing of a target word produces a set of candidates, weighted
according to stimulus evidence. At the next stage, context interacts
with these weights to produce a single best candidate, which is
then the word consciously perceived. However, it is difficult to see
why contexts in Experiments 1 and 2 should have led to the
selection of the target presented on each occasion as the best
candidate because none of these contexts specified either member
of a target pair. Indeed, performance across the members of each
pair showed no change across context conditions in either of these
experiments, indicating that legal contexts did not produce different weightings for target words relative to contextless baseline
(word replacement, nonword replacement) or transposed conditions. Moreover, congruent alternatives in Experiment 3 would
have received more weight than their incongruent pair mates when
presented in the strongly constraining contexts used in that experiment, thus decreasing the number of incongruent targets identified. However, as we have already pointed out, high constraining
contexts in Experiment 3 inspired more frequent report of congruent targets only when congruent targets were shown.1
1
A similar argument can be applied to other putative nonperceptual
influences on performance. For example, one virtue of the Reicher–
Wheeler task is that it minimizes the role of memory in performance (i.e.,
only one word need be remembered on each trial) but effects of memory
could conceivably be invoked to explain our findings. In particular, it could
be argued that targets in all contexts in our experiments were perceived
equally well but, when targets were not congruent with a context, memory
for the original (veridical) percept may be impaired and report accuracy
diminished. If this were occurring, target report should be particularly
inaccurate for targets (e.g., mild) in contexts that are particularly incongruent (“born to be”) where nonperceptual (in this case, memorial) processes support systematically the congruent alternative (wild) even when
42
JORDAN AND THOMAS
In contrast, the findings of this study are consistent with the
view that legal context facilitates the actual perception of words
and so support interactive theories of sentence processing. Theorists such as Morton (1969), Massaro (1979), and McClelland
(1987) have proposed an interactive architecture in which sensory,
lexical, and contextual information are used jointly to determine
the identity of a word before awareness. Although some of the
details of these models differ (see, e.g., Stolz & Besner, 1998), all
share the critical view that the context in which a word is encountered can affect the process of perceiving that word. In interactive
models, perceptual processing generates evidence for and against
each of a large number of words (in effect, all the words in the
lexicon). In parallel, information from sentence context generates
evidence for and against these same word items. The resulting
output from this interactive process is the candidate word that
achieves the highest level of activation.
From the findings of Experiments 1 and 2, legal sentence
contexts enhance word perception even when context provides
apparently little capacity to constrain the identity of each target
word (especially so in Experiment 1). This finding is in agreement
with the findings of previous studies in which facilitation of target
identification has been reported in the absence of lexical relatives
of each target (e.g., Foss, 1982; Hess et al., 1995; A. J. C. Sharkey
& Sharkey, 1992; Williams, 1988) and suggests that syntactic
coherence facilitated perception of words in our study. Moreover,
while Forster (1981) has argued that previous findings may be
explained by nonperceptual guesswork based on sentential and
lexical constraints, the present findings are not susceptible to this
caveat. First, we used the Reicher–Wheeler task to ensure that
accurate target identification could not be inferred nonperceptually
from either the context in which a target word was presented or
from information present in the noncritical parts of each target
word. Second, incongruent targets in Experiment 3 did not incur a
disadvantage relative to performance in baseline contexts (which
were selected in accord with the arguments for appropriate baseline contexts presented by Forster, 1981). As we have already
argued (and as Forster, 1981, also points out), if nonperceptual
guesswork were present, a disadvantage would be expected for
incongruent targets in legal contexts relative to contextless baseline controls.
Further evidence that syntactic coherence influences target perception comes from comparisons between performance in legal
and transposed contexts. In line with previous studies, transposed
contexts were included to provide a measure of influences of
lexical relatedness on target performance (e.g., Masson, 1986;
O’Seaghdha, 1989, 1997; Simpson et al., 1989). In line with the
findings of these previous studies, target performance was facilitated less in transposed contexts than in legal contexts in all three
experiments. However, in contrast to previous findings, transposed
contexts did produce an advantage for targets relative to baseline
contexts. At first sight, this suggests that intralexical activation
between context and target words provided some degree of facil-
incongruent targets are shown. However, report accuracy for targets in
incongruent legal contexts was no lower than in word replacement and
nonword replacement contexts that were incongruent with both types of
target and offered no systematic memorial support for one alternative more
than the other.
