Copyright 2000 Psychonomic Society, Inc. 766
Memory & Cognition
2000, 28 (5), 766-773
In 1976, Schvaneveldt, Meyer, and Becker demonstrated
that the effectiveness of a homograph as a related prime
depended on the context that preceded the homograph.
When a homograph was preceded by a word that was re-
lated to the same meaning as was the subsequent target
(e.g., SAVE–BANK–MONEY), responses were faster than re-
sponses to homographs preceded by an unrelated word
(e.g., DAY– BANK–MONEY). These in turn were faster than
responses to triples containing a prime related to the other
meaning of the homograph (e.g., RIVER–BANK–MONEY).
Subsequently, Onifer and Swinney (1981) reported that
homographs embedded in auditory disambiguating sen-
tences primed lexical decisions to visually presented words
related to either of the homograph’s meanings. However,
if the target was delayed for 1.5 sec, only the context-
appropriate meaning of the homograph was primed. Sev-
eral theorists have proposed that the selection of one
meaning of a homograph leads to the immediate active
suppression of alternative meanings (Gernsbacher, 1990;
Krueger (1990) found that processing the second of two
sentences containing the same homograph in alternate
meaning contexts was slower than processing following
a control sentence. Simpson and Kang (1994) determined
that a homograph that had primed the naming of a word
related to one of its meanings actually slowed performance
on a subsequent word related to an alternate meaning. On
the basis of their extensive review of homograph research,
they concluded, “Processing one meaning of a homograph
and responding to that meaning results in the active and
specific inhibition of competing meanings” (Simpson &
Kang, 1994, p. 376).
In the present investigation, we tested whether the in-
hibition of competing meanings observed by Simpson
and his colleagues in reaction time studies extended to
the discrete response selection required by the word as-
sociation test (WAT). The WAT enabled us to examine
whether priming a homograph in the direction of one of
its meanings makes associates related to its alternative
meaning less available. Using the WAT allowed us to eval-
uate the residual effect of responding to a homograph in
two different contexts on responding to that homograph
in a subsequent neutral context.
Early studies of priming in the WAT (Clifton, 1966;
Cofer, 1967; Cramer, 1968; Segal, 1967; Segal & Cofer,
1960; Storms, 1958) defined priming on the basis of a
specific preselected associate of the word association
stimulus. With the exception of Cramer, the preselected
word was presented in an orienting task. An increase in re-
sponse frequency of that word indicated priming. Segal
reported that the likelihood that homographs initially em-
bedded in an analogy test were produced on a subsequent
WAT depended on whether the association stimulus was
We thank Christian DeBiasi, Susan DeGuilio, and Sara Jamshidi,
who served as experimenters. Three reviewers of an earlier version of
this manuscript, Doug Nelson, Dani McKinney, and Greg Simpson,
provided thoughtful ideas that resulted in clarification of the ideas pre-
sented. We are grateful to them, as well as to Jim Erickson, for their crit-
ical reviews. Correspondence should be addressed to D.S. Gorfein, De-
partment of Psychology, University of Texas, Box 19528, Arlington,
TX 76019-0528 (e-mail: email@example.com).
The selection of homograph meaning:
Word association when context changes
DAVID S. GORFEIN
University of Texas, Arlington, Texas
College of Mount St. Vincent, Riverdale, New York
St. Peter’s College, Jersey City, New Jersey
In a study of lexical ambiguity processing, responses to homographs were examined in a word as-
sociation task. The context of repeated exposures of a homograph was manipulated by requiring a re-
sponse to a word related to a meaning of the homograph on the trial prior to homograph presentation.
A change of that relationship reduced the effectiveness of the contextual item as a prime on the sec-
ond occurrence of the homograph. In response to a third unprimed occurrence of the homograph, as-
sociations were consistent with a conclusion that when semantic contexts are opposed, a “primacy ef-
fect” is obtained. The overall effects in the studies reported are seen as consistent with the theoretical
view of Simpson and Kang (1994) that processing and responding to one meaning of a homograph re-
sult in the inhibition of alternative meanings. A mechanism to account for that inhibition is proposed.
