1_98� 1984, 2, 256-281

Document technical information

Format pdf
Size 1.0 MB
First found Jun 9, 2017

Document content analysis

Category Also themed
not defined
no text concepts found





2, 256-281
Cop) right
Journal of Ex�rimental Psychology: General
Vol. 113, No.
1_98� by tht
American Psycho1og�ca1 Assoc1at1on, Inc.
Resolving 20 Years of Inconsistent Interactions Between Lexical
Familiarity and Orthography, Concreteness, and Polysemy
Morton Ann Gernsbacher
University of Texas at Austin
Numerous word recognition studies conducted over the past 2 decades are examined.
These studies manipulated lexical familiarity by presenting words of high versus
low printed frequency and most reported an interaction between printed frequency
and one of several second variables, namely, orthographic regularity, semantic
concreteness, or polysemy. However, the direction of these interactions was incon­
sistent from study to study. Six new experiments clarify these discordant results.
The first two demonstrate that words of the same low printed frequency are not
always equally familiar to subjects. Instead, subjects' ratings of "experiential fa­
miliarity" suggest that many of the low-printed-frequency words used in prior
studies varied along this dimension. Four lexical decision experiments reexamine
the prior findings by orthogonally manipulating lexical familiarity, as assessed by
experiential familiarity ratings, with bigram frequency, semantic concreteness, and
number of meanings. The results suggest that of these variables, only experiential
familiarity reliably affects word recognition latencies. This in tum suggests that
previous inconsistent findings are due to confounding experiential familiarity with
a second variable.
Twenty years of research on word recog­
nition has repeatedly shown that the famil­
iarity of a word greatly affects both the speed
and the accuracy of its recognition. More fa­
miliar words can be recognized faster and more
accurately than less familiar words. Tradi­
tionally, lexical familiarity has been opera­
tionalized as the frequency with which a word
occurs in printed English text. Experimenters
typically construct their stimulus sets by con­
sulting one of three widely used indices:
Thorndike and Lorge's ( 1944) Teacher's Word
Book of 30,000 Words, Kucera and Francis's
( 1967) Computational Analysis ofPresent-Day
American English, or Carroll, Davies, and
Richman's ( 1971) American Heritage Word
Frequency Book. Within these corpora, one
This research was conducted while the author was sup­
ported by Predoctoral Training Grant MH-15744 from
the National Institute of Mental Health.
This article benefited from the insightful critiques pro­
vided by James C. Johnston, Thomas K. Landauer, and
Donald L. Scarborough. Donald E. Broadbent, John B.
Carroll , and Kenneth I . Forster provided feedback on an
earlier draft. I thank Donald J. Foss for first encouraging
me to submit this article and especially Arnold H. Buss
for helping me eliminate needless verbosities.
Requests for reprints should be sent to Morton Ann
Gemsbacher, who is now at the Department of Psychology,
University of Oregon, Eugene, Oregon 97403- 1 227.
would find that the English word amount oc­
curs relatively frequently (with an average fre­
quency score of 110 occurrences per million
words of text), whereas the word amour occurs
relatively infrequently (with an average fre­
quency score of I occurrence per million words
of text).
The Effect of Printed Frequency
Howes and Solomon ( 19 51) reported that
printed frequency could account for approx­
imately half of the variance found in tachis­
toscopic thresholds. Similarly, Rubenstein,
Garfield, and Millikan ( 1970) reported that,
on the average, lexical decision latency to a
high-printed-frequency word is significantly
shorter than that to a low-printed-frequency
word, such that words that differ in printed
frequency by a factor of 10 usually show a
75-ms difference in response latency. A less
conservative estimate has been given by Scar­
borough, Cortese, and Scarborough ( 1977): A
50-ms difference in response time occurs be­
tween words that differ by one logarithm ic
unit of printed frequency. According to an
average of these estimates, the word amount
should be recognized a little more than 100
ms faster than the word amour.
Despite wide evidence for printed frequen­
cy's potency in predicting both speed and ac­
curacy in word recognition, there is little
agreement about the mechanism underlying
its robust effect. There appear to be two broad
classes of theories. One theoretical camp sup­
ported the proposition that the effect of printed
frequency was perceptual; in simplistic terms,
high-printed-frequency words elicit superior
recognition performance because they are
more easily seen (e.g., Catlin, 1969; Newbig­
ging, 1961; Rumelhart & Siple, 1974; Savin,
1963; Solomon & Postman, 1952). Their op­
ponents argued that the effect of printed fre­
quency derived from response processes: High­
printed-frequency words can evoke responses
more rapidly (e.g., Adams, 1979; Broadbent,
1967; Morton, 1968; Treisman, 1971).
These theories were based on the implicit
assumption that high- and low-printed-fre­
quency words are equivalent along all other
relevant dimensions. But Landauer and Stree­
ter (1973) disconfirmed this assumption. They
demonstrated that the distribution of letters
and phonemes differs significantly in high- and
low-printed-frequency words. That is, high­
printed-frequency words are likely to contain
more regularly occurring phonemic and gra­
phemic patterns than low-printed-frequency
words. Landauer and Streeter's work sup­
ported Carroll and White's caveat: "Word fre­
quency may not be the simple variable that it
appears to be" ( 1973, p. 563).
To be sure, other variables do covary with
printed frequency, and the effect of printed
frequency may be partially attributable to these
secondary variables. Besides differing in or­
thographic and phonemic structure, high­
printed-frequency words also differ from low­
printed-frequency words along semantic and
lexicographic dimensions. Paivio, Yuille, and
Madigan ( 1968) noted that a greater propor­
tion of high-printed-frequency words are con­
crete or imageable rather than abstract,
whereas the reverse is true of low-printed-fre­
quency words. Furthermore, high-printed-fre­
quency words tend to have more individual
meanings (Glanzer & Bowles, 1976; Reder,
Anderson, & Bjork, 1974; Schnorr & Atkin­
son, 1970).
In the last 20 years, many researchers have
orthogonally manipulated the printed-fre­
quency variable with these other variables in
the hope of discovering the nature of the
printed-frequency effect. With few exceptions,
high-printed-frequency words were recognized
with a consistently high level of accuracy or
speed, regardless of their orthographic regu­
larity, semantic concreteness, or number of
meanings. Performance with low-printed-fre­
quency words has not been so consistent.
Rather, recognition of low-printed-frequency
words has often interacted with the above three
variables in paradoxical and inconsistent ways.
The Inconsistent Interaction Between
Printed Frequency and Bigram Frequency
Just as English words differ in frequency of
occurrence, so the components of those words,
individual letters and letter patterns, differ in
frequency of occurrence (Shannon, 1948). One
measure of orthographic frequency is bigram
frequency, that is, the frequency of two letters
occurring in tandem in a particular position
of a particular length word. As an illustration,
the bigram WH frequently occurs as the first
bigram of a five-letter word, but never as the
last bigram of a five-letter word.
Orsowitz ( 1963, cited in Biederman, 1966)
factorially combined printed frequency with
bigram frequency. Subjects were tachistoscop­
ically presented with five-letter words, and the
number of trials to accurately recognize each
stimulus word was recorded. Orsowitz found
that the effects of printed frequency and bi­
gram frequency were not additive but inter­
active and that the interaction was somewhat
paradoxical. For high-printed-frequency
words, bigram frequency had no effect, but
for low-printed-frequency words, more trials
were required to recognize words with high­
frequency bigram (high-bigram words) than
words with low-frequency bigrams (low-bi­
gram words). This result was corroborated by
Broadbent and Gregory ( 1968). Rice and
Robinson ( 1975) also corroborated the Or­
sowitz results, using a lexical decision para­
digm: Subjects were required to decide quickly
whether letter strings composed a word. The
mean reaction time ( RT) and percentage of
errors revealed that for high-printed-frequency
words, bigram frequency had no effect, but
responses to low-printed-frequencyjhigh-bi­
gram words were slower and less accurate than
those to low-printed-frequencyjlow-bigram
Table I
Resu/cs of Sl!ldies Thai Have Examined che Ejfecls of Prinled Frequency and Bigram Frequency
and Resu/cs of Experimenl 2
Orignal results
Original study
Broadbent & Gregory (1968)
HBF worse than LBF
Rice & Robinson (1975)
HBF worse than LBF
HBF better than LBF
Biederman ( 1966, Experiment 2)
Biederman (1966, Experiment I)'
Orsowitz (I 963, cited in Biederman, 1966)'
HBF better than LBF
Rumelhart & Siple (1974)•
HBF worse than LBF
HBF better than LBF
McClelland & Johnston (1977)•
HBF same as LBF
Noce. Data are for low-printed-frequency words only. HBF
Results of Experiment 2
HBF less familiar than LBF
HBF less familiar than LBF
HBF more familiar than LBF
HBF equally familiar as LBF
HBF equally familiar as LBF
high bigram frequency words, LBF
low bigram
frequency words.
The same stimulus words were used in these two original experiments. • Not examined in Experiment 2 because
their stimuli were not available.
Biederman ( 1966) tachistoscopically pre­
sented subjects with Orsowitz's five-letter
words and measured temporal threshold for
accurate identification, but found opposite re­
guessing works against them when recognizing
low-printed-frequency words composed of
high-frequency bigrams. That is, with low­
printed-frequency words, if the orthography
sults. Indeed, Biederman found the usual main
effect of printed frequency, but conversely
found that low-printed-frequency words con­
taining high-frequency bigrams were recog­
nized in fewer trials than those composed of
low-frequency bigrams. In a second experi­
ment, using only low-printed-frequency words,
Biederman again found an advantage for a
high-bigram frequency in recognizing low­
printed-frequency words. Rumelhart and Siple
(1974) reported the same interaction as Bie­
derman (1966, Experiment I). Adding further
resembles a high-frequency word (i.e., the word
is composed of high-frequency big rams), sub­
jects will be likely to guess a high-frequency
word, and of course, be incorrect. Sophisti­
cated guessing is believed to be even more
attractive when the low-printed-frequency
words are from a very low range of printed
frequency (cf. Rumelhart & Siple, 1974 ), cr
are preceded cr followed by a visual mask (cf.
McClelland & Johnston, 1977; McClelland &
Rumelhart, 198 I).
Rice and Robinson (1975) conceded that a
to the puzzle, McClelland and Johnston (1977)
reported no interaction. The results of these
studies are summarized in Table I.
Though contradictory, the results of the
sophisticated guessing strategy could also be
operating in their lexical decision task, though
their data suggest that sophisticated guessing
cannot fully account for the performance they
observed. The R Ts from their study revealed
the typical paradoxical interaction between
bigram frequency and printed frequency, yet
they found no effect of bigram frequency oo
their subjects' perf ormance with nonword
Biederman (1966), Rumelhart and Siple
(1974), and McClelland and Johnston ( 1977)
studies are straightf orward. The most puzzling
finding is the paradoxical interaction reported
by Orsowitz (1963, cited in Biederman, 1966),
Broadbent and Gregory (1968), and Rice and
Robinson (1975 ). It does not seem reasonable
that the greater the frequency of a word's hi­
grams, the worse its recognition will be.
However, an explanation has been offered:
Subjects are "sophisticated guessers" (cf.
Broadbent, 1967; Neisser, 1967; Newbigging,
stimuli. If the paradoxical disadvantage of high
bigram frequency in low-printed-frequency
words is caused by subjects' sophisticated
guessing, surely one would predict longer la­
tencies or more errors for nonwords composed
of high-frequency bigrams because they are
more apt to resemble real words.
1961; Solomon & Postman, 1952). When rec­
To summarize, the studies reviewed here
ognizing tachistoscopically presented words,
subjects are likely to guess at a partially rec­
have factorially manipulated printed frequency
and bigram frequency, but their results ha>e
been inconsistent. All studies reported that
high-printed-f requency words were recognized
significantly better than low-printed-frequency
words. Almost all reported an interaction be­
tween printed frequency and bigram frequency
such that with high-printed-frequency words,
there was no effect of bigram frequency. Or­
thographic regularity influenced the recogni­
tion of low-printed-frequency words but with­
out a consistent pattern. In some studies, high
bigram frequency facilitated the recognition
of low-printed-frequency words; in others it
led to poorer performance.
The favored explanation for the paradoxical
interaction or its absence has been sophisti­
cated guessing. Subjects dealing with inade­
quate visual information or under time pres­
sure are more likely to incorrectly report or
to delay responding to low-printed-frequency
words composed of letter patterns that occur
frequently. The purpose of Experiment I was
to test this explanation. If the paradoxical in­
teraction is caused by a sophisticated guessing
strategy, and this strategy is induced by pro­
cessing incomplete inf ormation due to brief
exposure or speeded decision making, remov­
ing these inducements should eliminate the
paradoxical interaction. There would be n o
need for sophisticated guessing i f t h e stimuli
are available for as long as subjects wish and
the responses are not time pressured. Thus,
subjects in Experiment I were presented with
the stimulus words used in the Rice and Rob­
inson ( 1975) study and were asked to give an
page. The words, typed in capitals, appeared down the
left-hand margin. Opposite each word was a 7-point nu­
merical scale, v.ith its ends labeled HIGHLY CONFIDENT IS
order of the five pages was randomized for each set, and
the pages were collated into a booklet that included a cover
sheet with written instructions and a space for name, session
number, and date.
