Memory & Cognition

/993, 2/ (5), 7//-7/9

Conceptual metaphors are not automatically
accessed during idiom comprehension
SAM GLUCKSBERG, MARY BROWN, and MATI'HEW S. McGLONE
Princeton University, Princeton, New Jersey
Do conceptual analogies motivate idiom use and comprehension in discourse? For example, a
story in which a person is described as fuming would be analogically consistent with an idiom
such as blew her top, but inconsistent with an idiom such as bite his head off. Earlier work by
Nayak and Gibbs (1990) had suggested that people use such analogical information during idiom
comprehension, We replicated their findings in an idiom choice task, suggesting that people can
indeed make use of such knowledge. However, when reading times were used to assess idiom
comprehensibility, no effects of analogical consistency were found, We conclude that conceptual
analogies play little, if any, role in idiom comprehension unless people have the time (and moti-
vation) to make considered judgments.
711
Idioms are typically described as frozen phrases whose
meanings are stipulated directly in a mental lexicon. Thus,
the meaning of a phrase such as to kick the bucket would
be represented in an idiom or phrasal lexicon simply as
"to die" (Bobrow & Bell, 1973; Swinney & Cutler,
1979), Many idioms, however, are more complex and dy-
namic. Consider, for example, idioms such as carrying
coals to Newcastle or locking the bam door after the
horses have been stolen. Such idioms may well have
meanings stipulated in a phrasal lexicon, but they are also
recognized as alluding to prototypical instances of the con-
cepts that they refer to (Glucksberg, in press). Thus, these
idioms are fully compositional in that they are both se-
mantically and syntactically productive. One can, for ex-
ample, say, in reference to the savings and loan scandal
of this decade, "they never did lock the barn door until
after the barn was stolen." Such novel and productive
twists on idioms are understood easily by anyone famil-
iar with the original idioms (Glucksberg, 1991; McGlone,
Glucksberg, & Cacciari, in press).
Another class of idiomatic expressions may also require
fairly complex representational assumptions. Expressions
for many abstract concepts seem to cluster in terms of
underlying conceptual analogies. Thus, we speak of time
in spatial terms, of relationships with others as journeys,
of successand failure as directionsof movement (Kovecses,
1986; Traugott, 1975, 1985), Many such abstract con-
cepts can be conceptualized in systematically different
ways. Anger, for example, can be conceptualized in any
one of several specific ways (Lakoff, 1987; Lakoff &
We are grateful for the financial support provided by Public Health
Service Grant HD 25826 to Princeton University. We also thank Ray
Gibbs for providing us with both his insights and his materials, and Su-
san Fussell, Boaz Keysar, Deanna Manfredi, and Sachi Kumon-
Nakamura, for their valuable comments and suggestions. Correspon-
dence should be sent to S. Glucksberg, Department of Psychology,
Princeton University, Princeton, NJ 08544-1010.
Johnson, 1980), One conceptual metaphor for anger is
that of a heated fluid under pressure. Idioms that seem
to reflect this conceptual metaphor include flip your lid,
let offsteam, blow your top, and get hot under the collar.
An alternative conceptual metaphor for anger is that of
animal-like behavior, reflected in such idioms as bite
someone's head offandjump down someone's throat. Al-
though broad conceptual metaphors or analogies seem to
motivate many idiomatic expressions, their functional role
in idiom use and comprehension remains unclear. When
people encounter an idiom such as blow his top in a con-
versation, is the conceptual analogy of anger as heated
fluid under pressure (1) available or (2) accessible? By
availability, we mean that a conceptual structure is rep-
resented in semantic or long-term memory and could be
retrieved under some, but not all, circumstances. By
accessibility, we mean that a conceptual structure not only
is available, but also can be accessed in a particular con-
text to participate in either the production or the com-
prehension processes (for a fuller discussion of the
availability-accessibility distinction, see Higgins, Rholes,
& Jones, 1977; Srull & Wyer, 1979).
By definition, the availability of a conceptual structure
is context independent: it is either stored in memory or
not. Again by definition, the accessibility of any specific
conceptual structure is context dependent. Any given item
in memory may be accessible in one context, but not ac-
cessible in another. For example, if there is a conceptual
analogy of anger-as-fluid-under-pressure in semantic
memory, this conceptual structure is available. It may be
accessible in some circumstances, but not in others. Thus,
it might be accessible and used when people have the time
to make considered, deliberate judgments. It might, how-
ever, be inaccessible and therefore not used in ongoing
speech comprehension and production when people do not
have the time for such judgments. In the work to be pre-
sented here, Experiment I asks whether people access
conceptual analogues when making considered judgments
Copyright 1993 Psychonomic Society, Inc.
712 GLUCKSBERG, BROWN, AND McGLONE
about idioms. Experiments 2 and 3 ask whether people
have [automatic] access to such analogical information
under conditions of speeded reading for comprehension.
There is some evidence that people do use conceptual
analogical information when the situation is conducive to
deliberate judgments. Nayak and Gibbs (1990, Experi-
ment 6) asked college students to judge the appropriate-
ness of idioms in specific contexts. The students were
given short narratives such as the following (emphases
added):
Mary was very tense about this evening's dinner party. The
fact that Bob had not come home to help was making her
fume. She was getting hotter with every passing minute.
Dinner would not be ready before the guests arrived. As
it got closer to five o'clock the pressure was really build-
ing up. Mary's tolerance was reaching its limits. When Bob
strolled in at ten minutes to five whistling and smiling,
Mary ...
blew her top.
bit his head off.