WORD ASSOCIATION WHEN CONTEXT CHANGES767
identical to, related to, or different from the meaning pre-
sented in the analogy. The word association stimulus RIVER
elicited the response BANK with greatest frequency if the
analogy had been “BANK is to RIVER,” a lesser frequency
if the analogy had been “BANK is to STREAM,” and at near
chance levels if the analogy had been “BANKis to MONEY.”
Cramer (1968) presented homographs in the context
of two words selected to guide the production of a par-
ticular response. Participants were told to read aloud the
first two words on a trial (e.g., ANGER, FIGHT) and then
produce a word association to a third word (TAP). Perfor-
mancewas scored for the response HIT. In another group,
the word TAP was preceded by WHISKEY and ALE and the
target word was BEER. Both contexts elicited their re-
sponses at greater than normative frequency.
In our studies priming is defined as an increase in the
frequency of WAT responses that are related to a partic-
ular meaning of a homograph (e.g., responses related to
the financial meaning of BANK). Cramer (1968) analyzed
the increase in responses related to the primed meaning
of the homograph (e.g., beverage-related responses to
the stimulus TAP given the primes WHISKEY and ALE). She
concludedthat beyond an increase in the specific response
BEER, the prime had little influence. However, Cramer’s
manipulation may have led participants to treat the task
as a problem-solving task and use the 6sec provided to re-
spond to find a common response word to WHISKEY, ALE,
Berger, Bubka, and Gorfein (1990) reported an increase
in responses from the primed meaning class of a homo-
graph. Words from a meaning class were more likely to
be elicited to a homograph when the previous trial re-
quired a response to a related word than when the prior
response was to an unrelated word. Responding to LIVER
on a WAT trial increased the frequency of responses such
as HEART to the homograph ORGAN on the next trial.
The three experiments in this paper used a similar
method of presenting homographs and primes in a WAT.
The participant viewed a series of words and was in-
structed to write the first word to come to mind on each
trial. An advantage of this method is that prime trials are
not distinguished from homograph trials. This is in marked
contrast to priming in the naming task and to typical sin-
gle-trial priming events in lexical decision. It also differs
from the priming technique employed in the WAT by
Cramer (1968) in which the participant only read the
primes and responded to the homograph.
In our experiments, each homograph initially followed
a word related to one meaning and on a second occurrence
was preceded by a word related to its alternative meaning.
For example, ORGAN was presented on trial n preceded by
the related word LIVER on trial n ? 1. The word MUSIC
occurred on trial n + 23 and the homograph ORGAN again
on trial n + 24. This allowed us to test the hypothesis that
the first occurrence of the homograph would result in a
diminished effect of the second-occurrence prime as con-
trasted with the effect of a first-occurrence prime. We
were also interested in whether the effect of earlier prim-
ing trials would carry over to trials where the homograph
was presented in a neutral context. Continuing the pre-
vious example, we examined which response class would
be elicited when ORGAN was presented on trial n + 64 fol-
lowing an unrelated stimulus (HAT) on trial n + 63. We
also examined whether the effect on the neutral primed
trials depended on the time between the homograph oc-
currences. Experiment 1 tested these conditions.
Thirty-two Adelphi University students participated in Experi-
ment 1 as part of a requirement for an introductory psychology
course. The participants in all experiments were native English
speakers. Participants were assigned to list at random.
Twenty-eight homographs were selected from the Adelphi norms
(Gorfein, Viviani, & Leddo, 1982). Items were selected so that the
two meanings of the homograph were reasonably balanced. The
mean proportion of occurrences for the more common (dominant)
meaning of the 28 homographs was .605, with a range of .50–.72.
We used homographs that had two distinct meanings, when all senses
of a meaning were scored as the same meaning. Therefore the sec-
ondary meaning of the homograph had a normative proportion of
.395. The mean frequency of occurrence per million of the homo-
graphs was 48.25 (Kuˇ cera & Francis, 1967).
