JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
Schütz, Baumeister / LANGUAGE OF DEFENSE
THE LANGUAGE OF DEFENSE
Linguistic Patterns in Narratives of
ROY F. BAUMEISTER
Case Western Reserve University
To investigate how defensive motivations would affect patterns of language use, the
a set of narratives about making someone happy. Compared to the happy stories, trans
gression narratives were more likely to describe actions occurring without deliberate
guidance or intention. Length in word count did not vary, but transgression narratives
had shortersentences,especially inthe sections describing thetransgression andits con-
sequences. They had longer introductions, presumably to explain background and miti-
gating circumstances. Transgression narratives featured the emotions and thoughts of
the narrator significantly more than did narratives of making someone happy (which
emphasize the narrator’s emotions. Transgression narratives had fewer specific details
but more (ostensibly) exact quotations.
It is generally accepted that people prefer to avoid discussing their
own misdeeds and transgressions—unlike the misdeeds and trans-
gressions of others, which people have long shown a keen interest in
discussing, as evidenced by gossip and scandal. The reluctance to dis
cuss one’s own misdeeds probably stems from the undesirability of
making oneself look bad. Sometimes, however, such revelations are
inevitable,andsopeoplemustconfront the dilemmaofhowtodescribe
the misdeeds of others may seem quite unsuitable for recounting one’s
the Alexander-von- Humboldt Foundation to the first author, and was supported by
grants MH 43826 and MH 51482 from the National Institute of Health to the second
author. The authors thank the University of Virginia for laboratory and subject pool re
sources, including special help by Darren Newtson, and we thank Margit Wenzler and
Thomas Kauper for their work coding the data. Address correspondence to A. Schütz,
Dept. of Psychology, University ofBamberg,Postfach 1549 D-96045Bamberg,Germany
(or astrid.schuetz@ ppp.uni-bamberg), or to R. Baumeister, Dept. of Psychology, Case
Western ReserveUniversity, ClevelandOH 44106-7123, USA,(or firstname.lastname@example.org), of
fice phone 216-368-2639, home phone 216-531-5185, fax 216-368-4891.
JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY, Vol. 18 No. 3, September 1999 269-286
1999 Sage Publications, Inc.
own acts, insofar as that sort of language dwells on the unsavory acts
and their deplorable consequences, whereas one presumably would
want to minimize anything that reflects unfavorably on the self. The
ple’s descriptions of their transgressions, particularly those patterns
that may help defend the self against the unsavory implications of the
actions being described.
PREVIOUS WORK ON DEFENSIVE LANGUAGE
The idea that people can manipulate language so as to conceal or
minimize unsavory aspects of their own history is not new, of course.
The self-serving manipulation of language was a major theme of
George Orwell’s novel 1984, in which the government sought to elimi
nate certain words, create new ones, and alter the meanings of others
so as to shape the thoughts of citizens. In psychology, a germane point
was made by Sullivan (1953), who coined the term verbalisms to refer
to patterns of speech that are designed to conceal, instead of to
Empirical research has already shown that people are sensitive to
the predicament of how to describe their own undesirable behaviors.
Theydenyorminimizetheirresponsibility,justifytheir actions asless
wrong than they seem, blame others, emphasize mitigating circum-
stances,apologize(especially ifaccepting blame doesnot impugn their
moral integrity, as in accidental transgressions), selectively omit
uncongenial facts, distance the act from the global self, and generally
manipulate both the facts and the attributional implications (e.g.,
Baumeister, 1998; Schönbach, 1985; Schütz, 1991).
Recent evidence has begun to examine the manipulation of lan
guagein service of psychological ends. Schmid and Fiedler (1996) ana
lyzed prosecutors’ and defense attorneys’ closing speeches in the Nur
emberg war crimes trials, and they found differences in language
patterns. For example, defense attorneys used a higher level of
abstraction (which implies generality) when referring to defendants’
positive attributes than their negative attributes, and they avoided
referring directly to the person when discussing negative attributes
a linguistic intergroup bias, whereby positive ingroup behaviors and
negative outgroup behaviors are described in broad, abstract terms,
whereas narrowly concrete speech is preferred for negative ingroup
and positive outgroup behaviors (Arcuri, Maass, & Portelli, 1993;
of speech makes the ingroup look good and the outgroup less good
(Maass, Milesi, Zabbini, & Stahlberg,1995).