Inthis story, the protagonist's (Mary's) anger is described
in terms of increasing pressure and heat-"making her
fume," "getting hotter," "pressure was really building
up," and so forth. The story is thus stylistically consis-
tent with idioms that could instantiate the concept of anger
as heated fluid under pressure-in this case, blew her top.
The following scenario was constructed to be consistent
with an alternative conception of anger, that of animal-
like behavior (emphases added):
Mary was getting very grouchy about this evening's din-
ner party. She prowled around the house waiting for Bob
to come home to help. She was growling under her breath
about Bob's lateness. Her mood was becoming more savage
with every passing minute. As it got closer to five o'clock,
Mary was ferociously angry with Bob. When Bob strolled
in at 4:30 whistling and smiling, Mary ...
blew her top.
bit his head off.
Here, the description of Mary's behavior in bestial
terms- "prowled," "growling," "savage," and so
forth-is consistent with idioms that instantiate animal-
like behavior, such as bite his head off.
Consistent with the hypothesis that idiomatic concep-
tual analogies are available and accessible when people
have the time to make deliberate judgments, the students
rated analogically consistent idioms as more appropriate
than comparable but analogically inconsistent idioms.
Thus, blew her top was rated as more appropriate as a
completion when anger was described in heat and pres-
sure terms, and bit his head offas more appropriate when
anger was described in bestial terms.
On the basis of this finding, Nayak and Gibbs (1990)
concluded that readers not only have relevant conceptual
information available, but also use this information to
facilitate idiom comprehension. The differences in ap-
propriateness ratings are taken to reflect the relative dif-
ficulty subjects had in interpreting the competing idiom
completions. Idioms in story contexts that were matched
for analogical information were considered easier to in-
terpret than idioms in contexts using a different concep-
tual analogy. The appropriateness ratings, on this account,
directly reflected on ease of interpretation.
There are, however, at least two additional competing
interpretations of these data. First, the appropriateness
ratings may not be the product of ease of comprehension
at all, but rather the outcome of postcomprehension de-
cision and judgment processes. After all, people might
readily notice the relation between pressure words in a
text and the semantic content of an idiom such as blow
one's top. If this were the case, the data might simply
reflect the ability of people to recognize a relation between
themes in a text. If the subjects in the Nayak and Gibbs
experiment did recognize relations between textual ele-
ments and the idiom choices, their choices may have been
based simplyon a preference for stylisticconsistency. Sec-
ond, the data may not even implicate deliberate choices
at all. As Kreuz and Graesser (1991) have pointed out,
the ratings data may be entirely attributable to simple lex-
ical priming rather than stylisticconsistencyper se. Words
such as prowled, growled, and savage are semantically
associated with the word bite, as in the idiombite his head
off. There is substantial evidence that inferences that
readers draw during text comprehension can be strongly
influenced by lexical priming (e.g., Ratcliff & McKoon,
1986; Potts, Keenan, & Golding, 1988). Thus, even if
there were no conceptual analogy in semantic memory
underlying the meaning of the idiom bite ones' head off,
the relationship between the words in a text and the words
in an idiom could influence subjects' appropriateness
ratings.
Experiment 1 was designed to control for simple lexi-
cal priming by varying the referent of the target idioms.
In the original Nayak and Gibbs (1990) materials, only
the central protagonist in a story could be the referent of
target idioms. We used these original materials as part
of a replicationof the Nayak and Gibbs study, but we also
used two other item types. The first was an other-person
referent version of the original story, in which the hypo-
thetical conceptual analogy information referred to the
original protagonist, but the target idioms referred not to
that protagonist, but instead to some other person. Thus,
if Mary were to be described as fuming in a particular
scenario, the target idiom blow one's top would refer to
another person, Chuck, in that scenario. If idiomappropri-
ateness judgments are simply a function of which kinds
of idiomatically relevant concepts are most accessible in
a given context, the specific referent of the target idiom
should make no difference. Thus, if anyone in a story is
described infuming-anger terms, anyone in that context
would tend also to be described in those terms. To con-
trol for lexical priming per se, we used a third type of
scenario, in which a situation or state of affairs, rather
than any person, was described with words relevant to
an idiomatic conceptual analogy-for example, the wind
"roared." Such nonperson descriptions should not evoke
any specific emotion concept such as anger, and so they
should not elicit analogically consistent idiomjudgments,
unless the lexical items themselves prime such judgments.
The target idioms referred invariably to a person in the
scenario-for example, she bit his head off. If lexical
priming were the only effective variable, subjects' judg-
ments of idiom-story appropriateness should not be af-
fected by an idiom's referent. Even when events or situ-
ations are described, say, in animalistic terms, subjects
shouldjudge that animal-consistent anger idioms are more
appropriate than idioms based on other metaphors for
anger.
To summarize the logic of Experiment 1, if judgments
of idiom appropriateness are based on the specific emo-
tion attributed to a specific protagonist in a given con-
text, analogically consistent idioms should be chosen as
most appropriate only in the original-person referent con-
dition. If people base their choices instead on the basis
of which emotion concepts are most accessible in mem-
ory in a given context, there should be no difference be-
tween the original- and the other-person referent condi-
tions. Either of these two patterns of results would be
consistent with the claim that people can use conceptual
analogical information for idiom interpretation when such
information is available. However, if readers' idiom pref-
erences can simply be primed by the words in a story con-
text, the same idiom preferences should be exhibited in
all three types of stories. People should, for example,
choose animal-behavior idioms to describe an angry per-
son even when the story uses a word such as roarto refer
not to anger, but instead to the sound of the wind. The
latter result would support Kreuz and Graesser's (1991)
contention that Nayak and Gibbs's findings are simply an
artifact of lexical priming.