Two primes were chosen for each homograph, one related to its
dominant meaning and the other to its secondary meaning. The
choice of prime word was somewhat arbitrary since relatedness
norms for the homographs did not exist at the time of list creation.
It was our aim to select words that while related to the homographs
chosen would be unlikely direct associates of the homograph, to
minimize what Cofer (1967) called “direct priming.” Across par-
ticipants the same prime preceded a first and a second occurrence
of a homograph equally often. We therefore compared the amount
of priming obtained from the same prime when it preceded a first
occurrence with priming when it preceded a second homograph oc-
currence. We look at a homograph MARCH following the prime DE-
CEMBERwhen the participant has earlier responded to ARMYpreced-
ing MARCH and in the absence of the earlier experience. Because of
the counterbalancing, the actual size of the relationship between any
prime and homograph is largely irrelevant to the questions asked.
Materials were divided into four subsets matched on meaning
dominance and were arranged into four unique word lists so that each
homograph appeared in each condition of the experiment once. Two
context conditions (mixed and single) were included in each list. In
the mixed condition, 14 homographs were presented three times.
These homographs were preceded by a prime related to one meaning
on their first occurrence, by a prime related to the alternate mean-
ing on second occurrence, and by a word unrelated to either meaning
on third occurrence. In the single-prime condition, the remaining 14
homographs were presented twice, preceded by a related prime on
the first occurrence and by a word unrelated to either meaning on
the second occurrence. On first occurrence, half (7) of the homo-
graphs in each condition were preceded by a word related to their
dominant meaning while the remaining homographs were preceded
by a word related to their secondary meaning.
Each list was constructed in three parts. Part 1 (Trials 1–84) of
each list included the first occurrence of the 14 homographs and their
primes in the mixed-prime condition and the 14 homographs and
768GORFEIN, BERGER, AND BUBKA
primes in the single-prime condition, as well as 28 filler items. Filler
words were unrelated to the homographs. A set of 12 filler items
was also included between Parts 1 and 2 to ensure a minimum sep-
aration of first and second occurrences of homographs. Actual sep-
aration varied from 24 to 142 trials with a mean of 82 events.
Part 2 (Trials 97–166) included the second occurrences of the 14
mixed-prime homographs preceded by their alternate meaning prime.
Homographs in the single-prime condition were replaced by unre-
lated words, which were preceded by the alternative meaning prime
of each replaced homograph. Part 2 included 28 filler trials, with
half of the fillers being repetitions of Part 1 filler words.
Part 3 followed Part 2 immediately or was delayed by 40 filler
trials. Delay was manipulated between lists and resulted in Part 3
trials of either 167–206 or 207–262. The 56 trials of Part3 included
the third occurrences of the 14mixed-prime homographs, 14second
occurrences of the single-prime homographs, and a third occurrence
of 14 fillers, as well as a second occurrence of another 14 fillers.
Counterbalancing homographs across the meaning primed on
first homograph occurrence (dominant or secondary), prime context
conditions (mixed or single), and delay (0 or 40 trials) resulted in
eight unique lists that were employed equally often in the experiment.
Participants were tested individually on the WAT. Single words
were presented centered on a computer screen, along with their trial
numbers. Participants were instructed to write the first word each
item brought to mind alongside the trial number on the answer sheet
provided. No mention was made of primes or homographs. From
the participant’s standpoint, each word was a unique item in the WAT.
After responding, the participant pressed the space bar and the next
item appeared on the screen. If no response was made and 6 sec
elapsed, the next trial was presented. Participants were informed that
some items would be repeated and were instructed that they should
always write the first word that came to mind. The task occupied
Results and Discussion
A rater blind to the prime assigned the response to each
homograph to a meaning class. When a response could
not be readily assigned to one of the two meanings of a
homograph, a second rater attempted to classify it. Omis-
sions and responses that could not be scored were dropped
from the analyses (2.1% of all responses). The reliability
of scoring was measured by having two raters score 28
homograph trials from the Part 3 data of a random sam-
ple of 10 participants. Of the 280 total judgments of re-
sponses, 278 were in agreement (99.3%).