itation for target perception. However, it is hard to reconcile this
explanation with the facilitation produced by the weak constraining contexts of Experiment 1 (e.g., “to began it flap/flop”) where
semantic or associative overlap between context words and targets
was essentially nonexistent. Moreover, levels of facilitation produced by transposed contexts relative to legal and baseline contexts remained virtually unchanged across Experiments 1–3, despite substantial changes in the constraint provided by legal
contexts and the content of transposed controls. Other objections to
a role for intralexical priming in sentence context effects have also
been raised by other researchers (e.g., Auble & Franks, 1983;
Gough et al., 1981; Ratcliff, 1987; Ratcliff, Hockley, & McKoon,
1985; Simpson et al., 1989), and the general appeal of this account
of sentence context effects appears limited.
An alternative explanation of the facilitation produced by transposed contexts in the present study is that when words in transposed contexts are presented simultaneously, as in normal text (cf.
the RSVP tasks used by Masson, 1986, O’Seaghdha, 1989, 1997,
and Simpson et al., 1989), processes of syntactic analysis attempt
initially to derive meaning from these contexts. Thus, when presented with the context “to began it” and the target flap or flop,
participants may have attempted initially to “legalize” this sentence, which gave rise to a residual legal context effect, although
further processing revealed the ultimate illegality of the display (as
evidenced by the essentially error-free reports of sentence legality
in all three experiments). Indeed, from the comments made by
many of our participants, words presented in transposed contexts
may actually be perceived (at least initially) in a syntactically legal
arrangement, which ultimately becomes resolved as further processing takes place. We are currently researching this issue.
The finding that target word perception was facilitated even
when legal contexts were low constraining (Experiments 1 and 2)
rather than only when they were high (Experiment 3) suggests that
the facilitative effect of contextual information on word perception
is a common occurrence in reading, requiring only that words are
presented in grammatically legal sentences. In particular, while it
may often be impossible to predict overtly the precise identity of
a word from the sentential context in which it is presented (e.g.,
Experiment 1; Gough et al., 1981; McConkie & Rayner, 1976),
influences from even low constraining contexts appear to be sufficient to facilitate processes of word perception by providing
additional activation for word recognition to achieve threshold. For
example, consider the low constraining context “it began to flap/
flop” used in Experiment 1. When the target word was presented
(e.g., flap), the visual information encoded from this word may
often have been sufficient to activate the node for the target word
more than nodes for other words, but insufficient, by itself, to
exceed recognition threshold. Specifically, although activation
would have been produced, this activation is likely to have occurred for a number of visually and orthographically consistent
words (including the nontarget alternative, flop) that may be mutually inhibitory, such that the net facilitation that resulted for the
appropriate word node was insufficient to exceed threshold. Similarly, a weak source of contextual constraint (“it began to”) may
also produce activation for a number of word nodes (e.g., constrained by syntax and, where appropriate, semantics) that also
result in mutual inhibition and little net facilitation for a particular
word node. However, if this weak source of contextual activation
were coupled with the activation produced by the visual informa-
SENTENCE CONTEXT AND WORD RECOGNITION
tion encoded from the target word, a single word node may receive
sufficient activation from both sources to surpass the interactive
threshold (see also Duffy et al. 1989; McClelland, 1987; McClelland & O’Regan, 1981; Morris, 1994; Simpson et al., 1989).
However, the findings of Experiment 3 suggest that the effects
of high constraining contexts were somewhat different from those
produced by the low constraining contexts of Experiments 1 and 2.
In particular, perception of only those targets (e.g., wild) that were
congruent with the high constraining contexts (e.g., “born to be”)
was facilitated; perception of incongruent targets (e.g., mild)
showed no benefit over control contexts, despite being presented in
a context that was still grammatically legal and despite being
visually very similar to their congruent counterparts. These findings suggest that, when sentential constraint was high, the facilitation produced was available only for the word that fitted each
context. Thus, while the visual information encoded from each
congruent and incongruent target word may have activated several
word nodes, and so been insufficient by itself to exceed threshold
in the node corresponding to the target, the activation produced by
the context was sufficiently specific to activate only the node
corresponding to the contextually defined target word and no
others. However, since performance with incongruent items was
no worse than in control contexts, this high degree of specificity
does not seem to have inspired a general bias towards selecting
each congruent alternative. Rather, when contextual constraint was
high, visual analysis of congruent targets seems to have been
facilitated but the role of visual input was not diminished, allowing
even visually similar incongruent alternatives to be rejected reliably. These findings have some resonance with those of Balota et
al. (1985), who investigated effects of sentential constraint on
parafoveal processing by varying the predictability of target words
and the availability of target information in parafoveal previews
using the boundary technique. Of particular relevance is the finding that the benefit of parafoveal preview was greater when a
target word was predictable, suggesting that identification of visual
information in the parafovea was more accurate when aided by
sentential context. These findings could be explained by nonperceptual influences of sentential (and lexical) constraint, which
enhanced the accuracy with which the orthographic content of
words in the parafovea was deduced. However, the complementarity of these findings to our own suggests that Balota et al. were
right to conclude that high sentential constraint enhances perception of visual information in text. Indeed, as Balota et al. point out,
the influence of context on performance they observed indicates
that, when reading, the threshold for word recognition is set rather
high, such that, even in high constraining contexts, considerable
importance is placed on the consistency of contextual and visual
information.