Procedure. Subjects were asked to rate their confidence
concerning the lexical status of letter strings. Specific in­
structions were read silently by each subject while the
experimenter read them aloud at the beginning of the
experimental session. These instructions encouraged sub­
jects to work at their own rate and to "please take as much
time to make each decision as needed."
Results and Discussion
Mean ratings were computed for each item
by averaging across all subjects' responses to
a given item. A 2 X 2 (Printed Frequency X
Bigram Frequency) analysis of variance (AN­
OVA ) was perf ormed on the ratings f or the word
stimuli. This analysis revealed a significant
main elf ect of printed frequency, F(l, 56)=
54.00, p < .00 I, a main effect of bigram fre­
quency, F(!, 56)= 6.41, p < .01, and a sig­
nificant interaction between the two variables,
F( l , 56)= 5.40, p < .02. Figure I compares
the mean lexical confidence ratings for the four
word conditions with the mean RT obtained
to these same items by Rice and Robinson
(1975). Bigram frequency affected lexical con­
fidence only for the low-printed-frequency
words. For high-printed-frequency words, the
mean lexical confidence rating for words with
unspeeded judgment of their confidence con­
high-frequency bigrams (M
cerning the lexical status of each word.
fer significantly from the ratings for words with
low-frequency bigrams (M
Experiment I
Subjects were 45 native English speakers at
the University of Texas at Austin who were enrolled in an
introductory psychology course and who participated in
the experiment to fulfill a course requirement.
6.6) did not dif­
6.6), t(28) 0.33,
.70. For low-printed-frequency words,
subjects were less confident that Rice and
The materials were the 60 words and 60
nonwords used by Rice and Robinson ( 1975). Half of the
60 real words occurred frequently in printed material; half
OCcurred infrequently. Half of each frequency set contained
high-frequency bigrams, the other half, low-frequency bi­
&rams. In addition, half of the non words contained high­
frequency bigrams, and the other half, low-frequency hi­
The 120 words and nonwords were randomly arranged
and typed on live pages, 24 words to a page, with the
constraints that no more than 2 words or nonwords ap­
Deared consecutively and that an equal number of items
from each of the original six conditions appeared on a
Robinson's high-bigram words (M= 5.6) were
real words than that their low-bigram words
6.1) were, t(28)
2.56, p < .02. This
pattern mirrored the latency data reported by
Rice and Robinson. Percentage response to
the top of the confidence scale reveals these
effects more dramatically. To the high-printed­
frequency /high-bigram words, 90% of the
subjects responded HIGHLY CONFlDENT IS A
WORD, compared with 89% to the high­
printed-frequency /low-bigram words. In re­
sponse to the low-printed-f requency words,
63% of the subjects responded
to those composed of low­
frequency bigrams, compared with 45% to
5 .5
:i! 600
Figure 1. Mean reaction time from Rice and Robinson's ( 1975) study and mean lexical confidence ratings
from Experiment l for word stimuli.
on literary samples of word usage. For ex­
ample, the word comma occurs only once or
twice per million words of text, but the word
chapter occurs 50 to 100 times. It is doubtful
that chapter is 50 to 100 times more familiar
than comma. Consider also the changes in
contemporary English usage since printed fre­
quency counts were first assembled. Only a
few years after the Thorndike and Lorge ( 1944)
count was published, Howes (1954) ques­
tioned, "to what extent can word frequencies
based on the linguistic behavior of writers in
the 1930's represent the average base proba­
bilities of Harvard students in 1 948?" (p. 106).
The problem must be more serious in the
1980s, yet in psycholinguistic research pub­
lished from 1970 to the present, the older
Thorndike and Lorge count was still favored
over the newer Kucera and Francis (1967)
count by approximately 2 to I (White, 1983).
(The Carroll et al., 1971, count was based on
grade school literature and is rarely used in
experiments with adult subjects.)
Another problem with counts of printed
frequency is that they are, by definition, sam­
ples and so are subject to sampling error. Low­
The Reliability of Printed Frequency
printed-frequency words are subject to the
A potential problem of counts of printed greatest sampling bias (Carroll, 1 967, 1 970),
frequency is that they are, by definition, based both in the original collection of the corpora
those composed of high-frequency bigrams,
t (28) 2.29, p < .03.
This paradoxical interaction between hi­
gram frequency and printed frequency seri­
ously challenges the sophisticated guessing ex­
planation of this result. The present subjects,
unlike those in Rice and Robinson's (1975)
experiment, performed the task without any
speed pressure. Moreover, the stimulus words
were not presented briefly, as in the Orsowitz
( 1 963, cited in Biederman, 1966) and Broad­
bent and Gregory ( 1 968) studies, nor were they
visually masked, as in the McClelland and
Johnston ( 1977) study, and they were in the
same frequency range as those in the Bie­
derman ( 1 966) study. The only procedure
common to all these studies was the presen­
tation of high- and low-printed-frequency
words that differed in bigram frequency. Even
more striking, the present study and that by
Rice and Robinson ( 1 975) used the same
words. Thus, the source of this 20-year dis­
crepancy may reside in the stimulus words
and in the subsequent selection by experi­
menters. For example, consider the words,
boxer. icing, and joker as opposed to loire,
gnome, and assay Intuitively, it seems the
words in the first set would be familiar to most
college undergraduates, whereas those in the
second would be unfamiliar. Yet both groups
of words have frequency scores of I in both
the Thorndike and Lorge ( 1944) and Kucera
and Francis ( 1 967) counts.
A second sampling error can occur when
low-printed-frequency words are selected for
material sets that manipulate other properties
of the stimulus words. For example, Rice and
Robinson (1975) selected two groups of low­
printed-frequency words, each occurring once
per million and hence matched for printed
frequency. One group was composed of words
such asfitmble, mumble, giggle, drowsy, snoop,
and lava. A second group contained words
such as cohere, heron, rend, char, cant, and
pithy The words in the first group comprised
low-frequency bigrams; the words in the sec­
ond comprised high-frequency bigrams. Rice
and Robinson found slower RTs to words in
the second group and concluded that high hi­
gram frequency interfered with recognition of
low-printed-frequency words.
Another explanation may be that the words
in the first set are simply more familiar.
Gernsbacher (1983) had subjects rate their
subjective, termed "experiential," familiarity
with 455 low-printed-frequency words. The
reliability of these ratings was high; different
raters agreed closely. More important, the
range of ratings was broad and well distributed,
suggesting that words with the same low­
printed-frequency score can differ substantially
in their experiential familiarity.
A difference in the experiential familiarity
of the stimulus words used in previous studies
could explain not only the paradoxical inter­
action between printed frequency and bigram
frequency but also the reverse interaction or
even the absence of an interaction. That is,
given the sampling error that may occur with
printed frequency counts, the probability of
confounding experiential familiarity with hi­
gram frequency would be most likely to occur
in words selected from the low-printed-fre­
quency range. Studies reporting that low­
printed-frequencyjlow-bigram words were
better recognized might have used low-printed-
frequency/low-bigram words that were more
familiar than their low-printed-frequency/
high-bigram counterparts. Studies reporting a
significant interaction in the opposite direction
might have used materials with an opposite
confound. Studies reporting no interaction
probably avoided the confound. To test this
possibility, a measure of the experiential fa­
miliarity of the low-printed-frequency words
used in those studies was needed.
Experiment 2
Subjects. Subjects were 44 native English speakers at
the University of Texas at Austin who were enrolled in an
introductory psychology course and who participated in
the experiment to fulfill a course requirement. Data from
an additional subject were excluded because he failed to
perform the task carefully, as indicated by his responses
to the catch words.
Materials. The experimental set of words comprised
all the low-printed-frequency words from the materials
used by Orsowitz ( 1963, cited in Biederman, 1966), Bie­
derman ( 1966), Broadbent and Gregory (1968), and Rice
and Robinson (1975), and 40 low-printed-frequency words
used in a study by Rubenstein et al. ( 1970). Thus the
experimental set consisted of 42 low-printed-frequency
words composed of high-frequency bigrams and 42 low­
printed-frequency words composed of low-frequency bi­
grams taken from four of the studies reviewed earlier, as
well as 40 low-printed-frequency words from the Ruben­
stein et al. (1970) stimuli. In addition to the 124 words
from the five previous studies. 7 five-letter words of high
(AA) printed frequency were added as a check for the
validity of individual subject's rating. As a second validity
measure, 7 five-letter nonwords. which conformed to the
rules of English orthography, were constructed and added
to the stimulus list. An additional 37 low-printed-frequency
words, which matched the average letter length of the ex­
perimental words, were selected as filler words.
All 175 words were randomly arranged and typed on
seven pages, 25 words to a page, with the constraint that
no more than one of either type "control" (i.e., AA or
nonword) word appeared on a page. The words, typed in
capitals, appeared down the left-hand margin. Opposite
each word was a 7-point numerical scale, with its ends
of the seven pages was randomized, and the pages were
collated into a booklet that included a cover sheet with
written instructions.
Procedure. Subjects rated how familiar they were with
each word on the list. Specific instructions were read silently
by each subject while the experimenter read them aloud.
Subjects were then encouraged to work at their own rate.
Results and Discussion
Sums were computed for each word at each
level of the 7-point scale. Two subjects failed
to respond to every item in their booklets;
I), using the exact same stimuli, reported the
therefore, sums were converted to proportions
exact opposite finding, a bigram frequency ad­
vantage. The results of the familiarity ratings
by dividing the total number of responses for
a given level by the total number of subjects
responding to that item. Mean proportions
obtained for the original Orsowitz stimuli are
equivocal. In the present experiment, Orso­
were tabulated for the words within each orig­
witz's high-bigram words were rated as VERY
inal condition of a previous study. The results
of this experiment are compared with those
bigram words were rated as VERY FAMILIAR
of the original studies in Table I.
by 28.00% of the subjects. This difference is
not statistically significant.
Broadbent and Gregory (1968) and Rice
Robinson (1975) reported
by 31.25% of the subjects; the low­
To summarize, the low-printed-£requency
printed-frequency words composed of low­
words used in some previous experiments ap­
parently differ in their rated experiential fa­
frequency bigrams were better recognized than
low-printed-frequency words composed
high-frequency bigrams. Experiment 2 re­
vealed that the low-printed-frequency /low-bi­
miliarity. Irrespective of orthographic fre­
quency, the mean levels of experiential fa­
miliarity found in Experiment 2 could easily
Table 2
Mean Reaction Time (RT) and Percentage of Errors in Experiment 3
High familiarity
Low familiarity
RT (ms)
they performed below the a priori error criterion of no
more than 30% errors in any one of the six experimental
Design and rna/erial s.
Four groups of 20 five-letter
words were selected from the aforementioned corpus. One
group consisted of words that were rated as VERY FAMILIAR
by at least 75% of the subject raters and that comprised
high·frequency bigrams. A second group consisted of words
account for many of the observed interaction s
with low printed frequency reported in the
that were also rated as VERY FAMILIAR by at least 75% of
the present subjects, whereas their low-printed­
original experiments. Furthermore, experi­
FAMILIAR by no more than 15% of the raters and that
frequencyfhigh-bigram words were rated as
ential familiarity might well account for ar­
comprised high-frequency bigrams. The last group con­
by only 46.07% of these sub­
tifactual differences in other experiments, since
3.50, p < .00 I. In addition, the
the present experiment investigated only stud­
low-printed-frequencyflow-bigram words used
ies in which authors had published their
by Rice and Robinson were rated as VERY
jects, 1(28)
The purpose of Experiment 3 was to ex­
low-printed frequency/high-bigram words(74%)
amine this hypothesis more directly. As pre­
used in that study, 1(28)
3.09, p < .00 I.
the raters but that comprised low-frequency bigrams. A
third group consisted of words that were rated as VERY
sisted of words that were rated as VERY FAMILIAR bv no
more than 15% of the raters and that comprised low­
frequency bigrams. The bigram frequencies were obtained
from the data presented by Massaro, Taylor, Venezky, Jas­
trzembski, and Lucas(1980). The mean summed bigram
by more subjects (96%) than the
viously mentioned, Gemsbacher (1983) ob­
frequency was 8,395 for the high-familiarity/high-bigram
words, I ,069 for the high-familiarity/low-bigram words,
8,340 for the low.familiarity/high-bigram words, and I ,029
for the low·familiarity/low-bigram words. (Units for the
Biederman ( 1966, Experiment 2 ) reported
the opposite effect, namely, that low-printed­
tained experiential familiarity scores for all
bigram frequency scores are the number of occurrences
five-letter words indexed by Thorndike and
per million words for each of the four bigrams in a five­
frequency words composed of high-frequency
Lorge ( 1944) occurring once per million. The
bigrams were recognized better than low­
stimulus words were drawn from this corpus.
printed-£requency words composed of low­
fre q u e n c y b i g r ams. H i s l o w - p r i n t e d-fre­
In Experiment 3, subjects made lexical deci­
sions to words that were factorial arrangements
quency /high-bigram words were rated a s VERY
of experiential familiarity and bigram fre­
by 50.13% o f the subjects in the
present study, whereas his low-printed-f re­
quency, each at two levels. It was expected that
lexical decisions to words with high experi­
letter word. These are summed and positional.)