EXPERIMENT 1
Idiom Choice as a Function
of Analogical Consistency
Method
Subjects. Twenty-four undergraduate students at Princeton Uni-
versity were paid for their voluntary participation. All were native
English speakers, and none had participated previously in studies
of figurative language.
Materials and Procedure. Nine idiom pairs, two each referring
to the concepts of anger, fear, success, and failure were chosen
as targets for appropriateness judgments. Each pair of idioms in-
stantiated two alternative conceptual analogies for the concept of
interest. For example, the idioms flip your lid and bite someone's
head off presumably reflect two alternative metaphors for anger:
heated fluid under pressure and animal behavior, respectively. The
18 context stories used by Nayak and Gibbs (1990, Experiment 6)
were adapted for this study. Each story was written in two ver-
sions, one to (theoretically) activate one conceptual analogy, the
other to activate the alternative analogy (see examples, Appendix A).
This yields 36 story-idiom pair items. Three referent versions of
each story were then prepared: (I) An original-referent version,
in which the target idiom would refer directly to the person in the
story who was described in terms analogically consistent with just
one of the two alternative idioms. This version was essentially iden-
tical to the stories used by Nayak and Gibbs (1990); (2) An other-
person referent version, in which the target idiom referred to some
IDIOM COMPREHENSION 713
person other than the one described in the story as conceptually
consistent with one of the two alternative idioms. This version con-
trols for specificity of the conceptual analogue. If the conceptual
analogue for, say, anger as heated fluid under pressure is specifi-
cally associated with person A in a story, but the final idiomatic
statement refers to person B, there should be no reason to choose
a pressure-anger idiom over an animal-behavior-anger idiom. If,
however, idioms are chosen on the basis of whatever idiomatic ana-
logue is activated during the course of the story, pressure-anger
idioms should be chosen, as in the original-referent condition.
(3) The nonperson referent version, a condition to control for lexi-
cal priming per se. In this version, the words that were associated
with the conceptual analogue of a target idiom referred to a non-
person entity in the story, such as the weather, as in the windroared.
In all three versions, the target idioms always referred to a person.
Examples of these materials are presented in Appendix A.
The 36 story-idiom pairs each appeared in the three referent ver-
sions to provide a total of 108 experimental items. In addition to
modifying the materials to provide the three different referent con-
ditions, we also modified the experimental task. Nayak and Gibbs
(1990) obtained ratings of idiom appropriateness. We decided to
use a measure that more closely reflected discourse production pro-
cesses; at the end of each story, the subjects were given a choice
of two idioms and were asked to choose the idiom that best com-
pleted the story. In all cases, either idiom provided a sensible com-
pletion; for example, subjects could choose between flipped her lid
and jumped down his throat to describe a protagonist's anger. In
addition to providing a measure, albeit indirect, of idiom produc-
tion, this alternative measure also provides a converging operation
that could extend the generality of Nayak and Gibbs's findings.
The 108 experimental items were divided into six sets of 18 each.
The sets were counterbalanced so that only one of the factor com-
binations for a story and idiom-choice combination appeared in a
given set. The sets contained equal numbers of original-, other-
person, and nonperson referent stories. Each set of 18 experimen-
tal items was presented in a booklet along with 52 filler items, for
a total of 70 items per booklet. The fillers, like the experimental
items, were story and idiom-choice combinations, but the stories
provided no apparent cues to guide idiom choice.
The subjects were randomly assigned to one of the six item book-
lets. They were instructed to read each story carefully andto choose
the idiom they considered to be the more appropriate completion
for each story. The subjects circled their choices directly in the ques-
tionnaire booklet. This procedure took about I h, after which the
subjects were debriefed about the purpose of the experiment.
Results and Discussion
The data of interest are the proportions of conceptu-
ally consistent idiom choices as a function of referent con-
dition (see Figure 1). In the original-referent condition,
the analogically consistent idiom was chosen 69.4% of
the time. This percentage is significantly greater than
chance (p < .01, binomial distribution). This finding
replicates Nayak and Gibbs (1990), even though we used
a different dependent measure-namely, forced choice in-
stead of appropriateness ratings. In the other-person refer-
ent condition, 6 of the 24 subjects apparently noticed that
the topic of the story had been abruptly switched, and on
seven occasions they substituted the original-person refer-
ent for the switched-person referent by crossing out the
other-person referent's name and writing in the original
protagonist's name. On five other occasions (one occa-
sion for 5 ofthese 6 subjects), either no choice was indi-
cated, or the subject wrote "neither makes sense." The
714 GLUCKSBERG, BROWN, AND McGLONE
7S
......

70
-
c
6S
0
'zj
U
QJ
60
Qj
V'l
E
SS
0
j2
oJ
QJ
Cl
...
(l)
....
o
Referent Condit ion
Figure 1. Idiom selection as a function of referent condition.
choice data for these 11 occasions were discarded from
the analysis. The corrected data, as shown in Figure 1,
indicate a preponderance of analogicallyconsistent choices
(60.2 %) in the other-person referent condition (p < .05,
binomial distribution). These data strongly suggest that
our subjects were generally sensitive to analogical con-
sistency when a person in a story was described in idio-
matically relevant terms, but that it did not matter much
which person this was. This effect, however, is not sim-
ply attributable to lexical priming, because idiom choice
was at chance level in the nonperson referent condition
(50.93% analogically consistent). It seems that if a par-
ticular conceptual analogy is activated in a given context,
idioms consistent with that analogy will be preferred when
both the story context and the idiom involve a person
referent. However, when the story context involves a non-
person referent and the idiom itself refers to a person,
idiom choice is not affected.