The score in each condition was the proportion of scor-
able responses that were judged to be related to the dom-
inant meaning of the homograph. Priming was defined
as the difference between the chance value (.605) and the
observed proportion of each condition. For the dominant
prime, a priming score was the proportion of responses
in the dominant direction1minus .605. Priming for sec-
ondary primes was defined as .605 minus the proportion
of dominant responses. The alpha level for all analyses
was set at .05 in all experiments.
Table 1 reports the proportion of dominant meaning
responses to the homographs as a function of the prime
history in each part of the experiment as well as a prim-
ing score for each condition. Each of the eight priming
scores was significantly greater than zero priming by a
two-tailed t test. This indicated that our method, which
defined priming in the WAT in relation to the meaning
class of the response, produced results similar to those of
earlier studies that defined priming with respect to spe-
cific response targets (Cramer, 1968; Segal, 1967).
The nature of the design necessitated several analyses
of variance (ANOVAs). An ANOVA2of Part 1 indicated
that the observed proportion of dominant responses was
significantly different when the homograph was preceded
by a prime trial related to the dominant meaning (.767)
than when the prime trial item was related to the secondary
meaning [.427, F(1,30) ? 165.17, MSe? 0.0112].
All homographs in Part 2 were in the mixed-context
condition and so were primed with the meaning alternate
to that primed in Part 1. The proportion of dominant re-
sponses to each of the Part2 primes differed significantly
[F(1,30) ? 18.26, MSe? 0.0424], with the proportion
of dominant responses when the prime was dominant
(.725) being greater than when the prime was secondary
(.504). The magnitude of priming in Parts 1 and 2 was
compared in an ANOVA of the priming scores (Part 1 vs.
Part 2, by current prime). The mean priming effect in
Part 1 (.170) was significantly different from the mean
priming effect in Part 2 [.121, F(1,60) ? 5.32, MSe?
0.0216]. These results indicate a reduction in the effec-
tiveness of a prime of one homograph meaning givenprior
exposure to an alternate meaning prime. This finding is
consistent with the loss of prime effectiveness in word
naming (Simpson & Kang, 1994; Simpson & Kellas,
1989). The implications of these results are addressed in
the General Discussion.
Part 3 tested how the prime history of a homograph
influenced subsequent responses to that homograph in a
neutral context. Priming in Part 3 was defined as the pro-
portion of responses that were related to the meaning
Proportion of Dominant Meaning Word Association Responses
(Score) and Observed Priming in Each Part of Experiment 1
Part 3 Unprimed
Note—Priming is the deviation from the normative baseline of .615.
*Priming in Part 3 is defined with respect to the Part 1 prime for all
WORD ASSOCIATION WHEN CONTEXT CHANGES 769
primed in Part 1. We then compared the amount of prim-
ing in Part3 to the amount in Part1. A 2prime type (dom-
inant, secondary) ? 3 prime history (Part 1, Part 3 single,
Part 3 mixed context) ? 2 delay (0 or 40 items between
Parts 2 and 3) mixed-model ANOVA was performed.
Neither the effect of delay [F(1,30) < 1] nor any of its in-
teractions approached significance (F < 1, in all cases).
The effect of prime history was significant [F(2,60) ?
11.10, MSe? 0.0279]. A post hoc analysis employing
Tukey’s procedure for pairwise comparisons (Myers &
Well, 1991, pp. 84–86) indicated that the higher propor-
tion of primed responses in Part1 (.170) was significantly
different from the priming observed in Part 3 to items in
the single-prime condition (.089) or to those in the mixed
prime history condition (.053). The difference within
Phase3 was not significant. However, the amount of prim-
ing in the single-prime condition [t(31) ? 4.61] and the
mixed-prime condition [t(31) ? 2.39] was significant in
the direction of the Part 1 meaning. The effect in the
mixed condition is remarkable in that not only was there
a departure from the more recent Part 2 prime, but also
there was a significant “primacy effect.” This effect was
not unlike the classical effect of primacy in trait descrip-
tions in the social psychological literature (see Anderson,
1996). The results were also consistent with results we
have reported elsewhere for both spelling and word asso-
ciation responses to nonhomographic homophones (Gor-
fein & Walters, 1989; Walters, 1988).