The non word stimuli were constructed in the same way
as those of Rice and Robinson(1975). Five-letter non words
the nonwords were first·order approximations to English
words. Nonlexicality in this and all subsequent experiments
was defined as failure to appear in the unabridged Webs/ers
New l�or/d DICtionary ( 1981). I n addition, nonwords that
ential familiarity would be faster than to words
contained embedded real words of three letters or more
were not used. Forty nonwords were selected to match
However, perhaps because there were only
effect of or interaction with bigram frequency
the mean of the high-bigram word stimuli, collapsed over
eight words per cell, this 20% difference in
would result.
by only 2 9.57% of these subjects.
mean ratings is only marginally significant at
Yet, when the familiarity data for the Bied­
erman high-bigram stimuli are added to those
generated by the Broadbent and Gregory
(1968) low-bigram stimuli, and when the
Biederman low-bigram stimuli are added to
the mean of the word items with low bigram frequency.
The mean bigram frequency of these nonwords was I ,040.
The subjects in this and the subsequent three
experiments were drawn from the same population ofth�
who had generated the familiarity ratings (Gernsbacher
the Broadbent and Gregory high-bigram stim­
1983), and all were native English speakers. N:J subject
who had served in the original rating experiment served
uli, the combined test is highly significant,
in any of the present experiments, nor did any subjeo
1(44) = 3.41, p < .001.
Finally, Orsowitz (1963, cited in Biederman,
familiarity. The mean bigram frequency of these nonwords
was 8,358. Another 40 nonwords were selected to match
Experiment 3
a conservative level, 1(14) = 1.38, p < .091.
more than one experiment. The subjects
in Ex·
periment 3 were 19 undergraduate students enrolled ir
introductory psychology at the University of Texas at AuS
1966) reported a bigram frequency disadvan­
tin, who participated to fulfill a course requirement.
tage, whereas Biederman (1966, Experiment
data from 3 additional subjects were excluded becaus
front of the television screen. A stimulus trial consisted
of the presentation of a warning dot in the center of the
television screen, appearing coincident with a short warning
tone and followed 5 0 0 ms later by the stimulus item. A
millisecond timer was activated coincidentally with the
presentation of the stimulus item. The stimulus item re­
mained in view until subjects in both booths had re·
sponded. One second elapsed between the removal of a
stimulus item and the presentation of the warning dot and
tone of the next trial.
Subjects were informed of the sequence of events for
each stimulus trial. They were told that they would be
shown groups of letters and that their task was to decide
whether the letters formed a real word in English. All
subjects used the index finger of their preferred hand to
indicate "yes" and the index finger of their non preferred
hand to indicate "no." Subjects were informed that ap­
proximately half of the letter groups would indeed form
real words and half would not and that some of the real
words presented might be slightly unfamiliar to them.
Further instructions stressed speed as well as accuracy.
The experimenter answered any questions about the task;
subjects were given I 0 practice trials, which included at
least one stimulus item characteristic of each of the six
stimulus conditions, and then subjects were presented with
the experimental materials.
were generated by a computer program that selected letter
pairs according to their bigram frequency. By this method,
with low experiential familiarity, but that
quency/low-bigram words were rated as VERY
were rated as VERY FAMILIAR by 74.33% of
gram words used by Broadbent and Gregory
Results and Discussion
For correct RTs, a mean and standard de­
viation were computed for each subject and
for each item in the experiment. Any indi­
vidual RT that was more than 2.5 SD away
from both the mean performance for the sub­
ject in that condition and the mean RT to the
item across subjects was replaced, following
the procedure suggested by Winer (1971).
The experiment was therefore a 3 X 2 (Word Type X
Subjects' mean RTs and percentage of errors
Bigram Frequency) design, with both variables manipu­
f or each of the six experimental conditions are
lated within subjects.
Apparaws and procedure.
The experiment was con­
trolled by a Digital Equipment Corporation PDP·II/03,
which was responsible for stimulus randomization, stimulus
Presentation, and data collection. The five-letter strings
we re displayed in uppercase white Matrox letters on the
blacl background of a Setchell Carlson television screen.
Two subjects were tested in each experimental session,
wuh subjects occupying separate booths and the experi·
shown in Table 2. All ANOVAS conducted on
mean RTs were also conducted on mean per­
centage of errors, and no discrepancies were
found between the two sets of results. There­
fore, only the results of the ANOVAs performed
on mean R T are reported.
The mean RTs of the 19 subjects and the
menter monitoring the session from an adjacent room.
160 stimulus items were both submitted to a
Subjects were seated approximately 3 ft (0.9144 m) in
3 X 2 (Word Type X Bigram Frequency) AN-
OVA. In one analysis, subjects were treated as
random effects; in a second, items were treated
as random effects (Clark, 1973). In addition,
the item analyses of all three levels of famil­
iarity included a statistical procedure for un­
equal cell sizes. These ANOVAS revealed a sig­
nificant main effect of experiential familiarity
in both the analysis by subjects, F,(2, 36)=
25.02, p < .001, and the analysis by items,
F2(2, 1 57) = 41.8 1 , p < .001; F�;0(2, 73) =
17.2 1 , p < .00 I. As can be seen in Table 2,
high-familiarity/low-printed-frequency words
were recognized more than 250 ms faster than
those rated as less familiar yet of equal fre­
quency of occurrence in printed English.
The 3 X 2 ANOVA, with subjects as random
effects, also revealed a main effect of bigram
frequency, F1(1, 18)= 9.18, p < .007, and an
interaction between experiential familiarity
and bigram frequency, F1(2, 36)= 8.55, p <
.00I. However, these last two effects failed to
reach a conservative level of significance in
the analysis in which items were considered
random effects, F2(1, !54)= 3. 1 3, p < .079,
and F2(2, !54)= 2.43, p < .092. Inspection
of the six conditions' means revealed that the
difference in RT to high and low bigram fre­
quency was only 1 9 ms in the high-familiarity
conditions. For the low-familiarity items, this
difference was only 23 ms. The greatest dif­
ference between high and low bigram fre­
quency (87 ms) occurred with the nonword
stimuli. Therefore, planned comparisons were
performed separately on the data from the
word and the nonword conditions. These
planned comparisons revealed that the effect
of bigram frequency was significant only in
the nonword condition, F1( I, 1 8)= 21.42, p <
.00 I; F2( I, 78) = 8.67, p < .004; F�;0( I, 90)=
6. 1 7, p < .025. In contrast, in the word con­
ditions, bigram frequency was not significant
(F1 < 1.0, F2 < 1 .0), nor was the interaction
between experiential familiarity and bigram
frequency, F1(1, 18)= 2.70; F2(1, 76) 1 .22;
allps > .IO.
Two regression analyses clarify the effects
of experiential familiarity and bigram fre­
quency in the word data. In the first, com­
binations of the two independent variables,
the mean familiarity rating (percentage of rat­
ers responding VERY FAMILIAR ) and the
summed bigram frequency, were used to pre­
dict mean correct RT. In the second, the total
error rate for each stimulus word was the cri­
terion variable; the two predictor variables
were the same. These analyses revealed that
rated familiarity accounted for more than 55%
of the variance found in the RT data, F( I,
78)= 98.01, p < .001, and for approximately
44% of the variance found in the correspond­
ing error data, F( l , 78)
6 1.36, p < .00 1 .
Conversely, bigram frequency explained only
an additional 0.3% of the variance found in
either measure, and entrance of this variable
into either regression equation was not statis­
tically warranted (F < 1.0). All these analyses
show that lexical familiarity, operationalized
as experiential familiarity, is the more critical
variable affecting the speed and accuracy of
recognizing an English word.
Although bigram frequency did not affect
the recognition of real words, it did signifi­
cantly affect the recognition of non words. An
examination ofthe nonwords used in both the
Rice and Robinson (1975) study and the pres­
ent Experiment 3 revealed that non words gen­
erated by a computer program, though they
might be first-order approximations to English,
differ in pronounceability. In a critical study,
Rubenstein, Lewis, and Rubenstein (197 1a;
see also Rubenstein, Richter, & Kay, 1975)
demonstrated that within a lexical decision
task, pronounceable nonwords are harder to
reject as nonwords than are unpronounceable
ones. Thus, in Experiment 3, it might have
been the pronounceability rather than the hi­
gram frequency that affected performance.
To examine this possibility, the nonword
stimuli were first classified by two independent
judges as pronounceable or unpronounceable.
Their decisions agreed closely (r= .982). The
mean RTs and mean percentage of errors to
the nonwords were then analyzed by a one­
way ANOVA, with the independent variable of
pronounceability. The between-group differ­
ence found in both analyses was statistically
significant: for the RT data, F(l, 78) = 20.24,
p < .00 I; for the error data, F(I, 78)= 1 !.51,
p < .00 I. A post hoc analysis verified that the
mean RT to the pronounceable nonwords
(I,078 ms) was significantly greater than that
to the unpronounceable nonwords (925 ms),
4.4 1 , p < .001, and that the mean
error rate to the pronounceable nonwords
(2.26%) was significantly higher than that
to the unpronounceable nonwords (1.47%),
Mean Reaction Time (RT) and Percentage of Errors in Experiment 4
High familiarity
Low familiarity
RT (ms)
1(62)= 3.31, p < .00 1 . In addition, regression
analyses indicated that pronounceability in­
dependently accounted for 20% of the variance
in RTs, F( I, 78) 20.24, p < .00 I, and for
!5% of the variance in error rate, F( l , 78)=
11.51, p < .00 I . Bigram frequency was a
weaker independent predictor: It accounted
for 10% of the RT variance, F( l , 78) 9.04,
p < .0 I, and 6% of the error rate variance,
F(l, 78) 4.97, p < .03. When added to the
regression on pronounceability, bigram fre­
quency predicted only an additional 5 % of the
RT variance, F( l , 77) = 5 .24, p < .03, and
an insignificant 3% of the error rate variance,
F(l, 77)= 2.56, p > . 1 0.
These results seem to suggest that pro­
nounceability, as opposed to bigram frequency,
was responsible for the main effect of bigram
frequency revealed in the nonword data, but
caution is needed here. Massaro, Venezky, and
Taylor ( 1 979a, !979b) noted that pronounce­
ability is so often correlated with bigram fre­
quency, as well as single-letter frequency, that
it is difficult to separate the independent con­
tribution of either measure of orthographic
structure (cf. Krueger, 1979; Mason, 1 975).
Experiment 4 was conducted to investigate this
question. Experiment 4 was a replication of
Experiment 3, without the question of pro­
nounceability interfering with interpreting any
possible bigram effect. Subjects were presented
with the same real words as those in Experi­
ment 3. However, in order to control for the
possible confounding of pronounceability and
bigram frequency in the nonwords, all non­
words presented in Experim ent 4 were un­
Errors (%)
Errors (%)
quirement. Data from 2 additional subjects were excluded
because they performed below the a priori error criterion.
Design and materials. The real word stimuli used in
Experiment 3 were used again in Experiment 4. Again,
of the 80 five-letter words, 20 were high-familiarity/high­
bigram words, 20 were high-familiarity/low-bigram words,
20 were low-familiarity /high-bigram words, and 20 were
low-familiarity/low-bigram words. The nonword stimuli
consisted of the 34 nonwords used in Experiment 3 that
had been rated as unpronounceable and an additional 46
non words chosen from a pool offive-letter strings generated
by a computer program. These additional non words were
similarly rated by two independent judges, and only those
unanimously judged as being unpronounceable were re­
tained. The mean summed bigram frequencies for the two
sets of nonwords were 8,340 for the 40 high-bigram non­
words and 1,020 for the 40 low-bigram non words.
Apparatus and procedure. The apparatus and proce­
dure were identical to those used in Experiment 3.
Results and Discussion
Correct RTs were edited in the same manner
as in Experiment 3, and all ANOVAs conducted
on mean RTs were also conducted on mean
percentage of errors. No discrepancies were
revealed between the two sets of results, and
so only the mean RT results are reported.
The mean RTs of the six experimental con­
ditions are presented in Table 3. A 2 X 2 AN­
OVA on the responses to the real words revealed
a strong main effect of experiential familiarity,
F1(1, 17)
1 65.43, p < .00 1 ; F2(1, 76) =
45.35, p < .00 I; F�;n( I, 92)= 35.93, p < .00 I .