This interpretation is consistent with an analysis of vari-
ance (ANOVA) of the choice proportions. A one-way
repeated measures ANOVA, with both subjects (Fs) and
items (Fi) as random factors, revealed a significant effect
of referent condition [Fs(2,46) = 5.42, p < .01;
Fi(2,54) = 3.72, p < .05]. Neuman-Keuls tests revealed
that the proportions of analogically consistent choices dif-
fered between all three conditions (p < .05) for the sub-
jects analysis. In the items analysis, the only significant
difference occurred between the original-person and non-
person referent conditions.
The most parsimonious interpretation of these data is
that the conceptual analogues for an idiom are accessed
only when the appropriate idiomatic concept itself (e.g.,
anger) is activated. Thus, when a person is described as
fuming, anger idioms that are consistent with heated fluid
under pressure, such as blow one's lop, are preferred to
inconsistent idioms. The particular person or persons in-
volved do not seem to matter all that much, suggesting
that the effective mediating factor is the general kind of
idiom involved, not its specific referents. In contrast,
when words such as fuming are used to describe nonper-
son entities, they presumably do not activate person-
relevant concepts such as anger. This in turn implies that
the choice patterns in our person referent conditions were
not an artifact of lexical priming per se.
Conceptual analogical information thus seems to be ac-
cessible, at least when people make considered, non-
speeded judgments or choices. Is such information acces-
sible during relatively unreflective comprehension? More
specifically, is such information automatically activated
when people are not asked for deliberate judgments but
insteadsimply read for comprehension? Experiment 2 was
designed to address this issue. If an idiom's hypothesized
conceptual analogy is automatically accessed during com-
prehension, comprehension should be facilitated when an
idiomis analogicallyconsistent with prior text, as opposed
to when an idiom is analogically inconsistent with that
text. In Experiment 2, we used a speeded task rather than
a reflective judgment task, and reading time was used to
assess whether analogically consistent text-idiom relations
facilitate comprehension.
EXPERIMENT 2
Does Analogical Consistency
Facilitate Idiom Comprehension?
Method
Subjects. Thirty-two Princeton undergraduates served as paid
volunteers. All were native English speakers, and none had previ-
ously participated in an experiment involving figurative language.
Materials and Design. The original-referent and other-person
referent items of Experiment I were adapted for this experiment.
Stimulus stories were created by combining original-referent and
other-person referent versions of each story with one member of
a given idiom pair. The particular idiom could be analogically con-
sistent or inconsistent with the story context. This yielded four story
conditions for each idiom. The context stories were edited so that
they could be presented line by line on a computer screen. The sen-
tence containing the target idiomatic phrase appeared either at the
end of the story or as the next to the last sentence in the story (for
examples, see Appendix B).
These materials were used in a 2 x 2 within-subjects design. The
two factors were referent version (original vs. other-person) and
analogical consistency between story and idiom (consistent vs, in-
consistent). On the basis of the results of Experiment I, we would
expect that analogically consistent idioms would be read more
quickly than inconsistent idioms irrespectively of referent condi-
tion, because specific kinds of anger should be activated in both
conditions. The original- versus other-person referent comparison
serves as a manipulation check, because switching the topic or fo-
cus of a story in midstream should disrupt performance (Garrod
& Sanford, 1988). Garrod and Sanford found that reading times
were slowed when a story topic shifted in a text. If reading times
in our task are not affected by a switch from original- to other-
person referents, this would indicate that our dependent measure
(reading time) is simply not sensitive enough to detect differences
in comprehension difficulty. The nonperson referent condition that
we used in Experiment I to control for lexical priming per se did
not lead to analogically consistent choices, so we elected not to use
this condition in Experiment 2, in order to maximize the likelihood
of finding an effect of analogical consistency.
Sixty-four stories (\6 idioms x 4 story conditions) were randomly
assigned to one of four lists, so that only one of the story condi-
tions for a particular idiom appeared on a given list. Each list con-
tained 4 items of each story type for a total of 16experimental items.
Interspersed among the experimental stories were 16 filler stories,
3 of which appeared as practice items. These stories were adapted
from the fillers of Experiment 2 by randomly selecting one mem-
ber of the original idiom pair for each filler story. As with the ex-
perimental items, the sentence containing the idiomatic phrase ap-
peared either as the last or the next to the last sentence of the filler
stories.
Procedure. The subjects were randomly assigned to one of the
four counterbalanced lists, with the constraint that all lists were used
equally often. The stories were presented in a different randomorder
for each subject. The subjects were seated in front of a Samsung
MA 2565 monochrome monitor at a comfortable reading distance,
with the right hand resting on the" +" key of the numeric keypad
of an Epson Equity I + computer. The stories were presented line
by line at the center of the screen in standard upper- and lowercase
type.
The subjects were given the following instructions:
In this experiment, you will read a number of very short stories, each
about a paragraph long. These stories will appear on the screen one
sentence at a time, at a pace that you will determine. When a line ap-
pears, read it, and when you are ready for the next line, press the gray
plus (+) key on the extreme right side of the keyboard. You should
try to read the stories at a quick pace, but do not move to the next
sentence until you have understood the sentence currently on the screen.
Once you move to a new sentence, you will not be able to backtrack
to any previous sentences.
The end of each story was signaled by the words "END OF
STORY" following the last sentence. To advance to the next story,
the subjects again pressed the "+" key. Reading times for each
line were measured from the onset of the line to the keypress for
advancing the next line.
Following four of the filler stories, true/false questions concern-
ing the content of the preceding story were presented, for which
subjects provided their answers on a paper-and-pencil score sheet.