Experiment 1 established the effect of the initial pro-
cessing of a homograph on selecting an alternate mean-
ing of that homograph in a new context. Further, process-
ing homographs in two different contexts prior to a test in
a neutral context resulted in a primacy effect. In the single-
prime condition, responses to homographs in a neutral
context showed a return toward the normative baseline.
In Experiment 2, we attempted to characterize these
findings more precisely. The interval between first and
second occurrences of homographs in the mixed-prime
condition was large and variable in Experiment 1. In Ex-
periment2, the separation between first and second occur-
rences of homographs was fixed at either 10 or 30 items.
A control condition was added in which previously
primedhomographs were presented in neutral contexts at
the same lags as the first two occurrences in the mixed-
prime condition. This allowed a comparison of priming in
the mixed-context condition to control items with similar
We reexamined the delay between the alternate mean-
ing context in Part 2 and the neutral context in Part 3 in
the mixed-prime condition. We hypothesized that as the
delay between the second and third occurrences increased,
the primacy effect would become stronger due to a loss
in effectiveness of the most recent prime. We saw this as
similar to the loss of recency effects in free recall over a
filled interval (see, e.g. Glanzer & Cunitz, 1966).
In Experiment 1, we manipulated type of prime (dom-
inant, secondary) as a within-subjects condition. In Ex-
periment2, the meaning of the first-occurrence prime was
manipulated as a between-groups variable; one experi-
mental group was assigned dominant primes and the
other group secondary primes.
Forty-eight Adelphi University students and 64 students at the
University of Texas at Arlington served in this experiment to fulfill
a course requirement. Participants were assigned to the four exper-
imental conditions in a counterbalanced manner. Each of the four
between-groups conditions included an equal proportion of the sam-
ples drawn from the two universities.
We created lists with the 28 homographs and the 56 prime words
from Experiment1. Four basic list types were constructed by manip-
ulating lag between the first and second occurrences of a homograph
(10 or 30 trials) as a between-lists variable and delay between the sec-
ond and third occurrences (3 or 20 trials) as a within-lists factor.
In one set all 28 homographs were primed in the dominant direc-
tion on their first occurrence and in another set of lists, items were
primed in the secondary direction. On the second occurrence, half
of the items on each list were primed by their alternate meaning
prime (mixed-context condition) and half by the unrelated alternate
prime of another homograph (unrelated-prime control condition).
Counterbalancing to ensure that every homograph appeared in each
condition resulted in 16 different lists.
Each list consisted of 222 trials. Homograph sets were interwoven
in the list. Therefore the second and third occurrences of some ho-
mographs preceded the first occurrence of others. Filler items, half of
which were repeated at varying intervals, occupied 82 trial positions.
The procedure and instructions were identical to those of Exper-
iment 1 except that a fixed 6-sec trial was employed.
The design of the study was a mixed model. Lag between first and
second occurrences (10 or 30 items) and Part 1 prime meaning
(dominant, secondary) were between-groups variables. Contextual
history (mixed or control) and delay between second and third oc-
currences of a homograph (3 or 20 trials) were within-groups vari-
ables. The task required 23 min to complete.
Results and Discussion
The proportion of scorable responses in the direction
of the normative dominant meaning of the homograph
was used for all conditions.
We collapsed the third-occurrence data across the delay
intervals between second and third occurrences and re-
served this delay variable for a separate ANOVA. A 3 (1st,
2nd, or 3rd occurrence) ? 2 (dominant or secondary prime
on first occurrence) ? 2 (prime history, mixed or con-
trol) ? 2 (lag, 10 or 30 trials between 1st and 2nd homo-
graph presentation) mixed model ANOVA was performed.
The lag between first and second homograph occur-
rences did not produce a significant main effect and did
not enter into any interaction. Table 2 presents the pro-
portion of responses scored as dominant meaning re-
sponses to the homograph collapsed over lag. Since, in
this experiment, the primes were not equated for rela-
770 GORFEIN, BERGER, AND BUBKA
tionship to the homographs, we restrict our reporting of
results to the effectiveness of the repetition conditions.