As in Experiment 3, high-familiarity words
were recognized more rapidly than low-fa­
miliarity words. Bigram frequency had no sig­
nificant main effect, nor did it interact with
experiential familiarity: for main effect, F1( 1 ,
1 7) 3. 78, F2( I , 76) 2.68; for interaction,
F1( 1 , 17) = 3.15, F2(1, 76)
2.28; all ps >
Experiment 4
.I 0. The analyses of the nonword data also
failed to reveal a main effect of bigram fre­
quency, F1(1 , 1 7)= 2.68; F2(1, 76)
1 .08;
Subjects. The subjects were 18 undergraduate students
enrolled in introductory psychology at the University of bothps > . 1 0.
Texas at Austin. They participated to fulfill a course reThe failure ofthe bigram frequency variable
to significantly affect response latencies in ei­
ther the word or nonword conditions supports
the hypothesis that the effect of bigram f re­
quency in the nonword condition of Experi­
ment 3 was simply due to a failure to control
for pronounceability across the high- and low­
bigram conditions. Moreover, the lack of a
significant effect of bigram frequency and,
more important, the lack of an interaction of
bigram frequency with the familiarity variable
support the hypothesis that the interaction be­
tween bigram frequency and printed frequency
found in previous studies was due to a failure
to control for the experiential familiarity of
their low-printed-frequency words. Taken to­
gether, the results of Experiments 3 and 4
strongly suggest that bigram frequency has of ­
ten been confounded with experiential famil­
iarity. This in turn has led to the inconsistent
findings of an i nteraction between the two
The Inconsistent Interaction
Between Printed Frequency and
Semantic Concreteness
Rubenstein et a!. ( 1970) provided a third
pattern of results. In that study, lexical decision
had used unpronounceable nonwords. In his
fourth experiment, he used a preexperiment
Effects of Printed Frequency and Semantic
RTs indicated a main effect of printed fre­
quency, no main effect of concreteness, and
no interaction. And four experiments by James
familiarization task (subjects were presented
with each word, were asked to create a sentence
using it, and were supplied with a definition
of any word they claimed was unfamiliar). The
familiarization task was assumed to have the
Table 4
Results of Sl!ldies That Ha1'f! Examined the
Original study
Concrete worse than abstract
Paivio & O'Neill
Concrete worse than abstract
Concrete better than abstract
Experiment I
Concrete better than abstract
Concrete better than abstract
Concrete equal to abstract
Concrete equal to abstract
Experiment 4
Rubenstein, Garfield,
& Millikan
Richards ( 1976,
Concrete equal to abstract
Concrete equal to abstract
Note. Data are for low-printed-frequency words only.
tically concrete words than semantically ab­
stract words; this difference was exaggerated
in subjects' performance with the low-printed­
frequency words.
Richards ( 1976) reported the results of two
similar experiments. The temporal threshold
higher probability of occurring in printed text
than words referring to abstract or intangible
data of the first also indicated a main effect
for printed frequency, no main effect of se­
mantic concreteness, and a significant inter­
action between the two. However, the direction
of the interaction in Richards's study was dif­
ferent from that in Winnick and Kressel's
(1965) and Paivio and O'Neill's (1970): Fcr
vides a summary of these results.
Winnick and Kresse! ( 1965) found a sig­
nificant main effect of printed frequency but
no main effect of semantic concreteness on
tachistoscopic thresholds. However, there was
a marginally significant interaction: Concrete
low-printed-frequency words took longer to
recognize than abstract low-printed-frequency
words. Paivio and O'Neill ( 1970) also corrob­
orated the well-established finding that high­
printed-frequency words were recognized in
fewer trials. In addition, their subjects required
significantly more trials to recognize seman-
(1975) provided an even broader spectrum of
results. James's first experiment revealed no
Winnick & Kresse!
Another variable that covaries with printed
frequency is semantic concreteness. Words re­
ferring to concrete or tangible items have a
items (Glanzer & Bowles, 1976; Paivio et al.
1968). During the past decade or two, re­
searchers have examined the effects of printed
frequency and semantic concreteness on word
recognition. Like the experiments investigating
the effects of printed frequency and orthog­
raphy, the results of the experiments manip­
ulating printed frequency and semantic con­
creteness have been inconsistent. Table 4 pro­
concrete words, thresholds declined system­
atically as a function of printed frequency, but
for abstract words they did not. I n a second
main effect of concreteness but did show a
significant interaction mirroring the interac­
tions discovered by Richards (1976). The sec­
ond experiment revealed the same interaction,
as well as a main effect of concreteness. Con­
versely, the third and fourth experiments re­
vealed neither a significant interaction nor a
main effect of concreteness.
James ( 197 5) attributed these results to the
differential levels of processing required by the
demands of his paradigm: the lexical decision
task. James ( 1975) likened responding in a
lexical decision task to searching for a word
words accounts for the main e ffe<.:t of printed
frequency found in all four experiments but
cannot account f or an i nteraction between
printed frequency and semantic concreteness,
James termed "lexical processing," is sufficient
for making a response. In other situations, a
much less a main effect of the latter variable.
deeper level of processing, what James termed
"semantic processing," might be required. In
semantic processing of abstract words should
take longer than that of concrete words. Even
his dictionary analogy, this deeper semantic
processing was likened to going a step beyond
merely locating the desired entry to perhaps
"reading" the appropriate definition of the
target word. Deep semantic processing should
granting that low-printed-frequency words re­
quire deeper semantic processing, why should
take longer than the more superficial lexical
processing and this should be reflected in lon­
ger latencies.
James (197 5) proposed that in his four ex­
periments he had manipulated depth of pro­
cessing by varying the familiarity of the stim­
ulus words and the type of catch trials (the
nonwords). With highly familiar words, op­
erationalized as high-printed-frequency words,
little or no semantic processing should be re­
quired, only lexical processing. In contrast,
actions between printed frequency and con­
v.ith low-printed-frequency words, deeper se­
creteness were significant. Richards explained
the inconsistency by pointing out that in his
first experiment, only 2 concrete words and 2
mantic processing should be required because
ment were possibly artifactual, whereas those
of his second were not.
The notion that additional semantic processing
is required for the low-printed-frequency
in a dictionary. In some experimental situa­
tions, merely locating a lexical entry, what
experiment, Richards found main effects fcr
printed frequency and concreteness. But un­
like in his first experiment, none of the inter­
abstract words were presented at each of eight
levels of printed frequency. In contrast, in the
second experiment, 16 and 9 words were pre­
sented at each of two or three levels. Richards
concluded that the results of his first experi­
effect of "tem�orarily raising the subjective
frequency" (p. 134) of the real words. Ac­
cordingly, James surmised that the optimal
level of processing need not extend past the
more superficial lexical processing; thus no
effect of nor interaction with the semantic
concreteness variable would be realized.
Yet, the theoretical framework proposed by
James (1975) only partially explains his results.
merely locating a lexical entry is insufficient
fcr discriminating a low-printed-frequency
word from a highly similar nonword distractor.
However, according to James ( 1975), pro­
cessing need not be at the deeper level even
k>r low-printed-frequency words when the
nonwords are unpronounceable and thus ex­
tre mely dissimilar to the target words. In his
third experiment, unlike in his first two, he
That is, his theory lacks a rationale for why
the abstract meanings of these low-printed­
frequency words be more difficult to "read"
than the concrete meanings?
framework posited by James ( 1975) is insuf­
ficient in accounting for the results reported
by Winnick and Kresse! ( 1965) and Paivio
and O'Neill (1970). Both studies reported that
recognition performance with low-printed­
frequency/concrete words differed f rom that
with low-printed-frequency/abstract words·
but neither study presented pronounceabl
nonwords nor nonwords of any type. More­
over, in James's terminology, both found that
concrete meanings of low-printed-frequency
words were
difficult to "read" than ab­
stract meanings.
To summarize, all of the studies reviewed
in this section have factorially manipulated
printed frequency and semantic concreteness.
Their results have been inconsistent. Many ex­
perimenters have reported an i nteraction be­
tween the two variables, but neither this in­
teraction nor its direction has been replicated
across all experiments, even those performed
by the same experimenter.
The source of these inconsistent interactions
could be the same as the source of the incon­
sistent interactions between printed frequency
and bigram frequency: the inadequacy of
printed frequency counts in reflecting expe­
riential familiarity. Direct evidence that ex­
periential familiarity has been confounded
with semantic concreteness was found in post
hoc analyses conducted by Paivio and O'Neill
( 1 970). They too questioned the reliability of
printed frequency and so they obtained ratings
of subjective familiarity for each of their stim­
ulus words. Rated familiarity correlated
strongly with both the concreteness values and
the recognition scores. When rated familiarity
was partialed out, the correlation between the
concreteness values and recognition scores
dropped dramatically to zero.
Other studies reviewed in this section might
also have been flawed by relying on printed
frequency as a reliable index of lexical famil­
iarity, and their results might be better attrib­
uted to experiential familiarity than semantic
concreteness. Experiment 5 was intended to
test this possibility. In order to manipulate
lexical familiarity, the stimulus words used in
Experiment 5 were also selected from the
rated, low-printed-frequency words collected
by Gernsbacher (1983). Experiment 5 also di­
rectly tested James's ( I 975) assertions con­
cerning the differential effects of nonword
pronounceability in a lexical decision task.
Experiment 5
Subjects. The subjects were 20 undergraduate students
at the University of Texas at Austin, enrolled in introductory
psychology, who participated in the experiment to fulfill
a course requirement. Eleven subjects were randomly as­
signed to the unpronounceable nonword condition; 9 were
assigned to the pronounceable nonword condition. Data
from two additional subjects in the pronounceable con­
dition were excluded: One subject failed to perform above
the a priori error criterion. and the other subject's mean
latencies, in all conditions, were well above 2.5 s.
Design and materials. The word stimuli were selected
from the aforementioned corpus of low-printed-frequency,
five-letter words. The selection of abstract as opposed to
concrete nouns was accomplished in the following manner.
Two independent judges were given 125 high-familiarity
nouns, namely, all the nouns to which 50%-9 3% of the
raters had responded VERY FAMILIAR, and 125 low-fa­
miliarity nouns, namely, all the nouns to which only 7%-
Table 5
Mean Reaction Time (RT) and Percentage of
Errors to Words in Experiment 5
Non word
20% of the raters had responded VERY FAMILIAR. From
each of these two lists, the judges were instructed to select
40 nouns that "specifically referred to a tangible object,
person or thing" and 40 nouns that "primarily referred
to an intangible person, object or thing." The judges were
supplied with the definition of each noun, taken from
Webster's New Collegiate Dictionary ( 1976), to aid them
in their decision. From these four lists of 40 nouns each,
four experimental groups of 20 nouns each were selected
by factorially combining high and low familiarity with
semantic abstraction and concreteness. This selection was
made with the constraints that each stimulus noun must
have appeared on both judges' lists and that across the
concrete or abstract conditions, the noun sets were matched
for mean familiarity ratings. The mean familiarity ratings
for the high-familiarity, semantically concrete or seman­
tically abstract nouns were 64.55% and 64.30%, respec­
tively: the mean familiarity ratings for the low-familiarity,
semantically concrete or semantically abstract nouns were
1 3.32% and 13.68%, respectively.
The nonword stimuli were selected from a pool generated
by a computer program that produced second-order ap­
proximations to real English words. Eighty nonwords were
selected that were unpronounceable, and 80 nonwords
were selected that conformed to English pronunciation
rules. Both groups of nonwords were matched for their
summed positional bigram frequency: The means of the
unpronounceable and pronounceable nonwords were 3,364
and 3,517, respectively. Half of the subjects were randomly
assigned to the pronounceable nonword condition and
half, to the unpronounceable nonword condition.
Apparatus and procedure. The apparatus and proce­
dure used in Experiment 5 were identical to those used
in Experiment 3.
Results and Discussion
Correct RTs were edited as in Experiment
3. Subjects' mean RTs and percentage of errors
to the word items in each of the four exper­
imental conditions are shown in Table 5 . All
ANOVAs conducted on mean RTs were also
conducted on percentage of errors, and no dis­
parity was revealed between the two sets of
results from any of the A NOVAs performed on
the two dependent measures. Again, only the
results of the ANOVAs performed on mean RTs
are reported.