Language can be used to manipulate impressions (for an overview,
see Fiedler & Semin, 1996). More specifically, previous research sug-
270 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
gests that certain linguistic patterns may be used to conceal a perpe
trator’s responsibility, such as including details that depict the victim
aspartially responsible,use ofpassive voice,nominal style,and avoid
ing explicit references to what happened (for an overview see Bohner,
tering women and found the following patterns: nominalization (e.g.,
“the abusive cycle”), diffusion of responsibility or avoiding explicit ref
erence to the deed (e.g., “violent relationships”; “domestic disputes”),
and use of passive voice.
Lamb and Keon (1995) tested the effect of linguistic patterns on
attributionofguilt. Useof passivevoicefailed toreduce attributions of
toward the batterer. Bohner (1998) used an experimental approach,
investigating how people report on rape and how different types of
(inGermany)who attributemore responsibility tothe victimmorefre
quently use passive voice, avoid explicit references to the word rape,
and use nominal style. However, passive style was very rare overall,
whichmay reflect adifference between Englishand German language
use. In a second experiment, Bohner had subjects evaluate reports on
anincident of rape.He found that the useof passive voiceled to attrib-
uting more responsibility to the victim only if people were prone to
accept myths about rape and if the report included details that sup-
ported such a stereotypic interpretation.
The fact that different types of verbs and adjectives have different
cognitive implications has been supported in workby Semin and Fied-
ler (1988) and Semin, Rubini, and Fiedler (1995). It is reasonable to
expect that people learn to use and manipulate these differences for
their own social and personal purposes.
Recent research has shown that linguistic features do have an
impact on perception (Berry, Pennebaker, Mueller, & Hiller, 1997).
Subjects were videotaped while describing themselves, and other sub
jects rated those tapes. Results showed that, for example, impressions
of competence were positively related to the use of present tense and
negatively related to negations, negative emotion words, and self-
also shown in a study by McNeil, Pauker, and Tversky (1988) where
identical data were framed in terms of positive survival outcome or in
terms of mortality rates.
Thus, past work justifies the conclusion that defensive and asser
tive language differs in context of what is reported, and there is some
evidencesuggestingthatdefensiveness may affectthelevelof abstrac
tion used, the use of passive voice, the inclusion of details, or explicit
references to a negative deed. Many of those defensive linguistic
devices were observed with respect to transgressions of law, however,
Schütz, Baumeister / LANGUAGE OF DEFENSE 271
andit is notclear whether moreeveryday interaction would be subject
to such distortions, too. Furthermore, several other linguistic aspects
that may very well be relevant to defending one’s action have not been
considered yet, such as how much detail an event is described in (e.g.,
length of description, elaboration and inclusion of details), or how
much consideration is given to the perspective of victim and perpetra
tor (i.e., reporting thoughts and emotions of victim and perpetrator).
The present study asked a sample of university students to write
about an incident in which they had hurt someone. For comparison
purposes, other participants were asked to write about an incident in
which they had made someone happy. We assumed that stories about
and so the requirement to write such a story would constitute a pre
dicament. We expected that people would respond to this predicament
with various linguistic devices intended to reduce the degree to which
the story made them seem blameworthy and malicious.
Ourfirstpredictionconcerned intentionalaction.Ithas been shown
in prior research that perpetrators like to depict their actions as unin-
tentional (Baumeister, Stilwell, & Wotman, 1990; Schütz, 1991). Sen-
tences referring to deliberate volition, such as saying “I decided...,”
should therefore be rarer in the transgression stories than in the com-
parison (happy) stories, because volition and conscious decision casts
theselfas responsible,andtransgressors should bemotivatedto avoid
responsibility. In contrast, passive, external constructions such as “It
eventsevolvedon theirownmomentum, ratherthanas aresultof con
scious or deliberate decisions, so that blame is reduced.
Opposing pressures affect story length. The literature on coping
pointsouttwoopposingways of dealingwithdisagreeableorthreaten
inginformation, namely avoidantversus refutational defenses(Wyer &
Frey, 1983; see also Baumeister & Cairns, 1992; Frey, 1986). In this
line of thinking, it can be argued that, on one hand, having to talk
sors may want to minimize the amount of time they spend talking
about their misdeeds (see Gonzales, Pederson, Manning, & Wetter,
transgressors may seek to furnish explanations and justifications for
their actions, including a description of all mitigating circumstances
(see Baumeister et al., 1990; Schönbach, 1985), and so they might
write longer stories than others.