These questions were included to ensure that the subjects read for
comprehension. After answering a question, the subjects pressed
the "+" key to advance to the next story. On the average, the sub-
jects took approximately 35 min to complete the experiment. They
were then debriefed and told the purpose of the experiment.
Results and Discussion
All 32 subjects answered the content questions cor-
rectly. The data for these subjects were trimmed by ex-
cluding data points 2.5 standard deviations from the rel-
evant condition means. This eliminated only six of the
observations ( < I %). No differences were found between
analyses of the trimmed and untrimmed data, so we re-
port only the analyses of the trimmed data. An initial
ANOYA that included lists as a factor revealed a signifi-
cant main effect of lists, but no significant interactions
involving lists. We therefore report the results of analy-
ses collapsing across this factor.
Mean reading times for the target idioms as a function
of the referent and analogical consistency conditions are
presented in Figure 2. Two-way ANOYAs were per-
formed on these data with referent version (original vs.
other person) and analogical consistency (consistent vs.
inconsistent) as repeated within-subjects factors. Two such
ANOYAs were done, one with subjects as a random fac-
tor (F
s
) and one with items as a random factor (Fi).
Reading times were reliably slower in the other-person
referent version than in the original version (2,323 msec
IDIOM COMPREHENSION 715
--. 2200
u
<l>
E2000
'-"'
<l> 1800
E
i= 1600
OJ
c
is 1400
ro
<l>
a::: 1200
1000
Original Person Other Person
Referent Condit ion
Figure 2, Mean reading times as a function of referent condition
and idiom-text consistency, Experiment 2 (bars indicate standard
errors of the means).
vs. 1,988 msec, respectively). This main effect of refer-
ent version is reliable [Fs(l,31) = 12.62, P < .001;
Fi(l,l5) = 5.93,p < .05]. This result is to be expected.
The stories would lead readers to expect material that fo-
cused on the main protagonist. When the story shifts
abruptly to another person, reading should be momen-
tarily disrupted by the break in story coherence (cf. Gar-
rod & Sanford, 1988). Although expected, this result also
provides a manipulation check on our method, indicating
that the obtained reading times were sufficiently sensi-
tive to permit the detection of variations in comprehen-
sion difficulty.
Analogical consistency had no discernible effect on
reading times. The mean reading times for analogically
consistent and analogicallyinconsistent idiomcompletions
were 2,137 msec and 2,174 msec, respectively. This dif-
ference was not reliable [F
s(1,31)
= 1.78, P > .05;
Fi(1,15) = .80]. More importantly, the interaction of
referent version and analogical consistency did not ap-
proach significance [F
s(l,31)
= 2.01,p > .05; Fj(l,15) =
.90]. Even though we did not obtain a significant inter-
action, we performed t tests on the difference between
analogically consistent and inconsistent idiom conditions
in the original-referent condition. The difference between
these two conditions is 171 msec, roughly half the mag-
nitude of the 335-msec difference obtained for the
original- and other-person referent conditions. The effect
of analogical consistency within the original-person refer-
ent conditions did not approach significance [t
s
(31) =
1.38, P > .16; ti(l5) = 1.01, P > .26]. These results
replicate those of Gibbs (1992), who reported a similar
failure to find effects of analogical consistency on com-
prehension performance, as measured by reading times.
The absence of any main effects or interactions involv-
ing analogical consistency, together with a robust effect
of referent version, suggests that even when a specific
716 GLUCKSBERG, BROWN, AND McGLONE
conceptual analogue for an emotion is available in a story,
that conceptual analogue is not automaticallyaccessed dur-
ing idiom comprehension. Persons in a story may have
been described, for example, as being angry in the hot
and fuming sense, yet this did not facilitate the compre-
hension of an analogically consistent idiom such as blow
one's top as opposed to a different and analogically in-
consistent idiom such as bite someone's head off. As far
as the reading times indicate, anger is anger. Our sub-
jects apparently did not take advantage of any conceptual
analogue consistencies between story elements and idio-
matic expressions when comprehending those expressions.
There are, however, two possible reasons for treating
these conclusions with caution. First, null findings are ul-
timately ambiguous, even when replicated across labora-
tories. Second, and perhaps more important, the particu-
lar set of conditions that we used in Experiment 2 may
have effectively prevented the detection of analogical con-
sistency effects. As indicated by the results of Experi-
ment 1, subjects foundthe other-person referent texts odd.
These texts were odd enough to produce significantly
elevated reading times in Experiment 2. There is the pos-
sibility that the disruptive effect of the other-person items
in Experiment 2 produced enough variance in responses
that any effects of analogical consistency in the original-
referent condition were simply undetectable in the resul-
tant (statistical) noise.
Accordingly, we performed a third experiment with the
original-referent condition to test for effects of analogi-
cal consistency on reading time, together with the non-
person referent condition to control for lexical priming
effects. Recall that the nonperson referent stories con-
tained scenarios in which a situation or state of affairs,
rather than any person, was described by words that are
associated with an idiom's conceptual analogy (see Ap-
pendix A). Becausethese stories did not involvean abrupt
switch of topic, we did not expect them to have disrup-
tive effects on reading times.
EXPERIMENT 3:
A Further Search for
Analogical Consistency Effects
Method
Subjects. Thirty-three Princeton undergraduates served as paid
volunteers. All were native English speakers, and none had previ-
ously participated in an experiment involving figurative language.
Materials and Design. The original referent stories from Ex-
periment 2 were retained for this experiment, and the nonperson
referent stories from Experiment I were adapted for line-by-line
presentation (see Appendix B). As in Experiment 2, stories were
created by combining referent versions of each vignette with one
member of a given idiom pair. The particular idiom could be ana-
logically consistent or inconsistent with the story context. This
yielded four story conditionsfor each idiom. The sentencecontaining
the target idiomatic phrase appeared either at the end of the story
or as the next to the last sentence in the story.