Because of the incomplete factorial design,
the data from the words-only conditions were
first analyzed separately from those of the
nonword conditions. The mean RTs of the 20
subjects and 80 items were both submitted to
a 2 X 2 X 2 (Familiarity X Concreteness X
Pronounceability) ANOVA. The ANOVA per­
formed with subjects as random effects in­
cluded a statistical procedure for unequal cell
size. These ANOVAS revealed a significant main
effect of experiential familiarity, F t ( J , 18)
23.32, p < .001; F2( J , 76) 30.90, p < .00 1 ;
F:,;n( l , 49) 13.29, p < .00 1 , such that high­
familiarity words were recognized more than
143 ms faster than low-familiarity words. In
addition, a significant main effect of pro­
nounceability was obtained, F1 ( I , 18) 8.27,
p < .010; F2( l , 76)
45.60, p < .001; F:,;n( J ,
25) 7.00, p < .025, such that subjects' re­
sponses were 1 25 ms slower to pronounceable
nonwords than to unpronounceable nonwords.
In interpreting this result, the important fact
is that the word stimuli were the same across
the two pronunciation conditions.
More germane to resolving the previous in­
consistent findings are two other aspects of the
present data. First, the concrete versus abstract
variable had no main effect (all Fs < 1.0), nor
did it reliably interact with any other exper­
imental variable (all Fs < 1.0). Indeed, when
collapsing over the other two experimental
variables, subjects' mean RT to concrete words
differed from that to abstract words by an av­
erage of only 12 ms, with the largest concrete
versus abstract RT difference observed in any
of the four conditionalized comparisons being
approximately 24 ms.
Second, the only significant interaction
found in these data was an interaction between
familiarity and pronounceability, F 1( 1 , 18)
9.70, p < .007; F2(1, 76) 10.04, p < .002;
F:,;n( l , 57) 4_93, p < .037. This interaction
is displayed in Figure 2. In the pronounceable
nonword condition, low-familiarity words were
recognized 190 ms more slowly than high­
familiarity words. But in the unpronounceable
nonword condition, this difference was re­
duced to 96 ms. Thus, the manipulation of
pronounceability differentially affected rec=
ognition performance with respect to the
words' experiential familiarity, not their con­
This interaction was also suggested by the
data of Experiments 3 and 4. The only dif­
ference between those two experiments was
the pronounceability of their non words. And
like the present experiment, there was a larger
difference in mean RT between the high- and
low-familiarity word conditions when the
nonwords were pronounceable (Experiment 3)
than when they were unpronounceable (Ex­
periment 4). This interaction provides an al­
ternative explanation of the experiments re­
ported by James ( 1 975).
As in any decision-making task, the more
closely the lures resemble the targets, the
stricter the criterion employed to decide be­
tween the two must be, and vice versa. In RT
tasks, relative differences in criteria are man­
ifested in both speed and accuracy (Kiger &
Glass, 1 981; Laming, 1979; Ratcliff, 1978).
So, in these lexical decision tasks, a stricter
criterion was probably needed to decide be­
tween the real words and the more wordlike
pronounceable nonwords than between the
real words and the less wordlike unpro­
nounceable nonwords. When this stricter cri­
terion must be employed, although responses
to high-familiarity words are also made more
slowly, responses to low-familiarity words are
made even more slowly. This is simply because
the low-familiarity words are even harder to
discriminate from the lures. Thus, the presence
of pronounceable nonwords accentuates the
difference between high and low familiarity.
Returning to James's ( 1 975) data, one hy­
pothesis is that his low-printed-frequencyI
concrete words differed from his low-printed­
frequency/abstract words in their overall level
of experiential familiarity though not in their
printed frequency. If so, the presence of pro­
nounceable nonwords would accentuate this
difference, creating the spurious interaction
between printed frequency and concreteness.
In other words, the mechanism underlying the
differential effects caused by manipulating
pronounceability was probably a shift in sub­
jects' decision criteria rather than a shift to a
level of semantic processing.
To evaluate this hypothesis, data from the
present experiment were used to estimate how
much James's (1975) low-printed-frequency/
concrete words would need to differ in fa­
miliarity from his low-printed-frequency/ab­
stract words in order to produce his results.
Two regression equations were calculated from
multiple regression analyses performed on the
mean RTs from both the pronounceable and
unpronounceable nonword conditions of Ex­
periment 5. The predictor variables in both
equations were experiential familiarity (en­
tered as a continuous variable, i.e., percentage
of subjects who considered the word HIGHLY
FAMILIAR ) and semantic concreteness (entered
as a dichotomous variable). Only the famil­
iarity variable satisfied the equation's signifi­
cance criterion for entrance; the variable of
semantic concreteness was not significant ei­
ther when entered alone or when added to the
familiarity variable (all Fs < 1.0). Both equa­
tions using only the familiarity variable were
highly significant: for the pronounceable con­
dition, F( l , 78) 32.36, p < .00 I ; for the un­
19 .5 7 ,
pronounceable condition, F( I , 78)
p < .001.
Mean familiarity ratings were predicted for
the low-prin ted-frequency /concrete and low­
printed-frequency/abstract words used in the
James ( 1 975) study by substituting the RTs he
reported for those two conditions (in the ex­
periment with pronounceable nonwords) into
the first regression equatio n. The predicted fa­
miliarity values were 36% for the low-printed­
frequency/concrete words and 27% for the low­
printed-frequency/abstract words, a difference
of only 9%. That his two groups of words ac­
tually differed in familiarity by this predicted
amount is suggested by the range of familiarity
ratings obtained in Experiment 2. If his two
groups did differ by this amount, the difference
1 000
Experiment 5 as a function of familiarity and pro­
Figure 2. Mean reaction time to words presented in
nounceability of nonwords.
in predicted mean RT to the two groups when
unpronounceable nonwords were presented
would be 17 ms. This predicted value was ob­
tained by substituting the predicted familiarity
values of the two groups of words into the
second regression equation, that is, the equa­
tion based on the data from the unpronounce­
able nonwords condition. The difference in
mean RT actually obtained by James, in the
experiment when unpronounceable nonwords
were presented, was 14 ms, which is close to
the predicted 17 ms. Thus it appears that the
effect of experiential familiarity not only pro­
vides a simpler, more tenable explanation of
the data reported by James ( 1 975) but also
quantitatively predicts those results.
The Inconsistent Interaction Between Word
Frequency and Number of Meanings
Printed frequency correlates strongly with
multiplicity of meanings: The higher the prob­
ability of a given word appearing in printed
English text, the more likely it has more than
one meaning (polysemy). Polysemy is of major
interest to theorists who attribute the effect of
printed frequency to the process of retrieving
words from lexical memory. They postulate
that either the structure of the lexicon (how
words are stored) or the processes that operate
on that proposed structure (how words are
retrieved) is a function of a word's frequency
of usage and its multiplicity of meanings.
Rubenstein and his colleagues (Rubenstein
et al., 1970; Rubenstein, Lewis, & Rubenstein,
197 1 b) reported the results of lexical decision
experiments with high- and low-printed-fre­
quency words that were either homographs
(e.g., water and gauge) or nonhomographs
(e.g., money and denim). Both printed fre­
quency and homography independently afl fected RTs. Rubenstein et al. (1970) and Ru­
benstein et al. (197 1 b) proposed a model of
word recognition in which the lexicon is ar­
ranged by printed frequency and a separate
entry exists for each semantically distinct
meaning of a given orthographic pattern. The
j finding of relative independence between a
1 word's printed frequency and its number of
meanings led them to assume, with Sternberg's
(1969) additive factors logic, that these vari­
ables operate in separate stages.
Forster and Bed nail ( 1 976) also measured
: lexical decision latencies to high- and low-
27 1
printed-frequency words that were either ho­
mographs or nonhomographs. In agreement
with the results of Rubenstein et al. (1970)
and Rubenstein et al. (197 1 b), Forster and
Bednall also found a significant main effect of
printed frequency. In contrast to the Ruben­
stein et al. results, they found neither a main
effect of homography nor an interaction be­
tween the two variables. However, an addi­
tional experimental task verified Rubenstein's
proposal of separate lexical entries for each
meaning of a homograph. Forster and Bednall
suggested that the effect of homography ob­
tained by Rubenstein et a!. in their lexical
decision tasks was attributable to "accidental
item sampling errors" ( 1 976, p. 56), as pre­
viously suggested by Clark ( 1 973). In their re­
vised model, Forster and Bednall retained the
general conception that the effect of printed
frequency is realized during retrieval and the
proposal that the multiple meanings of a given
word are stored at different locations. They
discarded the notion that lexical retrieval in­
volved two distinct processing stages; they
proposed instead a single search process that
is not random but serial, exhaustive, and di­
rected by frequency.
Jastrzembski and colleagues (Jastrzembski,
1981; Jastrzembski & Stanners, 1 975) argued
that the results of Rubenstein et a!. ( 1 970,
197 1 b) and Forster and Bednall ( 1 976) were
marred by use of a weak criterion of polysemy,
namely, whether the stimulus word was com­
monly considered to be a homograph. Jas­
trzembski suggested that a more powerful test
of the relation between printed frequency and
polysemy would not entail using lexical stimuli
with double as opposed to single meanings,
but rather lexical stimuli with numerous as
opposed to relatively few meanings. The op­
erational scaling of the number of meanings
variable preferred by Jastrzembski was the to­
tal number of individual definitions for a given
orthographic string, as listed in an unabridged
Thus Jastrzembski ( 1 98 1 , Experiment I)
collected lexical decision RTs to words of high
and low printed frequency that were indexed
as having either many or relatively few indi­
vidual definitions in an unabridged dictionary.
He found a significant main effect of printed
frequency, a significant main effect of number
of meanings, and a significant interaction be-
tween the two variables. The difference be­
tween RTs to words with many dictionary
meanings and RTs to words with few was
greater for words of low printed frequency.
Although Jastrzembski ( 1 9 8 1 ) proposed no
new model, he concluded that any tenable
model of word recognition must account for
all three significant effects he reported. But a
few troublesome issues remain to be resolved.
One major theoretical tenet remains unclear.
How psychologically valid is the dictionary
count definition of polysemy? Consider, as il­
lustration, the words, gauge, cadet. and fudge.
These three words were considered highly fa­
miliar by an average of more than 65% of the
undergraduate raters (Gernsbacher, 1 983). Yet
in reality, how many of these subjects are likely
to have stored in memory all 30 dictionary
meanings of the word gauge, all 1 5 dictionary
meanings of the word cadet, or even all 1 5
dictionary meanings of the word fudge? An
informal survey I conducted revealed that sev­
eral college professors could on the average
provide only 3 definitions of the word fudge,
2 of the word gauge, and 1 of the word cadet.
Thus, it appears that even well-educated sub­
jects can report only a relatively small pro­
portion of the total number of unabridged dic­
tionary meanings of three relatively familiar
Moreover, it is difficult to intuit how many
unabridged dictionary definitions may be
found for any given word. Consider, as illus­
tration, the words, souse, shunt, and thrum,
all of which were rated as being highly familiar
by only 2% to 3% of the subjects, although
they are indexed by 1 7 , 14, and 1 3 respective
meanings in an unabridged dictionary. Con­
versely, several words that received consider­
ably higher familiarity ratings, such as liter,
baggy, and lapel, are indexed by only I dic­
tionary meaning.
A more empirical issue arising from Jas­
trzembski's ( 1 98 1 ) work remains unsettled.
How effective is the manipulation of number
of dictionary meanings? More specifically, does
the difference between the number of diction­
ary meanings operationalized as many and the
number of dictionary meanings operational­
ized as few predict a main effect? In addition
to the two experiments reported by Jas­
trzembski and Stanners ( 1 97 5) and J as­
trzembski ( 1 98 1 ) that have been discussed, six
other experiments in which number of dic­
tionary meanings was manipulated were re­
ported by Jastrzembski ( 1 98 1 ). These nine ex­
periments, and the Rubenstein et al. ( 1 970)
and Rubenstein et al. ( 1 97 1 b) experiments
for which Jastrzembski and Stanners tallied
the number of dictionary meanings possessed
by the stimulus words, are catalogued in
Table 6.
As can be seen in Table 6, the magnitude
of the effect of the number of meanings vari­
able (as indicated by the F:n;n value) is, for the
most part, independent of the magnitude of
the difference in number of meanings manip­
ulated. In order to discern which factor or
factors might be critical in explaining the oc­
currence of a significant main effect, several
one-way ANOVAs were performed on these re­
sults. In all these analyses, each of the 1 1 ex­
periments was considered an individual case,
and the presence or absence of a significant
main effect was considered the grouping vari­
able. These analyses revealed that there was
no discernible difference in the mean number
of meanings manipulated between the two
groups of studies that had or had not obtained
a significant effect, F(l , 9)
3.29, p > . 1 0,
nor were there any differences between the
two groups of studies in the mean number of
meanings possessed by their words with many
meanings or by their words with few meanings
(all Fs < 1 .0). Surprisingly, the difference be­
tween mean RT to words with many meanings
and mean RT to words with few meanings
barely differed between the studies that had
or had not obtained a significant effect, F( I ,
9 ) 4.43, p < .06. Yet what did differ greatly
between these two classifications of studies
were the relative differences in errors produced
in response to the words with many as opposed
to few meanings, F(l , 9) = 1 3 .08, p < .0 1 . As
shown in Table 6, the error rates reported for
words with many meanings did not differ as
vastly across studies, F( 1 , 9) < 1 .0, as did error
rates for words with few meanings, F( 1 , 9)
1 1 .00, p < .0 1 .