Althoughcompeting predictionscould bemade aboutlength of story,
the two theoretical bases do agree in predicting the different distri-
272 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
want to explain their reasons andjustifications inorder to make their
actions seem more understandable (and hence perhaps forgivable),
they should focus their efforts on the backgroundevents leading up to
the transgression. Meanwhile, their effort to minimize the blamewor
thiness of their act should make them spend less time detailing the
consequences and aftereffects. Accordingly, the simple hypothesis
about differences in total length could be elaborated by a more fine-
grained hypothesis, namely, that the accounts of transgression would
son group. To measure length, we counted both words and sentences.
We assumed that the word count would reflect the total amount of
meaningful units of information.
Anotherinteresting aspect is sentence length,which isindicative of
the grouping of information into meaningful units. Short sentences
presumably reflect a fragmented, deconstructed presentation of infor
mation,inwhichthe informationispresentedin many small,separate
units. In contrast, reliance on long sentences represents a more inte-
ing all the information into larger units. It can be argued that trans-
gressors would want to avoid an integrative style of thinking, because
that style would render apparent the unsavory implications of their
actions. Therefore, we predicted that transgressors would use shorter
that they acted from positive motives but somehow produced undesir-
able outcomes, and this effort could result in longer, more complex
thinking (e.g., Tetlock, 1985).
Besides story length, furnishing details is another aspect of elabo
rating on an event. If transgressors are indeed motivated to present
their misdeeds in a vague and less elaborated way, they should gener
ally be reluctant to furnish all the relevant details. We predicted that
in the comparison stories.
with respect to whose perspective is represented in the narrative and
how much consideration is given to the transgressor’s and the target’s
inner states. People writing stories about doing a good deed for some
one else should presumably dwell on how the other person (the target)
thought and felt as a result of the narrator’s actions, because those
thoughtsand feelings will dramatizethe goodness of thegood deed (cf.
enhancements, Schlenker, 1980). In contrast, people writing about
hurting someone may want to downplay the target’s thoughts and
feelings, because these mental and emotional states would convey the
suffering the narrator caused (Baumeister et al., 1990). Hence we
Schütz, Baumeister / LANGUAGE OF DEFENSE 273
predicted less emphasis on the target’s inner states in the transgres
Furthermore,whenone describesone’s pastmisdeed,one maybe at
pains to explain how one’s own emotional state or mental processes
ally accepted, even in courts of law, that misdeeds committed under
emotional distress are less blameworthy than those committed in the
absence of emotion (Averill, 1982), and so people may seek to reduce
their blame by emphasizing their own emotional distress. In extreme
cases, perpetrators can depict themselves as victims too, if they can
about hurting someone elsecan end up coming across as a story about
joint or mutual victimization (Baumeister, 1997). We predicted that
the stories about hurting someone would show an increased emphasis
on the narrator’s emotional state, particularly with regard to negative
SUMMARY OF PREDICTIONS
1. Transgression stories will include fewer statements describing inten-
tional action, such as “I decided to...”butmore statements describing
external causation, such as “It happened. . . .”
2. Overall stories will not differ in length. However, transgressions will
3. Transgressions will be presented in a more fragmented style. That is,
they will consist of shorter sentences than the comparison stories.
4. Transgressions will contain fewer details than the comparison stories.
5. Target thoughts and feelings will be deemphasized in the transgression
stories while the narrator’s inner states, especially negative emotions,
will be emphasized.
Participants. The experiment was run in groups of 4 to 6 persons.
They were undergraduates who participated in connection with their
introductory psychology course. The total sample consisted of 122 stu
dents (52 male, 70 female) and reflected an assortment of ethnic
Procedure. The experimenter (the first author) explained to each
group that the study was concerned with the effects of mood on mem
on a scaleconsisting of 16 items (calm, happy, content, nervous, tired,
sad, etc.), each of which had a 7-point scale. The mood scale was used
simply to make the cover story plausible, and to allow a little extra
time for late-arriving participants to get settled.
274 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
rience. By random assignment, each person was assigned to write
anevent in which his orher actions made someone happy.As a control
condition to the transgressions, we used positive events to avoid the
problem of comparing meaningful and important to relatively mean
ingless, unimportant events. The randomization procedure was tar
geted at the entire group, because the instructions were given orally,
tion in the Hurt condition was “Please describe an event in which your
actions hurt someone. Ideally, you would choose the most important
and extreme event you can remember.” In the Happy condition, the
words “hurt someone” were replaced by the words “made someone
happy.” Further instructions asked the participants to give thorough
and detailed descriptions. We asked for the most important events to
receivea sampleof events, and to excludetrivial eventsthat narrators
may not have cared enough about to describe in a comparable way.