Finally, the similar patterns of serial position performance observed across the four context conditions in each experiment
suggest that advantages for target words in legal contexts were not
produced by fundamental changes in the way words were perceived. Rather, improvements in legal contexts appear to reflect
enhancement of the same processes of orthographic encoding in all
context conditions.2 Moreover, the advantage for exterior letters
we obtained is consistent with the view that exterior letters of
words are unusually important in word recognition and resembles
findings obtained previously with isolated displays (e.g., Butler et
al., 1987; Campbell & Mewhort, 1980; Carr et al., 1976; Jordan
43
& Bevan, 1994; Jordan & de Bruijn, 1993; Jordan et al., 1995;
Merikle, 1974; Merikle & Coltheart, 1972; Merikle et al., 1971;
Merikle & Glick, 1976; Mewhort & Campbell, 1978; Prinzmetal,
1992; Prinzmetal & Silvers, 1994; Rumelhart & McClelland,
1982; see also Jordan 1990, 1995; Jordan et al., 1999; Jordan et al.,
2000) and by Ehrlich and Rayner (1981) using textual displays. It
should be noted, however, that we make no claim that an exterior
letter advantage will persist throughout the perceptual processing
of words. Indeed, it has been known for some time that serial
position curves can change as exposure durations increase (e.g.,
Merikle & Glick, 1976) and identification of all letters will presumably eventually reach asymptote. However, this does not detract from the finding that target word performance in our experiments was underpinned by the same “outside–in” pattern of
orthographic analysis in all context conditions and that this pattern
of analysis was sufficient to produce high levels of target identification in legal contexts.
In summary, the findings of this article indicate that legal
sentence contexts facilitate the identification of words when tested
under conditions that suppress contamination from nonperceptual
influences of sentential and lexical constraint. Moreover, the advantage for legal contexts over transposed controls indicates a role
for syntactic coherence rather than merely intralexical priming,
although transposed contexts did provide a residual advantage over
baseline contexts. Consequently, although nonperceptual influences (e.g., of explicit prediction, expectation, or guesswork)
based on sentential and lexical constraints may be present in a
range of reading situations, the findings of this study support the
notion that the very perception of words can be enhanced when
words are presented in legal sentence contexts.
2
Although the actual influence of nonperceptual processes on serial
position performance is hard to predict, the consistent pattern of serial
position performance observed across all contexts is further evidence that
target performance in legal contexts was not distorted by nonperceptual
enhancement.
References
Auble, P., & Franks, J. J. (1983). Sentence comprehension processes.
Journal of Verbal Learning and Verbal Behavior, 22, 395– 405.
Balota, D. A., Pollatsek, A., & Rayner, K. (1985). The integration of
contextual constraints and parafoveal visual information in reading.
Cognitive Psychology, 17, 364 –390.
Balota, D. A., & Rayner, K. (1983). Parafoveal visual information and
semantic contextual constraints. Journal of Experimental Psychology:
Human Perception and Performance, 9, 726 –738.
Baron, J., & Thurston, I. (1973). An analysis of the word superiority effect.
Cognitive Psychology, 4, 207–228.
Becker, C. A. (1980). Semantic context effects in visual word recognition:
An analysis of semantic strategies. Memory & Cognition, 8, 493–512.
Becker, C. A., & Killion, T. H. (1977). Interaction of visual and cognitive
effects in word recognition. Journal of Experimental Psychology: Human Perception and Performance, 3, 389 – 401.
Bjork, E., & Estes, E. K. (1973). Letter identification in relation to
linguistic context and masking conditions. Memory & Cognition, 1,
217–223.