Elsewhere, Gernsbacher ( 1 9 84) argued that
a vast majority of errors produced during cog­
nitive RT tasks (e.g., lexical decision, picture­
naming latency, sentence verification, and cat­
egory membership verification) are not always
due to motoric "slips of action" (e.g., Norman,
1 98 1 ; Rabbitt & Vayas, 1 970) but are often
"' N \Q
C""'; O o.r. N -n
0 O N V --:
r--: -.c; o O M .,.) M
N \0 - t- 'o:::t" 'o:::t" OO
...O «::i O O N � «i
- O M t"i \0 0 0
oO O o\ o.r\ o\ � �
M M N f""'' N M M
'-0 \0 0 0\ 1.1"1 0\ 0
r--: -.D N «i «i ...O r---:
V O O\ N - V'I N
_; N .....: � N � N
V'l \0
_: r--:
"'=t" N
0 0\ V'I OO OO O O
� oO -= � N ...O ...O
V M "==t on � � �
due to caref ully conceived, well-executed, and
honest but nonetheless incorrect answers. For
example, in a lexical decision task, if a subject
were asked to determine whether the letter
string VIAND was a real English word, the re­
sponse "is not a word" would be an error.
However, the most likely cause of this erro­
neous response is not that the subject executed
a poorly planned motor response pattern or
that the subject erred while attempting to trade
Table 7
Mean Re action Time (RT) and Percentage of
Errors to J..J!ords in Experiment 6
76) < 1 . 0 . I n addition, the interaction between
the number of meaning variables and the ex­
Many meanings
One meaning
one dictionary meaning, F1(1, 20) < 1.0;Fil,
VIAND is indeed an English word. In reference
to the finding that error rate, particularly error
rate to words with few dictionary meanings,
was a good discriminator of studies that had
or had not found a significant effect of the
number of meanings variable, one plausible
hypothesis is that the studies characterized by
the highest probability of error rate could also
be the studies with the highest probability of
presenting words that subjects did not know
were English words.
Do all these unknown words have lo w
printed frequencies? Th� answer is not .avail­
able from the
published reports of these studies. However,
in most of the experiments that found a mam
effect of number of meanings, stimulus words
were chosen from a wide range of printed fre­
quencies, including words of very low printed
frequency. Hence, the expected questiOn re­
mains to be asked. Given the occurrence of
several previous discrepancies in t e wo.rd �ec­
ognition literature, and given t e Imphcauon
that these former inconsistencies commonly
occurred with the manipulation of printe
frequency, and given the fact that Jastrzembski,
like other researchers, relied on pnnted
frequency as a reliable measure of lexical fa­
miliarity, and in doing so presented low­
printed-frequency words, can the findmgs
reported by Jastrzembski ( !981) also be ex­
plained by experiential familiari ty?' Ex�n­
was designed to explore this possibihty.
only variable that had a significant effect m
these analyses was experiential familiarity,
F1(1, 20)
FAMILIAR or FAMILIAR by an average 75% of the . raters
and that had at least 10 or more ind1v1dual d1c11onary
meanings. One group consisted of words that were also
rated as VERY FAMILIAR Or FAMILIAR by an average 75%
of the raters but that had only I individual d1ct1onary
meaning. One group consisted of words that were rated
as VERY FAMILIAR by an average 15% of the raters and
that had more than 10 individual dictionary meamngs.
The final group consisted of words that were also rated
as VERY FAMILIAR by an average 15% of the raters but
that had only 1 individual dictionary meamng. The number
of meanings was computed from the unabridged Webster's
New World Dictionary (1981 ).
The maximal difference in average number of meaning;
manipulated was constrained lJj the composition ci the
stimulus word pool. However, Jastrzembski and Stanners
to 20.
( 1975) observed the largest difference in mean RT
words with 1 to 10 meanings versus those w1th II
The nonword stimuli used in Experiment 6 were all or­
thographically lega� pronounceable five-letter strings.
Apparatus and procedure.
The apparatus and proce.
dure used in Experiment 6 were identical to those used
in Experiment
Results and Discussion
Correct RTs were edited in the same manner
as used in Experiment 3. Sub jects' mean RTs
and percentage of errors to words in each of
the four experimental conditions are shown
in Table 7 . All ANOVAS conducted on mean
RTs were also conducted on percentage of er­
rors, and no discrepancies were found between
the two sets of analyses.
at the Univer.;ity of Texas at Austin, enrolled in introductory
' Experiential familiarity ratings were collected on
(126) of the low-printed-f requency homographiC and no
homographic words presented in the Rubenstem et al.
(1970) and Rubenstein et al. (1971 b) and Forster and Bed­
nail (1976) studies, with the same procedures descn"
in the Method section a Expenment 2. The results of t
psychology, who participated in the expenment to fulfill
a course requirement. Data f rom I subject were excluded
The subjects were 21 undergraduate students
because he failed to perform above the a priori error cn­
22.56,p < .OOI;F2(1, 76)
p < .OOJ; F;nin(l, 6 6 )
Design and materia ls. Four groups of 20 five-letter
words each were selected from the aforementiOned corpu s.
One group consisted of words that were rated as VERY
ment 6
Experiment 6
periential familiarity variable was not signif­
icant, F1(1, 20) < 1.0; F2(1, 76) < 1.0. The
speed at the expense of accuracy, but rather
that the subject simply did not know that
A s can b e seen in Table 7, t h e mean RT to
words with many dictionary meanings did not
differ from the mean RT to words with only
analyses performed on these ratings mnrored the resul�
in the original lex1cal dec1s1on taSk (for the Forst
& Bednall, 1976, study) and the results presented by ClarK
( 1973) in his reanalysis of the Rubenstem et al. studies
11.28, P < .005.
General Discussion
Six experiments were designed to help clar­
ify three sets of inconsistent findings in the
word recognition literature. These i nconsis­
tencies arose from experiments in which lex­
ical familiarity was orthogonally manipulated
with a second variable of interest. More spe­
cifically, each concerned the difference in rec­
ognizing low-familiarity words (operational­
ized as low-printed-frequency words) as a
function of orthographic regularity, semantic
concreteness, or polysemy.
Experiment I tested the sophisticated
guessing hypothesis that had been proposed
to explain why in some but not all experi­
performance with low-printed-fre­
quency words composed of high-frequency
bigrams was worse than performance with low­
printed-frequency words composed of low­
frequency bigrams. When the results of Ex­
periment I failed to support this hypothesis,
an alternative hypothesis was entertained: The
two groups of low-printed-frequency words
used in previous experiments could have dif­
fered in their subjective or experiential fa­
miliarity. The results of Experiment 2 sup­
ported this alternative hypothesis: Many of the
low-printed-frequency words used in those
previous studies, though matched f or printed
frequency, did differ substantially in rated ex­
periential familiarity. Experiment 2 also dem­
onstrated that the pattern of inconsistent find­
ings could easily be accounted for by the pat­
tern of differences in experiential familiarity
Experiments 3 and 4 provided further sup­
port for this explanation. In lexical decision
tasks, bigram frequency did not affect perfor­
miliarity) or high-familiarity words (opera­
tionalized as words with high experiential
familiarity). Experiential familiarity did sig­
nificantly affect performance, but it did not
interact with bigram frequency.
Experiment 5 investigated a similar pattern
of inconsistent findings. Some researchers had
reported that semantic concreteness facilitated
recognition of low-printed-frequency
words, whereas others had reported that it in­
terfered, and still others had reported that it
had no effect. In Experimen t 5, lexical fa­
miliarity was again operationalized as rated
experiential familiarity, and it solely affected
lexical decision RTs That is, no main effects
of or interactions with semantic concreteness
were observed. In addition, Experiment 5 cast
doubt on a previous hypothesis proposed to
explain why in at least one study, low-printed­
frequency /concrete words were recognized
better than low-printed-frequency/abstract
words. The previous elfect appeared to be bet­
ter attributable to differences in the words'
experiential familiarity and the subjects' re­
sponse criteria when making lexical decisions.
In the same vein, Experiment 6 investigated
a series of inconsistencies concerning the ef­
fects of a word's number of meanings on its
recognition. Even when polysemy was oper­
ationalized as the number of a word's defi­
nitions in an unabridged dictionary, only ex­
periential familiarity affected lexical decision
Two major conclusions can be drawn from
this series of expe riments. The first is that rated
experiential familiarity appears to be a potent
predictor of word recognition. Dependmg on
the experimental criterion of high versus low
experiential familiarity and on the use of pro­
words the effect was as great as a 250-ms dif­
in RT and an 18% difference in ac­
To provide a more precise estimate of the
relation between experiential familiarity and
word recognition and to examine the shape
of this function across the entire range of
available experiential familiarity ratings, the
following additional experiment was con­
ducted. Each of the 455 words in the corpus
was randomly placed in one of four material
mance to either low-familiarity words (oper­
ationalized as words with low-experiential fa-
2 Details of this experiment are available from the author.
sets. Including 1 1 4 pronounceable nonwords,
each material set was presented to a different
group of 1 8 subjects in a lexical decision task.
Experiential familiarity correlated strongly
with a combined measure of latency and ac­
curacy (see Gemsbacher, 1 984); specifically,
the correlations were -.86, -. 89, -. 87, and
-.78 for the four material sets. Within each
set, the same linear relation was observed: The
higher the familiarity of the stimulus word,
the quicker and more accurate were responses
to it. Across all 4 5 5 words and 72 subjects,
experiential familiarity was successful in ac­
counting for more than 7 1 % of the variance
found in performance.
This prediction compares favorably with
previous predictions of word recognition per­
formance made with the more traditional
measure of lexical familiarity, printed fre­
quency. For example, Howes and Solomon
( 1 95 I) accounted for an average SO% of the
variance found in tachistoscopic thresholds,
and Rosenzweig and Postman ( 1 956) reported
a slightly larger prediction (6 1 %) for the vari­
ance associated with auditory thresholds.
Somewhat lower predictions were reported by
Whaley ( 1 978) for lexical decision perfor­
mance (46%) and by Carroll and White ( 1 97 3 )
for picture-naming latency (39%). It is really
only the predictions made via multiple mea­
sures that rival the present 7 1 % estimate. For
example, Whaley also accounted for 7 1 % of
the variance in lexical decision, but this was
with a multiple regression based on 1 6 different
predictor variables. Thus experiential famil­
iarity is indeed a powerful single predictor of
word recognition performance.
In addition, in the experiment conducted
with all 4 5 5 words (see Footnote 2), neither
bigram frequency nor single-letter frequency
correlated significantly with performance (r
-.0 1 and - .08, respectively), further verifying
the results of Experiments 3 and 4. Regression
analyses also demonstrating a null effect of
bigram frequency on tachistoscopic recogni­
tion were reported by Johnston ( 1 97 8 ; see also
Carr, Posner, Hawkins, & Smith, 1 979). In the
present data, total number of dictionary
meanings correlated moderately with perfor­
mance (r = .25, p < .03), but when the effects
of experiential familiarity were partialed out,
this relation was reduced to insignificance
(r = -.05). Because experiential familiarity
had a substantially stronger zero-order cor­
relation, and partialing out number of mean­
ings did not significantly reduce that predic­
tion, experiential familiarity can be considered
the stronger predictor. This verifies the results
of Experiment 5 .
These additional findings support the second
major conclusion that can be drawn from the
present body of work: Previous reports of an
effect of orthographic regularity, semantic
concreteness, or polysemy on recognizing low­
printed-frequency words were most likely the
result of confounding experiential familiarity
within some level ofthese other three variables.
Acknowledging the potential unreliability
of printed frequency, several have suggested
that these probable confounds are due to
regression to the mean, that is, the statistical
probability that with a different sample of an
independent variable, the extreme points on
a normal distribution will assume a "truer"
value, one closer to the mean of that distri­
bution (see, e.g., Landauer & Freedman, 1 968).
Regression to the mean is particularly probable
when two highly correlated variables are fac­
torially combined and when the measurement
of either independent variable is noisy. Ar­
ranging groups of stimuli that are extremely
high or low along one variable and simulta­
neously extremely high or low along its co­
variate variable, and vice versa, is often done
by capitalizing on the measurement error
found in either variable. Thus, though it is
believed that the values of each variable are
well matched within either level of the opposite
variable, it is possible that their "true" values
are not. Although the measurement properties
of experiential familiarity are not completely
known at this time, experiential familiarity is
not highly correlated with the other three vari­
ables manipulated here (r -.0 1 , - .05, and
. 2 8 for bigram frequency, semantic concrete­
ness, and number of meanings, respectively).