As in Experiment 2, the materials were used in a 2 x 2 design,
with referent version (original vs. nonperson) and analogical con-
sistency between story context and idiom (consistent vs. inconsis-
tent) as within-subjects factors. The 64 stories (16 idioms x 4 story
conditions) were randomly assigned to one of four lists, so that only
one of the story conditions for a particular idiom appeared on a
given list. Each list contained 4 items of each story type for a total
of 16experimental items. Interspersed among the experimental sto-
ries were the 16 filler stories used in Experiment 2, 3 of which ap-
peared as practice items.
Procedure. The procedure in this experiment was identical to
that of Experiment 2.
Results and Discussion
One of the subjects failed to answer all of the content
questions correctly, so this subject's reading time data
were discarded. The data for the remaining subjects were
trimmed by excluding data points 2.5 standard devia-
tions from the relevant condition means. This eliminated
10 of the observations (less than 1%of the data). No sig-
nificant differences were found between analyses of the
trimmed and untrimmed data, so we report only the anal-
yses of the trimmed data.
Mean reading times for the target idioms as a function
of referent and analogical consistency conditions are pre-
sented in Figure 3. Two-way ANOVAs were performed
on these data with referent version (original vs. non-
person) and analogical consistency (consistent vs. incon-
sistent) as repeated within-subjects factors. Two such
ANOVAs were conducted, one with subjects as a random
factor (Fs), and one with items as a random factor (Fs).
Reading times were slightly slower in the nonperson
referent versions than in the original versions (2,192 msec
vs. 2,126 msec, respectively). This difference, however,
was not significant in either the subjects or items analy-
ses [Fs(l,32) = 1.57, p > .22; Fj(l,15) = 0.84, p >
.38]. As expected, the nonperson versions of the stories
did not produce the disruption in reading time that was
found for their other-person version counterparts in Ex-
periment 2, which contained clear breaks in narrative con-
tinuity. Thus, the use of the nonperson versions in this
2400
,.-.. 2200
u
Q)
E2000
Q) 1800
E
i= 1600
OJ
c
'6 1400
ro
Q)
a:: 1200
1000
Original Person Non-Person
Referent Conditi on
Figure 3. Mean reading times as a function of referent condition
and idiom-text consistency, Experiment 3 (bars indicate standard
errors of the means).
experiment effectively eliminated this potential masking
factor for analogical consistency effects.
As in Experiment 2, there was no effect of analogical
consistency on reading times [Fs(1 ,31) =0.84; Fj(1, 15) =
.59], nor did analogical consistency interact with refer-
ent conditions [Fs(1,31) = 1.22, p > .25; Fj(1,15) =
1.04, p > .30]. Not only did we fail to find effects of
analogical consistency, both the difference in the overall
means for analogically consistent and inconsistent idiom
completions (2,172 and 2,146 msec) and the difference
in means within the original-referent condition (2,160
and 2,092 msec) are in the opposite direction of what
would be expected if analogical consistency facilitated
comprehension.
Taken together, the results of Experiments 2 and 3
suggest that even when a specific conceptual analogue for
an emotion is available in a story, that conceptual analogue
is not automatically accessed during idiom comprehension.
GENERAL DISCUSSION
The systematicity of idioms in various conceptual do-
mains is striking. Surely it is not coincidental that anger,
for example, can be described in such related ways asjlip
one'slid, bLow one'sstack, expLode, get hot under the coL-
Lar, Let offsteam, and so forth. At the very least, this sys-
tematicity implies that once anger has been expressed in
one of these ways, people could readily coin related vari-
ants of the original idiom, whether or not that original
idiom reflected any particular conceptualization of anger
(e.g., as heated fluid under pressure). Even minimally
motivated idiomatic expressions are commonly extended
in systematic ways, as in a recent newspaper headline
about the Chicago river flooding into the city's subterra-
nean tunnels. If one recalls Carl Sandburg's poem that
characterized Chicago as a "city of broad shoulders,"
the headline "Broad Shoulders and Wet Feet" makes per-
fect sense.
Other than reflecting people's penchants for systema-
ticity in linguistic expressions, what does a particular sys-
tematicity imply about conceptual structures? The sys-
tematicity of idiomatic expressions for such concepts as
anger, fear, success, and failure, among others, suggests
at least three possibilities.
1. Given any set of related idiomatic expressions, peo-
ple are able to recognize the relation and then may infer
a basis for that relationship.
2. Conceptual metaphors for idioms are available in se-
mantic memory and accessible in some contexts, but not
in others. Specifically, they are not automatically accessed
and thus do not playa necessary role in either production
or comprehension.
3. Conceptual metaphors underlie idiom meanings: they
not only are available, but also are automatically accessed
during idiom production and comprehension, regardless
of context. They playa central role in idiom use.
We will consider each of these possibilities in tum.
First, at the simplest level, systematic clusters of idio-
matic expressions may imply nothing more than that peo-
IDIOM COMPREHENSION 717
ple can recognize a specific relation among expressions
for a given concept and infer a plausible basis for that
relationship. Thus, if one were to notice that the expres-
sions jlip your lid and bLow one's top referred to anger
and also could be used to describe something in a pressur-
ized container, one might infer that some aspect of the
concept "anger" is analogous to heated fluid under pres-
sure. This would not only enable a person to make judg-
ments about the relative appropriateness of idioms in con-
texts, as we found in Experiment 1, but also enable people
to coin new expressions based on the same set of rela-
tions. This is the most parsimonious interpretation of the
results of our first experiment, in which subjects preferred
idioms that were stylistically consistent with a preceding
text to idioms that were stylistically inconsistent.