Briihl, D., & Inhoff, A. W. (1995). Integrating information across fixations
during reading: The use of orthographic bodies and of exterior letters.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 21, 55– 67.
44
JORDAN AND THOMAS
Butler, B. E., Mewhort, D. J. K., & Tramer, S. C. (1987). Location errors
in tachistoscopic recognition: Guesses, probe errors or spatial confusions? Canadian Journal of Psychology, 41, 339 –350.
Campbell, A. J., & Mewhort, D. J. K. (1980). On familiarity effects in
visual information processing. Canadian Journal of Psychology, 34,
134 –153.
Carr, T. H., Lehmkuhle, S. W., Kottas, B., Astor-Stetson, E. C., & Arnold,
D. (1976). Target position and practice in the identification of letters in
varying contexts: A word superiority effect. Perception & Psychophysics, 19, 412– 416.
Carroll, J. B., Davies, P., & Richman, B. (1971). The American Heritage
word frequency book. Boston: Houghton Mifflin.
Duffy, A. A., Henderson, J. M., & Morris, R. K. (1989). Semantic facilitation of lexical access during sentence processing. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 791– 801.
Ehrlich, S. F., & Rayner, K. (1981). Contextual effects on word perception
and eye movements during reading. Journal of Verbal Learning and
Verbal Behavior, 20, 641– 655.
Fischler, I., & Bloom, P. A. (1979). Automatic and attentional processes in
the effects of sentence contexts on word recognition. Journal of Verbal
Learning and Verbal Behavior, 18, 1–20.
Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT Press.
Fodor, J. A. (1998). Concepts: Where cognitive science went wrong. New
York: Oxford University Press.
Forster, K. I. (1979). Levels of processing and the structure of the language
processor. In W. Cooper & E. Walker (Eds.), Sentence processing:
Psycholinguistic studies presented to Merrill Garett (pp. 27– 85). Hillsdale, NJ: Erlbaum.
Forster, K. I. (1981). Priming and the effects of sentence and lexical
contexts on naming time: Evidence for autonomous lexical processing.
Quarterly Journal of Experimental Psychology: Human Experimental
Psychology, 33(A), 465– 495.
Forster, K. I. (1985). Lexical acquisition and the modular lexicon. Language and Cognitive Processes, 1, 87–108.
Forster, K. I. (1994). Computational modeling and elementary process
analysis in visual word recognition. Journal of Experimental Psychology: Human Perception and Performance, 20, 1292–1310.
Foss, D. J. (1982). A discourse on semantic priming. Cognitive Psychology, 14, 590 – 607.
Foss, D. J., & Ross, J. R. (1983). Great expectations: Context effects during
sentence processing. In G. B. Flores d’Arcais & R. J. Jarvella (Eds.), The
process of language understanding (pp. 169 –191). New York: Wiley.
Foss, D. J., & Speer, S. R. (1991). Global and local context effects in
sentence processing. In R. R. Hoffman & D. S. Palermo (Eds.), Cognition and the symbolic processes: Applied and ecological perspectives
(pp. 115–139). Hillsdale, NJ: Erlbaum.
Francis, W. N., & Kučera, H. (1982). Frequency analysis of English usage:
Lexicon and grammar. Boston: Houghton Mifflin.
Gough, P. B., Alford, J. A., & Holley-Wilcox, P. (1981). Words and
contexts. In O. J. L. Tzeng & H. Singer (Eds.), Perception of print:
Reading research in experimental psychology (pp. 85–102). Hillsdale,
NJ: Erlbaum.
Hess, D. J., Foss, D. J., & Carroll, P. (1995). Effects of global and local
context on lexical decision processing during language comprehension.
Journal of Experimental Psychology: General, 124, 62– 82.
Humphreys, G. W., Evett, L. J., & Quinlan, P. T. (1990). Orthographic
processing in visual word identification. Cognitive Psychology, 22,
517–560.
Johnston, J. C. (1978). A test of the sophisticated guessing theory of word
perception. Cognitive Psychology, 10, 123–153.
Jordan, T. R. (1990). Presenting words without interior letters: Superiority
over single letters and influence of postmask boundaries. Journal of
Experimental Psychology: Human Perception and Performance, 16,
893–909.
Jordan, T. R. (1995). Perceiving exterior letters of words: Differential
influences of letter-fragment and non-letter-fragment masks. Journal of
Experimental Psychology: Human Perception and Performance, 21,
512–530.