Several experimenters have suggested that
consulting two counts of printed frequency
and selecting only those words having the same
printed frequency score in both should elim­
inate the possible confounds. Although cross­
checking printed frequency counts would pre­
vent some of the potential sampling errors,
this solution would still be inadequate.
Consider, for example, the 4 5 5 five-letter
words indexed by Thorodike and Lorge ( 1 944)
as occurring only once per million. The dis­ an experiential familiarity
rating taps this me­
tribution of their Kucera and Francis ( 1 967) morial frequency record
(cf. Attneave, 1 9 5 3).
frequency scores is, indeed, much broader
If experiential familiarity, like printed fre­
5 . 1 4 ). And this second measure of quency, is indeed
an estimate of
printed frequency correlates moderately with counters, then the two measur previous en­
es should be
both experiential familiarity ratings (r .26), highly correlated. Such appears
to be the case.
and lexical decision performance (r -.26). In still another experiment, I
randomly se­
However, across only those five-letter words lected I five-letter word at each
half-log unit
with printed frequency scores of one in both interval (according to Carroll's
, 1 970, Standard
indices (N I 02), a wide distribution of ex­ Frequency Index) from Kucera
and Francis's
periential familiarity still exists. In fact the ( 1 967) printed frequency count.
variance of experiential familiarity found be­ familiarity ratings were obtained
for each of
tween those words that both counts index as these 1 30 words. Experiential familiarit
occurring once per million does not differ sig­ related highly with printed frequency y cor­
(r .8 1 ).
nificantly from that between the words not The function relating the two was
consistently indexed, F( l , 453)
2.42, p > linear, except in the low-printed-frequency
. 1 0. Hence even after cross-checking printed range. Here the relation
was less linear. These
frequency counts, an experimenter would have data corroborate Carroll's ( 1
9 7 1 ) subjective
an equal probability of selecting words that magnitude estimates of printed
are indexed by both counts as occurring only When his data are plotted, a function
with the
once per million but that still differ in rated same shape appears.
Interestingly, half of his
experiential familiarity. Thus cross-checking subjects were professio
nal lexicographers.
printed frequency counts does not appear to Thus it appears
that the relation between
adequate solution to this confound.
printed frequency and experiential familiarity
Given that experiential familiarity is both (or subjectiv
e magnitude estimates) breaks
a robust predictor of word recognition per­ down in the
range in which printed frequency
formance and a probable source of artifact in is
considered to be the least reliable. In the
previous contradictory studies, the next logical present
article, I have argued that within this
question is, What exactly is experiential fa­
tial familiarity is the more re­
miliarity? In order to obtain experiential fa­
miliarity ratings, subjects are simply asked to liable measure.
Experiential familiarity should also be a
"rate how familiar you are with each word."
The assumption underlying the present re­ more sensitive measure of actual frequency of
search is that this instruction is a simple tool encounters. Subjective ratings of familiarity
for collecting a measure of the extent and type must obviously be more contemporary than
of previous experience a subject has had with printed frequency counts. Perhaps they also
automatically take into account the number
each word.
By extent of previous experience, I am sup­ of times the word has been spoken, written,
posing that experiential familiarity is in part, or heard, in addition to read (recall the chapter
hke printed frequency, a measure of how often vs. comma illustration). Of course, experiential
a subject has encountered a word. There is, familiarity ratings are probably affected by de­
as Hasher and Zacks ( 1 979) observed, a large mographic biases particular to the population
body of data to the effect that information of subjects from which they were obtained.
about a stimulus's frequency of occurrence is This would not be reflected in any of the data
accurately stored in memory, often indepen­ presented here because the subjects who per­
dent of its other attributes or the conditions formed each word recognition task were drawn
surrou nding its presentation (see Hintzman, from the same population that generated the
1976, for a review). Moreover, numerou s stud­ ratings. Only further investigation with more
ies have demonstrated that retrieval of stored varied subject populations will identify the ex­
frequency information is relatively facile, is tent of these potential biases. However, these
perhaps automatic, and occurs rapidly (see results do argue strongly for the use of rated
Hasher & Zacks, 1 979, for a review). So it is experiential familiarity as either a substitute
highly plausible that asking a subject to supply for or a complement to printed frequency,
the low-printed-frequency
processes in memory. Journal of Experimental Psy­
chology: General. 108. 356-388.
Hintzman, D. L. ( 1 976). Repetition and memory. In
G. H . Bower (Ed.}, The psychology of/earning and mo­
tivation (pp. 47-9 1 ). New York: Academic Press.
Howes, D. ( 1 954). On the interpretation of word frequency
Adams, M. J. ( 1979). Models ofword recognition. Cognitive
Psychology, I I . 1 33- 1 76.
Attneave, F. ( 1 953). Psychological probability as a function
of experienced frequency. Journal ofExperimental Psy·
chology, 46, 8 1 -86.
Biederman, G. B. ( 1 966). The recognition of tachisto­
as a variable affecting speed of recognition. Journal of
Experimental Psychology. 48, 1 06- 1 1 2.
Howes, D. H.,
& Solomon,
R. L. ( 1 9 5 1 ). Visual duration
thresholds as a function of word-probability. Journal of
Experimental Psychology, 41, 40 1 -4 1 0.
James, C.
( 1 975). The role of semantic information in
digram frequency. Journal of Verbal Learning and Verbal
lexical decisions. Journal of Experimental Psychology:
Human Perception and Performance, 104, 1 30- 1 3 6.
Jastrzembski, J. ( 1 98 1 ). Multiple meanings, number of
Broadbent, D. E. ( 1967). Word-frequency effect and re­
related meanings, frequency of occurrence, and the lex­
scopically presented five-letter words as a function of
Behavior. 5, 208-209.
sponse bias. Psychological Review,
Broadbent, D. E.,
& Gregory,
icon. Cognitive Psychology,
l - 1 5.
M. ( 1968). Visual perception
of words differing in letter digram frequency. Journal
of Verbal Learning and Verbal Behavior. 7, 569-57 1 .
H., Posner, M .
Hawkins, H . L.,
Jastrzembski, J.,
Stanners, R. ( 1 975). Multiple word
meanings and lexical search speed. Journal of Verbal
Learning and Verbal Behavior. 14, 534-537.
Johnston, J. C. ( 1 978). A test ofthe sophisticated guessing
M. E. ( 1 979). Perceptual flexibility in word recognition:
theory of word perception. Cognitive Psychology,
Strategies affect orthographic computation but not lex­
1 2 3- 1 53.
ical access. Journal ofExperimental Psychology: General,
Kiger, J.
1., & Glass, A. L. ( 1 98 1 ). Context effects in sentence
verification. Journal of Experimental Psychology: Hu­
Carroll, J. B. ( 1 967). On sampling from a lognormal model
& W.
man Perception and Performance. 7, 688-700.
Krueger, L. E. ( 1979). Features vs. redundancy: Comments
Francis, Computational analysis of present-day A mer·
ican English (pp. 406-424). Providence, RI: Brown
on Massaro, Venezky, and Taylor's "Orthographic reg­
ularity, positional frequency, and visual processing of
University Press.
letter strings." Journal of Experimental Psychology:
of word-frequency distribution. In H. Kucera
Carroll, J. B. ( 1 970). An alternative to Julliand's usage
coefficient for lexical frequencies, and a proposal for a
Standard Frequency Index (SF!). Computer Studies in
the Humanities and Verbal Behavior. 3, 6 ! -6 5 .
Carroll, J. B . ( 1 97 1 ) . Measurement properties of subjective
magnitude estimates of word frequency. Journal of Verbal
Learning and Verbal Behavior, 10. 722-729.
& Richman, B. ( 197 1 ). The
American heritage word frequency book. New York:
Carroll, J. B., Davies. P.,
Houghton Miflli n.
Carroll, J. B.,
White, M. N. ( 1 973). Word frequency
General, 108, 1 25 - 1 30.
Kucera, H.,
& Francis, W. N. ( 1 967). Computational anal­
ysis of present-day American English. Providence, RI:
Brown University Press.
Laming, D. R. J. ( 1 9 79). Choice reaction performance
following an error. Acta Psychologica,
T., &
Freedman, J . L. ( 1 968). I n formation re­
trieval from long-term memory: Category size and rec­
ognition time. Journal of Verbal Learning and Verbal
Behavior. 7, 29 1 -295.
T., & Streeter,
L. ( 1 973). Structural differences
and age of acquisition as determiners of picture-naming
between common and rare words: Failure of equivalence
latency. Quarterly Journal of Experimental Psychology,
assumptions for theories of word recognition. Journal
Catlin, J. ( 1 969). On the word-frequency effect. Psycho­
logical Review, 76, 504-506.
Clark, H . H. ( 1 973). The language-as-a-fixed-effect fallacy:
A critique oflanguage statistics in psychological research.
Journal of Verbal Learning and Verbal Behavior. 12.
Forster, K.
1., &
Bednall, E. S. ( 1 976). Terminating and
exhaustive search in lexical access. Memory & Cognition.
of Verbal Learning and Verbal Behavior. 12. 1 1 9- 1 3 1 .
Mason, M. ( 1 975). Reading ability and letter search time:
Effects of orthographic structure defined by single-letter
positional frequency. Journal of Experimemal Psy­
chology: General, 104, 1 46 - 1 66.
Massaro, D.
W., Taylor, G. A., Venezky, R. L., Jastr.zembski,
& Lucas, P. A. ( 1 980). Leuer and word perception:
The role of orthographic structure and visual processing
in reading. Amsterdam: North-Holland.
J. E.,
Gernsbacher, M . A . ( 1983, April). The experiential fa­
& Taylor, G. A. ( 1 979a).
Orthographic regularity, positional frequency, and visual
miliarity norms and their psychological reality Paper
processing of letter strings. Journal of Experimental
53-6 1 .
presented at the 29th Annual Meeting of the South­
Massaro, D. W., Venezky, R. L.,
Psychology: General, 108, 1 07- 1 24.
Gernsbacher, M. A. ( 1 984 ). On the use of a "new " per­
W., Venezky, R. L., & Taylor, G. A. ( l 979b).
Orthographic regularity, positional frequency, and visual
formance variable to measure cognitive processing.
processing of letter strings: A reply to Krueger's com­
Manuscript submitted for publication.
ments. Journal of Experimental Psychology: General.
western Psychological Association, San Antonio, TX.
Glanzer, M.,
& Bowles, N.
( 1 976). Analysis of the word­
frequency effect in recognition memory. Journal ofEx­
perimental Psychology: Human Learning and Memory,
2 1 -3 1 .
Hasher, L.,
& Zacks, R . T. ( 1 979). Automatic and effortful
Massaro, D.
1 3 1 - 1 32.
McClelland, J. L.,
Johnston, J. C. ( 1 977). The role of
familiar units in the perception of words and nonwords.
Perception & Psychophysics. 22. 249-26 1 .
McClelland, J. L . ,
Rumelhart, D . E . ( 1 9 8 1 ) . A n inter-
action activation model of visual perception. Psycho­
logical Review, 88, 3 1 5-40 1 .
Morton, J . ( 1968). A retest of the response-bias explanation
of the word frequency effect. British Journal of Math­
ematical and Statistical Psychology, 21, 2 1 -22.
Neisser, U. ( 1 967). Cognitive psychology. New York: Ap­
Newbigging, P. L. ( 196 1 ). The perceptual reintegration of
frequent and infrequent words. Canadian Journal of
Psycholog}: 15, 1 23 - 1 32.
Norman, D. A. ( 1 9 8 1 ). Categorization of action slips. Psy­
chological Review, 88. 1 - 1 5.
Paivio. A.,
& Psychophysics, 8, 273-275.
Paivio, A., Yuille, J., & Madigan, S. ( 1 968). Concreteness,
imagery, and meaningfulness values for 925 nouns.
Journal of Experimental Psychology Monograph, 76( 1 ,
Pt . 2).
preliminary taxonomy for some errors in choice RT
chological Review, 85, 59- 1 0 8 .
Reder, L., Anderson, J.,
& Bjork,
R. A. ( 1 974). A semantic
perimental Psychology, 1 02. 648-656.
& Robinson, D. 0. ( 1 975). The role ofbigram
frequency in perception of words and nonwords. Memory
5 1 3-5 1 8.
recognition. American Journal of Psychology,
Postman, L. ( 1 956). Intelligibility
as a function of frequency of usage. Journal of Exper­
imental Psychology, 54, 4 1 2-422.
Rubenstein, H., Garfield, L.,
the perception of speech. Journal of the Acoustical So­
ciety of America, 35. 200-206.
Scarborough, D. L., Cortese, C.,
Scarborough, H. S.
( 1 977). Frequency and repetition effects in lexical mem­
ception and Performance, 3, l - 1 7.
Schnorr, J. A.,
Atkinson, R. C. ( 1 970). Study position
and item difficulty in the short- and long-term retention
of paired associated learned by imagery. Journal of Verbal
Learning and Verbal Behavior. 9, 6 1 4-622.