This interpretation suggests that people can readily in-
fer conceptual motivations for idioms even when those
motivations were irrelevant to begin with. A recent study
by Keysar and Bly (1992) lends credence to this notion.
Keysar gave people novel idioms in contexts that enabled
them to infer the idioms' general meaning. One such
idiom, culled from British English, is The goose hangs
high. In one context that was given to one group of sub-
jects, two farmers are strolling home after a long day in
the fields. They express gratitude for the good growing
season and the bountiful crop. At the conversation's end,
one remarks to the other, "Aye, John, the goose hangs
high." In an alternative context, given to a different group
of subjects, the farmers bemoan the poor growing sea-
son and express dismay at the prospect of a dismal winter.
At the end of this conversation, one farmer remarks to
the other, "Aye, John, the goose hangs high."
If, as Lakoff and Johnson (1980) might contend, the
conceptual metaphor "good is up" has a controlling in-
fluence here, the positive interpretation of the goose idiom
should be more intuitively satisfying than the negative in-
terpretation. No such preference was evident. Each group
of subjects rated the idiom as "making sense," regard-
less of whether the meaning was positive or negative. Each
group also indicated that the opposite meaning of the idiom
would not make much sense. Finally, each group felt that
the meaning was transparent: a majority of the subjects
in each group predicted that other people would be able
to get the idiom's meaning even without a supporting con-
text. Keysar's results are reminiscent of young children's
attitudes toward word meanings. As adults, we know that
word meanings are assigned arbitrarily, and that what we
call, say, the moon could just as well be called by any
other name. For young children, a name "belongs" to
its referent-a phenomenon described by Piaget (1926)
as nominal realism. Perhaps we as adults retain a vestige
of childhood nominal realism in our willingness to read
systematicity and meaning into essentially arbitrary idio-
matic word strings.
An alternative reason for systematically related idio-
matic expressions within a domain is that (1) they are
motivated by common conceptual cores; (2) the concep-
tual cores are metaphorical in nature; and (3) they are
available and in certain contexts accessible in semantic
718 GLUCKSBERG, BROWN, AND McGLONE
memory. The results of Experiment 1 are certainly con-
sistent with this hypothesis. When faced with a choice of
analogically consistent or inconsistent idioms, our sub-
jects preferred the consistent idioms, perhaps because the
relevant conceptual metaphors for, say, anger, were ac-
tivated during comprehension of the context stories. The
data of Experiment l , however, do not permit us to dis-
criminate between the recognition-inference hypothesis
and the availability-in-memory hypothesis.
The third and strongest implication of idiom systema-
ticity is that conceptual metaphors underlying idiommean-
ings are automatically activated and accessed during pro-
duction and comprehension. As Lakoff (in press) clearly
puts it, "The systemof conventional conceptual metaphor
is mostly unconscious, automatic, and is used with no
noticeable effort, just like our linguistic system and the
rest of our conceptual system." The results of Experi-
ments 2 and 3 do not confirm this hypothesis. There was
no comprehension advantage for analogically consistent
idioms over analogically inconsistent ones. Since the de-
pendent measure, reading time, was sufficiently sensitive
to produce a robust effect of shifting focus (original pro-
tagonist vs. other person, Experiment 2), it is unlikely
that the lack of effect for analogically consistency is at-
tributable to the particular measure that we used.
Our failure to find an effect of analogical consistency
does not, of course, mean that people (1) do not have con-
ceptual metaphors available in semantic memory, or
(2) do not access those metaphors under certain circum-
stances. Our conclusion pertains specifically to the claim
for automaticity, which we feel is certainly wrong. Under
appropriate circumstances, conceptual metaphors may
well be accessible and used for production or compre-
hension purposes. What those circumstances might be re-
mains to be determined. Further work should enable us
to specify the necessary and sufficient conditions for the
activation of conceptual metaphors if, indeed, such
metaphors are represented in semantic memory.
REFERENCES
BoBROW, S. A., 8t BELL, S. M. (1973). On catching on to idiomatic
expressions. Memory & Cognition, I, 343-346.
GARROD, S., 8t SANFORD, T. (1988). Thematic subjecthood and cogni-
tive constraints on discourse structure. Journal of Pragmatics, 12,
519-534.
GIBBS, R. (1992). Metaphorical knowledge andunderstanding language.
Paper presented at the conference on Metaphor and Cognition: Studies
in Art, Linguistics, Literature, Philosophy and Psychology, Univer-
sity of Tel Aviv, Israel.
GLUCKSBERG, S. (199\). Beyond literal meanings: The psychology of
allusion. Psychological Science, 2, 146-152.
GLUCKSBERG, S. (in press). Idiom meanings and allusional content. In
C. Cacciari & P. Tabossi (Eds.), Idioms: Representation. compre-
hension and production. Hillsdale, NJ: Erlbaum.
HIGGINS, E. T., RHOLES, W. S., 8t JONES, C. R. (1977). Category ac-
cessibility and impression formation. Journal of Experimental Social
Psychology, 13, 141-154.
KEYSAR, 8., 8t BLY, 8. (1992). Conceptual metaphors andconventional
language: Why do some idioms seem transparent? Paper presented
at conference on Metaphor and Cognition: Studies in Art, Linguis-
tics, Literature, Philosophy and Psychology, University ofTe1 Aviv,
Israel.
KOVECSES, Z. (1986). Metaphors ofanger, pride andlove. Amsterdam:
John Benjamins.