Jordan, T. R., & Bevan, K. M. (1994). Word superiority over isolated
letters: The neglected case of forward masking. Memory & Cognition,
22, 133–144.
Jordan, T. R., & Bevan, K. M. (1996). Position-specific masking and the
word-letter phenomenon: Re-examining the evidence from the Reicher–
Wheeler paradigm. Journal of Experimental Psychology: Human Perception and Performance, 22, 1416 –1433.
Jordan, T. R., & de Bruijn, O. (1993). Word superiority over isolated
letters: The neglected role of flanking mask-contours. Journal of Experimental Psychology: Human Perception and Performance, 19, 549 –563.
Jordan, T. R., & Martin, C. (1987). The importance of visual angle in word
recognition: A “shrinking screen” modification for visual displays. Behavior Research Methods, Instruments, & Computers, 19, 307–310.
Jordan, T. R., Patching, G. R., & Milner, A. D. (1998). Central fixations are
inadequately controlled by instructions alone: Implications for studying
cerebral asymmetry. Quarterly Journal of Experimental Psychology:
Human Experimental Psychology, 51(A), 371–391.
Jordan, T. R., Patching, G. R., & Milner, A. D. (2000). Lateralized word
recognition: Assessing the role of hemispheric specialization, modes of
lexical access and perceptual asymmetry. Journal of Experimental Psychology: Human Perception and Performance, 26, 1192–1208.
Jordan, T. R., Smith, L. S., & Phillips, W. A. (1995). Exploring the
structure of the word recognition module: Psychological and computational investigations. Language and Cognitive Processes, 10, 393– 400.
Jordan, T. R., Thomas, S. M., & Scott-Brown, K. C. (1999). The illusory
letters phenomenon: An illustration of graphemic restoration in visual
word recognition. Perception, 28, 1413–1416.
Keefe, D. E., & Neely, J. H. (1990). Semantic priming in the pronunciation
task: The role of prospective prime-generated expectancies. Memory &
Cognition, 18, 289 –298.
Massaro, D. W. (1979). Letter information and orthographic context in
word perception. Journal of Experimental Psychology: Human Perception and Performance, 5, 595– 609.
Masson, M. E. J. (1986). Comprehension of rapidly presented sentences:
The mind is quicker than the eye. Journal of Memory and Language, 25,
588 – 604.
McClelland, J. L. (1987). The case for interactionism in language processing. In M. Coltheart (Ed.), Attention and performance XII: The psychology of reading (pp. 3–36). Hillsdale, NJ: Erlbaum.
McClelland, J. L., & O’Regan, J. K. (1981). Expectations increase the
benefit derived from parafoveal visual information in reading words
aloud. Journal of Experimental Psychology: Human Perception and
Performance, 7, 634 – 644.
McConkie, G. W., & Rayner, K. (1976). Identifying the span of the
effective stimulus in reading: Literature review and theories of reading.
In H. Singer & R. B. Ruddell (Eds.), Theoretical models and processes
of reading (pp. 137–162). Newark, DE: International Reading Association.
McCusker, L. X., Gough, P. B., & Bias, R. G. (1981). Word recognition
inside and out and outside in. Journal of Experimental Psychology:
Human Perception and Performance, 7, 538 –551.
Merikle, P. M. (1974). Selective backward masking with an unpredictable
mask. Journal of Experimental Psychology, 103, 589 –591.
Merikle, P. M., & Coltheart, M. (1972). Selective forward masking. Canadian Journal of Psychology, 26, 296 –302.
Merikle, P. M., Coltheart, M., & Lowe, D. G. (1971). On the selective
effects of a pattern masking stimulus. Canadian Journal of Psychology,
25, 264 –279.
Merikle, P. M., & Glick, M. J. (1976). Processing order in visual perception. Quarterly Journal of Experimental Psychology, 28, 17–26.
SENTENCE CONTEXT AND WORD RECOGNITION
Mewhort, D. J. K., & Campbell, A. J. (1978). Processing spatial information and the selective-masking effect. Perception & Psychophysics, 24,
93–101.
Morris, R. K. (1994). Lexical and message-level sentence context effects
on fixation time in reading. Journal of Experimental Psychology: Learning, Memory, and Cognition, 20, 92–103.
Morton J. (1969). The interaction of information in word recognition.
Psychological Review, 76, 165–178.
Norris, D. (1986). Word recognition: Context effects without priming.
Cognition, 22, 93–136.
O’Seaghdha, P. G. (1989). The dependence of lexical relatedness effects on
syntactic connectedness. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 73– 87.