Shannon, C. E. ( 1 948). A mathematical theory of com­
Verbal Learning and Verbal Behavior. 9, 487-494.
Rubenstein, H., Lewis, S. S.,
Rubenstein, M. ( 1 97 1 a).
Evidence for phonemic coding in visual word recog­
nition. Journal of Verbal Learning and Verbal Behavior.
Rubenstein, H., Lewis, S. S.,
Solomon, R. L., & Postman, L. ( 1 952). Frequency of usage
Journal of Experimental Psychology, 43, 1 95-20 1 .
Extensions of Dander's method. Acta Psycho/ogica,
276-3 1 5 .
Thorndike, E., & Lorge, I . ( 1 944). The teacher's word book
of 30,000 words. New York: (:olumbia University,
Teacher's College Press.
Rubenstein, M. ( 1 97 l b).
Homographic entries in the internal lexicon: Effects of
systematicity and relative frequency of meanings. Jour­
nal of Verbal Learning and Verbal Behavior, 10. 57-62.
Rubenstein, H . , Richter, M. L.,
ments on the papers by J. Catlin and L. H. Nakatani.
Psychological Review, 78, 420-425.
Webster's new collegiate dictionary (desk ed.) ( 1 976).
Springfield, MA: G.
Kay, E. J. ( 1 975). Pro-
C. Merriam.
Webster's new world dictionary ( unabridged) ( 198 1 ).
Springfield, MA: G.
Millikan, J. ( 1 9 70). Ho­
mographic entries in the internal lexicon. Journal of
ical Review. 81. 99- 1 1 8 .
Savin, H. B . ( 1 963). Word-frequency effect and errors in
Treisman, M. ( 1 9 7 1 ). On the word frequency effect: Com­
Richards, L. G. ( 1 976). Concreteness as a variable in word
7 1 8.
Siple, P. ( 1 974). Processes of rec­
Sternberg, S . ( 1 969). The discovery of processing stages:
interpretation of encoding specificity. Journal of Ex­
Rosenzweig, M. R.,
ognizing tachistoscopically presented words. Psycholog­
as a determinant of recognition thresholds for words.
Ratcliff, R. ( 1 978). A theory of memory retrieval. Psy­
& Cognition, 3,
65 1 -657.
munication. The Bell System Technical Journal,
Vayas, S. M. ( 1 970). An elementary
tasks. Acta Psychologica.
Rice, G. A.,
Rumelhart, D. E.,
ory. Journal of Experimental Psychology: Human Per­
O'Neill, B. J. ( 1 970). Visual recognition
thresholds and dimensions of word meaning. Perception
Rabbitt, P. M. A.,
nounceability and the visual recognition of nonsense
words. Journal of Verbal Learning and Verbal Behavior.
C. Merriam.
Whaley, C. P. ( 1978). Word-nonword classification times.
Journal of Verbal Learning and Verbal Behavior. 1 7,
143- 1 54.
White, M. J. ( 1 983). Prominent publications in cognitive
psychology. Memory & Cognition. 11, 423-427.
Winer, B. J. ( 1 9 7 1 ). Statistical principles in experimental
design (2nd ed.). New York: McGraw-Hill.
Winnick, W. A.,
& Kresse!, K. ( 1965).
Tachistoscopic rec­
ognition thresholds, paired-associate learning, and im­
mediate recall as a function of abstractness-concreteness
and word frequency. Journal of Experimental Psychol­
ogy, 70, 1 6 3- 1 68.
(Appendix follows)
High Familiarity/High Bigram
SMICE (7,838) CHEGG (9,453) BEFNG (4,929)
THRIM ( 10,735) PCHER (9,066) JOMER (9,4 38)
WHIBB ( 1 2,405) LARDT (3,535) SHEHT ( 1 2,5 75)
BOXER (8,239) CHORE (8,262) MIXER (8,862)
BELCH (6,509) BATCH (6,94 1 ) BOOST ( 5 , 1 86)
RACER (8,567) RHINO (9,686) LEACH ( 1 1 ,032)
CHESS ( 1 2,3 1 4) CIDER (9,578)
High Familiarity/Low Bigram
QNEND ( 1 ,96 1 ) JGFDS ( 1 ,736) BQRIL (977)
PSFTU ( 1 22) ERTIY (670) JOUIR ( I ,5 1 7) FRTTL
(599) LABHE ( 1 ,568) FYTCK ( 1 ,274) CINSS
( 1 ,392) PYRLT ( 1 ,568) MRAON ( 1 ,69 1 ) OTTYE
( 1 ,73 1 ) BRLAE ( 1 ,097) HRNIO (232) MTHRU
( 1 ,723) BOAUG ( 1 ,249) FLMAT ( 1 , 7 1 8) PIOSP
MNRTI ( 1 ,028) WERFD (695) PRRYT (679)
( 1 3 1 ) DERFV ( 1 ,0 1 6) PIUYT (297)
High Familiarity/One Dictionary Meaning
High Familiarity/Semantically A bstract
Low Familiarity/Many Dictionary Meanings
Low Familiarity/Semantically Concrete
DENIM ( 1 ,358) BULKY ( 8 1 0) KODAK (977)
DUIOP ( 1 , 1 05)
POKIL ( 1 ,805) FRTUI
SOAPY ( 1 ,543) SISSY ( 1 ,296) JERKY (85 1 )
TULIP (676) PUFFY (62) FUNGI (887) SUEDE
( 1 , 1 1 2) ULTRA (237) BAGGY ( 1 ,027) BLUFF
( 1 ,034) EXCEL ( 1 ,090)
YXEDF (248) TYIUR (258) JIKMR (93) SEDCF
(984) NIUTY ( 1 ,674) MOOHF ( 1 ,899)
Nonwords (Experiment 4)/High Bigram
Low Familiarity/High Bigram
HATER ( 1 2,535) SHIRE ( 1 2,346) FATED (8,03 1 )
AUGER (7,752) ADDER (9,679) SHEBA ( 1 0, 1 66)
TRICE (9,677) ROWER (9,044) CORSE ( 1 1 ,947)
ABASE (7,9 1 2) ASTER ( 1 0,946) TERSE (7,556)
BREST (7,22 1 ) MANSE (7,032) FIBRE (6,653)
BRINE (6,430) GUISE (6,0 3 1 ) STOKE (5,388)
ABRLD (9, 1 85) WHSIR (9,636) KFRSE (6,438)
GLTES (7,29 1 ) MHITD (7,426) SHRRE (6,887)
XRRES (5,059) QRASE (7,039) HLARE (8, 802)
DOUFC (8,099) THGIY (9,83 1 ) TIERS (6, 1 35)
WHREF ( 8 , 1 65) CHOSE (5,564) FHIYT (6,209)
GHIIP (6,258) ZSERT (5,350) NHITY (8,0 1 7)
GLINT (5,223) CIRCE (5,206)
DSERE (8, 5 1 0) GHNER (8,222) FSTER ( 1 0,475)
CHTER ( 1 1 ,400) MOUPF (8,885) GFDER (9, 1 60)
Low Familiarity/Low Bigram
PCHER (9,066) BHTER ( 1 0,475) IOUGE (9,357)
WHUSQ (7, 1 4 1 ) DMICE (7, 1 20) RBDER (9, 3 1 2)
AGAPE ( 1 , 520) DEIFY ( 1 ,484) MOGUL ( 1 ,425)
AGLOW ( 1 ,400) TABBY ( 1 ,365) BURKE ( 1 ,280)
AFFIX ( 1 , 1 80) TEMPS ( 1 , 1 82) ALLAH ( 1 , 1 3 5)
WHIXB ( 1 2,405) AOUNG ( 1 3,564) NCDER
(9, 1 60) SHPER (9,56 1 ) NLANT (4,690) SHETD
( 1 2,575) SHEBT (9, 1 66) CHRTE (8,770) STKEO
ASSAY ( 1 , 1 1 9) DITTY ( 1 ,025) DELHI ( 1 ,006)
(5,876) TRCKE (7,324)
FLUME (926) REFIT (852) ADEPT (840) SAVOY
(832) VOLGA (803) GNOME (752) TWIXT (287)
BYLAW ( 1 5 5 )
Nonwords (Experiment 4)/Low Bigram
Nonwords (Experiment 3)/High Bigram
PSFTU ( 1 22) DWNIS (578) ALWRT ( 1 ,872)
FRTTL (599) MIPWS ( 1 ,446) WSADE ( 1 , 1 23)
QNEND ( 1 ,96 1 ) JGFDS ( 1 ,736) HQRIL (977)
OTHEI (5, 1 32) ABRLD (9, 1 85) WHSIR (9,636)
BVIRT ( 1 ,332) BRTTY ( 1 ,633) MRAON ( 1 ,092)
WOZUT (6,565) GLTES (7,29 1 ) GGHER (9,063)
ALANG (7,508) ABDER ( 1 1 , 3 1 1 ) WOVEZ (7,702)
VIWRS ( 1 ,7 1 0) BRLAE ( 1 ,097) HRNIO (232)
MTHRU (760) DHRMU ( 1 , 1 2 1 ) FLMAT ( 1 ,7 1 8)
LIJHC (958) NOKLJ (604) TYUIP ( 1 ,0 1 5) RWQIO
STISH (5,305) GOUSE ( 1 3,792) QRASE (7,039)
HLARE (8,802) ZIXER (7,7 53) YOUGE (9,357)
ASDER (9,649) TIERS ,(6, 1 35 ) WHREF (8, 1 65 )
CHOSE (5,664) FASRE (5,669) GHIIP (6,258)
( 1 56) MNRTI ( 1 ,028) QUDDC (659) PRRYT (8 10)
(2 1 7)
( 1 ,0 1 6)
PIUYT (297)
ZSERT (5,350) IKMER (8,508) DSERE (8, 5 1 0)
DUIOP ( 1 , 1 05) GIVVM ( 1 , 7 1 8) FRTUI (549)
PLCNE ( 1 ,729) IKKLP (793) SHTUY ( 1 ,663)
GHNER (8,222) FSTER ( 10,475) CHTER
YXEDF (724) TYUIR (258) JIKMR (93) SEDCF
( I I ,400)
WHUSY (7, 1 4 1 ) TROUN ( 5 ,5 3 1 ) FRARE (8,646)
(984) NIUTY ( 1 ,674) SHTUY ( 1 ,663)
BANJO ( 1 ,739) ALBUM (948) BURNS ( 1 ,632)
( 1 ,729)
FAGOT ( I I ) GOUGE ( 1 4) LEACH ( 1 0) STALE
(24) SUPER ( 1 2) WAVER ( 1 6)
Nonwords (Experiment 3)/Low Bigram
( I ,965) AGHIX (720) TYUIP (287) RWQIO ( 1 56)
BOOZE ( 1 ,383) FUDGE ( 1 ,527) AMAZE ( 1 , 1 8 1 )
SUPER (8,03 1 ) CHILI (9,4 1 3) ROACH (7,225)
LOUSY (9,457) STALE (8,309) PRONG (7,362)
( 1 2) CLACK ( 1 0) CLAMP ( 1 2 ) CLOUT ( 1 4)
POUNG ( 1 3,564)
ICING (6,684) JOKER (8,702) ULCER (7,543)
BULGE ( 1 8) CADET ( 1 5) CANNY ( 1 4) CINCH
Experiment 5
High Familiarity/Semantically Concrete
Stimuli Used in Experiments 3-6
Experiments 3 and 4 (Summed bigram
frequencies are shown in parentheses)
Low Familiarity/Semantically Abstract
Experiment 6 (Number of dictionary definitions
is shown in parentheses)
( 1 5) BANDY ( I I ) BERTH ( 1 5 ) BRAWN ( I I )
( I I ) SHUCK ( 1 2) SHUNT ( 1 4) SOUSE ( 1 7)
(9) FIBRE ( 1 4) FLAIL ( I I ) FLECK
TABBY ( 1 0) THRUM ( 1 3)
Low Familiarity/One Dictionary Meaning
High Familiarity/Many Dictionary Meanings
ANNEX ( 1 6) BATON ( 1 2) BELCH ( 1 0) BLARE
( 1 2) BLOAT ( 1 0) BOOST ( 1 3) BRAWL ( 1 0)
Received February
I I , 1983
Revision received October 1 6 , 1 98 3 •
Interested in the Journal Review Process?
Reviewers are needed for all APA journals. Individuals who have not regularly par­
ticipated in the review process are encouraged to submit a letter of interest to the
appropriate editor of the journal of their choice. Women and minorities are especially
invited to submit letters of interest. A vita or description of relevant background
should be included with the letter. The names and addresses of journal editors are
published annually in the June issue of the American Psychologist or may be obtained
by contacting Arlene Stephens, American Psychological Association, 1 200 1 7th Street,
N. W., Washington, DC 20036. Also, each journal publishes the name and address of
its journal editor on the inside front cover.

Similar documents


Report this document