KREUZ, R. J., 8t GRAESSER, A. C. (1991). Aspects of idiom interpre-
tation: Comment on Nayak and Gibbs. Journal of Experimental Psy-
chology: General, 120, 90-92.
LAKoFF, G. (1987). Women, fire and dangerous things. Chicago: Uni-
versity of Chicago Press.
LAKoFF, G. (in press). The contemporary theory of metaphor. In
A. Ortony (Ed.), Metaphor and thought (2nd ed.). Cambridge: Cam-
bridge University Press.
LAKoFF, G., '" JOHNSON, M. (1980). Metaphors we live by. Chicago:
University of Chicago Press.
MCGWNE, M. S., GLUCKSBERG, S., 8t CACCIARI, C. (in press). Se-
mantic productivity and idiom comprehension. Discourse Processes.
NAYAK, N. P., 8t GIBBS, R. W. (1990). Conceptual knowledge in the
interpretation of idioms. Journal of Experimental Psychology: Gen-
eral, 119, 315-330.
PtAGET, J. (1926). The language and thought ofthe child (M. Warden,
Trans.). New York: Harcourt Brace.
POTTS, G. R., KEENAN, J. M., 8t GOLDING, J. M. (1988). Assessing
the occurrence of elaborative inferences: Lexical decision versus nam-
ing. Journal of Memory & Language, 27, 399-415.
RATCUFF, R., 8t McKooN, G. (1986). Automatic activation of episodic
information in a semantic memory task. Journal ofExperimental Psy-
chology: Learning, Memory, & Cognition, 12, 108-115.
SRULL, T. K., 8t WYER, R. S. (1979). The role of category accessibil-
ity in the interpretation of information about persons: Some deter-
minants and implications. Journal of Personality & Social Psychol-
ogy, 37, 1660-1662.
SWINNEY, D., 8t CUTLER, A. (1979). The access and processing idio-
matic expressions. Journal of Verbal Learning & Verbal Behavior,
18, 523-534.
TRAUGOTT, E. C. (1975). Spatial expressions of tense and temporal se-
quencing: A contribution to the study of semantic fields. Semiotica,
15, 207-230.
TRAUGOTT, E. C. (1985). "Conventional" and "dead" metaphors. In
W. Paprotte & R. Dirven (Eds.). The ubiquity ofmetaphor (pp. 17-
56). Amsterdam: John Benjamins.
APPENDIX A
Sample Materials from Experiment 1 (Idiom Choice)
Original Nayak and Gibbs Story (Original Referent)
Susan was waiting impatiently for Chuck to pick her up from
school. She had a final the next day and she wanted to get home
to study. She was really tense about how she would do on the
final. She was getting really hot with Chuck for his inconsider-
ate behavior. After an hour of waiting, Susan was reaching the
limits of her patience. She was fuming and fretting, wondering
how she would get home. When Chuck finally arrived with not
a hint of an apology, Susan ...
blew her top.
bit his head off.
Other-Person Referent Version
Susan was waiting impatiently for Chuck to pick her up from
school. She had a final the next day and she wanted to get home
to study. She was really tense about how she would do on the
final. She was getting really hot with Chuck for his inconsider-
ate behavior. After an hour of waiting, Susan was reaching the
limits of her patience. When Chuck finally arrived with not a
hint of an apology, Susan sat in the car fuming and fretting.
Chuck ...
blew his top.
bit her head off.
Nonperson Referent Version
Susan was waiting for Chuck to pick her up from school. The
weather was impatient: the rain began earlier than predicted.
Susan had a final the next day and she wanted to get home to
study. As she waited, the winds became tense and their breath
became hot. Susan realized that Chuck was being inconsiderate
by keeping her waiting in this weather. After an hour of wait-
ing, the clouds were reaching the limits of their patience and
Susan was getting drenched. The winds continued their fuming
and fretting, tossing Susan's hair into her face. When Chuck
finally arrived with not a hint of an apology, Susan ...
blew her top.
bit his head off.
APPENDIX B
Sample Materials from Experiments 2 and 3 (Reading Time)
Original Referent Story (Experiments 2 and 3)
Susanwaited impatientlyfor Chuck to pick her up from school.
She had a final tomorrow and wanted to get home to study. She
was really tense about how she would do on the final. She was
getting really hot with Chuck for his inconsiderate behavior.
After an hour of waiting, Susan reached the limits of her pa-
IDIOM COMPREHENSION 719
tience. She was fuming and fretting, wondering how she would
get home. Chuck eventually arrived, without a hint of an apol-
ogy. Susan blew her top/bit his head off.
Other-Person Referent Story (Experiment 2)
Susan waited impatientlyfor Chuck to pick her up fromschool.
She had a final tomorrow and wanted to get home to study. She
was really tense about how she would do on the final. She was
getting really hot with Chuck for his inconsiderate behavior.
After an hour of waiting, Susan reached the limits of her pa-
tience. Chuck eventually arrived, without a hint of an apology.
Susan sat down in the car, fuming and fretting. Chuck blew his
top/bit her head off.
Nonperson Referent Story (Experiment 3)
Susan waited for Chuck to pick her up from school. The
weather was impatient; the rain began earlier than predicted.
Susan had a final tomorrow and wanted to get home to study.
As she waited, the winds became tense, and their breath be-
came hot. Chuck was being inconsiderate to make her wait in
this weather. After an hour or so, the clouds reached their lim-
its of patience. As it rained, the winds continued fuming and
fretting. Chuck eventually arrived, without a hint of an apol-
ogy. Susan blew her top/bit his head off.
(Manuscript received June 29, 1992;
revision accepted for publication February 5, 1993.)