The initial priming direction interacted both with occur-
rence [F(2,216) ? 67.20, MSe? 0.016] and with prime
history [F(1,108) ? 21.47, MSe? 0.017]. There were
no significant main effects of occurrence or prime his-
tory and no interaction between the two.
However, these effects must be interpreted in light of
the significant three-way interaction between occurrence,
prime history, and initial prime direction [F(2,216) ? 6.23,
MSe? 0.016]. In the control condition (unrelated con-
text on second occurrence), the effect of the initial prime
was reduced on the second occurrence as control item re-
sponses shifted toward the norm and then remained un-
changed on the third occurrence. In the mixed condition,
the alternate meaning was successfully primed on the
second occurrence. Further, on the third occurrence, per-
formance moved back toward the first-occurrence mean.
A second ANOVA examined the primacy effect based
on initial homograph meaning. As shown in Table 2, the
means for the mixed items moved in different directions
from the second to the third occurrence of the homo-
graph based on initial meaning primed. Means for homo-
graphs initially primed in the dominant direction moved
toward the dominant meaning, while means for homo-
graph initially primed in the secondary direction changed
in that direction. A 2 (dominant, secondary prime) ? 2
(second or third occurrence) ANOVA was performed to
see whether this observed effect was significant. The
significant interaction [F(1,108) ? 15.32, MSe? 0.026]
indicates that responses to the unprimed third occur-
rences of homographs in the mixed condition are in the
direction of the initial priming experience.
We had hypothesized that the longer the delay between
the second and third occurrences of the homograph, the
greater the change toward the initially primed meaning.
The hypothesis was not supported; the change toward the
initial prime was .06 in the Lag 3 delay condition, while
the corresponding change in the Lag20 condition was .04.
Neither the effect of the lag between first and second
occurrences nor the delay prior to testing the third oc-
currence in the mixed prime history condition had a sig-
nificant effect on performance. Nevertheless, in the control
condition the effect of the prime declined significantly
(.09) from first occurrence to the neutral primed second
occurrence. In the mixed condition, there was a similar
decline of second prime effectiveness (.05) relative to the
neutral context third occurrence. It must be concluded
that decline was relatively rapid at a lag of 3 or fewer
items. A possible explanation of the rapid change is that
a portion of the immediate priming effect depends on the
N ? 1 prime being in the active state. Like the “auto-
matic priming” of Posner and Snyder (1975), that activa-
tion may decay in a matter of seconds or be disrupted by
another word presentation. In that event, the prior prime
would be inactive before the next occurrence of a homo-
graph even at short lags.
In the mixed condition on second occurrences, .583 of
the responses were in the dominant direction despite the
presence of a secondary meaning prime. Responses to the
dominant primed condition were near baseline at .601.
These data are compatible with the “inhibition of compet-
ing meanings” hypothesis of Simpson and Kang (1994).
Alternatively, it might be argued that computing priming
from the normative baseline is misleading because the
first-occurrence prime produces a temporary shift in the
normative frequency of the homograph meanings, lead-
ing to an underestimate of the amount of priming. The
results with respect to second occurrences in Experiment2
speak directly to this point. As shown in Table 3, com-
putations of priming based on the control condition, which
reflect any change in baseline, do not support this expla-
nation. The amount of priming in relation to the appro-
priate control was .09. This value was slightly more than
half of the priming obtained (.17) for the first-occurrence
prime. Some form of inhibition of alternative meanings
results from the processing of a homograph within a spe-
cific semantic context.
Proportion of Dominant Meaning Word Association Responses
(Score) and Observed Priming in Each Part of Experiment 2
Part 3 Unprimed
Note—Priming is the deviation from the normative baseline of .615.
*Priming in Part 3 is defined with respect to the Part 1 prime for all
Effect of Context on Second-Occurrence Responses to
Homographs (Score): Change in the Direction of
Second-Occurrence Prime Event
First Second Mixed