O’Seaghdha, P. G. (1997). Cojoint and dissociable effects of syntactic and
semantic context. Journal of Experimental Psychology: Learning, Memory, and Cognition, 23, 807– 828.
Potter, M. C., Moryadas, A., Abrams, I., & Noel, A. (1993). Word
perception and misperception in context. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 3–22.
Potter, M. C., Stiefbold, D., & Moryadas, A. (1998). Word selection in
reading sentences: Preceding versus following contexts. Journal of Experimental Psychology: Learning, Memory, and Cognition, 24, 68 –100.
Prinzmetal, W. (1992). The word-superiority effect does not require a
T-scope. Perception & Psychophysics, 51, 473– 484.
Prinzmetal, W., & Silvers, B. (1994). The word without the tachistoscope.
Perception & Psychophysics, 55, 296 –312.
Ratcliff, J. E. (1987). The plausibility effect: Lexical priming or sentential
processing? Memory & Cognition, 15, 482– 496.
Ratcliff, R., Hockley, W., & McKoon, G. (1985). Components of activation: Repetition and priming effects in lexical decision and recognition.
Journal of Experimental Psychology: General, 114, 435– 450.
Ratcliff, R., & McKoon, G. (1988). A retrieval theory of priming in
memory. Psychological Review, 95, 385– 408.
Rayner, K., & Well, A. D. (1996). Effects of contextual constraint on eye
movements in reading: A further examination. Psychonomic Bulletin &
Review, 3, 504 –509.
Reicher, G. M. (1969). Perceptual recognition as a function of meaningfulness of stimulus material. Journal of Experimental Psychology, 81,
275–280.
Rumelhart, D. E., & McClelland, J. L. (1982). An interactive activation
model of context effects in letter perception: II. The contextual enhancement effect and some tests and extensions of the model. Psychological
Review, 89, 60 –94.
Rumelhart, D. E., & Siple, P. (1974). The processes of recognizing tachistoscopically presented words. Psychological Review, 81, 99 –118.
45
Seidenberg, M. S. (1985). The time course of information activation and
utilization in visual word recognition. In D. Besner, T. G. Walker, &
E. M. MacKinnon (Eds.), Reading research: Advances in theory and
practice (Vol. 5, pp. 199 –252). New York: Academic Press.
Seidenberg, M. S., Waters, G. S., Sanders, M., & Langer, P. (1984). Preand postlexical loci of contextual effects on word recognition. Memory
& Cognition, 12, 315–328.
Sharkey, A. J. C., & Sharkey, N. E. (1989). Lexical processing and the
mechanism of context effects in text comprehension. In Proceedings of
the 11th Annual Meeting of the Cognitive Science Society (pp. 466 –
473). Hillsdale, NJ: Erlbaum.
Sharkey, A. J. C., & Sharkey, N. E. (1992). Weak contextual constraints in
text and word priming. Journal of Memory and Language, 31, 543–572.
Sharkey, N. E., & Mitchell, D. C. (1985). Word recognition in a functional
context: The use of scripts in reading. Journal of Memory and Language,
24, 253–270.
Sharkey, N. E., & Sharkey, A. J. C. (1987). What is the point of integration? The loci of knowledge-based facilitation in sentence processing.
Journal of Memory and Language, 26, 255–276.
Simpson, G. B., Peterson, R. R., Casteel, M. A., & Burgess, C. (1989).
Lexical and sentence context effects in word recognition. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 15,
88 –97.
Stanovich, K., & West, R. F. (1983). On priming by a sentence context.
Journal of Experimental Psychology: General, 112, 1–36.
Stolz, J. A., & Besner, D. (1998). Levels of representation in visual word
recognition: A dissociation between morphological and semantic processing. Journal of Experimental Psychology: Human Perception and
Performance, 24, 1642–1655.
Thompson, M. C., & Massaro, D. W. (1973). Visual information and
redundancy in reading. Journal of Experimental Psychology, 98, 49 –54.
West, R. F., & Stanovich, K. W. (1982). Sources of inhibition in experiments of the effect of sentence context on word recognition. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 5, 385–
399.
Wheeler, D. D. (1970). Processes in word recognition. Cognitive Psychology, 1, 59 – 85.
Williams, J. N. (1988). Constraints upon semantic activation during sentence comprehension. Language and Cognitive Processes, 3, 165–206.
Received October 6, 2000
Revision received May 5, 2001
Accepted May 16, 2001 䡲