All participants were assured of confidentiality and were exhorted
not to use real names in their stories. No identifying information was
collected. The experimenter allowed participants 30 minutes to com-
plete their essays. After 30 minutes, the essays were collected and the
ment and their expectations. Then they were given a full debriefing
that included an explanation of the purpose of the study.
quency of various language features. One assistant coded all the sto-
ries. A subsample (10%) was coded independently by a second assis
tant, in order to permit the calculation of reliabilities. The two
assistants were blind to the hypotheses of this study. Training of the
coders involved explication of the coding categories as they are
described below, as well as supervised coding of a sample of six stories
that were not used in the present analyses.
To check on the use of language to emphasize or diminish the
appearance of volition, we coded first for sentences that asserted that
the actor had acted consciously or intentionally, such as “I decided
to...”or“Ichoseto...”or“Iwantedto....”Wealsothen counted sen
almost an unconscious act,” “it wasn’t an intentional thing,” “it just
happened,” and “before I knew it, I was signing. . . .”
ofsentences.For eachstory, wecomputedthe averagesentence length
by dividing the number of words by the number of sentences. For each
story,we thenidentified the transitionpoint betweenthe introduction
or background, and the incident itself (plus any explanation of
Schütz, Baumeister / LANGUAGE OF DEFENSE 275
aftermath or consequences). Both coders identified this transition
point independently. They agreed in 92% of the cases. Dissent was
adjudicated through discussion. We then did the word count, sentence
To examine elaboration, we analyzed three aspects possibly repre
senting elaboration that emerged during the coding process: (a) the
number of adjectives used in each story; (b) the number of specific
details, such as names, locations, ages, distances, times, frequencies,
and dates; (c) the number of specific quotations (i.e., supplying actual
words allegedly spoken during the incident).
emotions pertaining to either the narrator or the target. We also
counted the use of words or phrases to emphasize the strength of the
emotions, such as “terribly upset,” “happy,” or “happier and happier.”
Last, we tallied the allusions to specific thoughts by the characters in
the story. We made separate counts for the narrator’s own thoughts
and for the target’s thoughts.
RESULTS AND DISCUSSION
written pages long. About 40% of the targets were friends, 25% were
family members, and 20% were romantic partners. The rest were
acquaintances and strangers. Most of the events involved rather
recent but not ongoing events. Because most stories did not provide
exactdates, coders estimatedhow recent eventswere. Over 80% of the
stories were estimated to have occurred less than two years but more
target was were not associated with story type. All of the events
appeared to be rather meaningful to the narrators and thus con
formed to the task given. However, none of the events involved very
severe events such as physically injuring someone or other legal
Stories about hurting someone referred to insensitive, cold, rude
behaviors; to norm transgressions such as stealing, betraying secrets,
gossiping, or harming someone’s reputation; or failing to meet the
expectations of others. Stories about making someone happy included
major gifts or presents, participants’ own successes in academics or
athletics (which made their parents happy), important or well-timed
communications (especially phone calls or letters) to parents or
friends, engaging in fun activities with someone, and lending support
to someone in trouble.
276 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
Reliabilityofcodingand analysisstrategy.Cohen’sKappa wasused
to assess the reliability of codings. Across the coding dimensions, the
average Kappa was .88, and the range was .70 to 1.00. These figures
simple tallies of relevant behaviors. The hypotheses were tested by
comparing the mean frequency in the hurt versus happy stories.
Before proceeding to specific hypotheses, we conducted a MANOVA
with the story type as independent variable and the various coding
dimensions as dependent variables. This was highly significant, F(17,
105) = 14.54, p < .001. The main results are summarized in Table 1.
Intentional action. Ourfirst hypothesis concerned the use of phras
ingto deny oraffirm that thenarrator’s actions were intentional. Con
gression stories (i.e., stories about hurting someone) to use phrases
thatevadedordeniedintentional action,suchas“beforeI knewit,...”
p < .001. Thus, narrators of transgressions spoke so as to conceal or
minimize their control of events. The related hypothesis concerning
phrasingsthat emphasize responsible volitionwas not supported. Use
of phrases such as “I decided...”were not used at significantly differ-
ent rates in the two sets of stories, t < 1, ns.
The results provide partial support for the view that accounts of
make it seem as though events followed along on their own and pro-
duced unwanted consequences without the active guidance of the
transgressor. With such language, people can send the implicit mes-
sage that they hurt someone without fully intending to do so.
Story length. The amount of information furnished by participants
wascoded in severalways. The simplecount of the number of words in
the story did not yield a significant difference between the two types,
t(121) = 1.13, ns. There was, however, a significant difference in the
numberofsentencesperstory,t(121)=2.73,p < .01.Thetransgression
stories had more sentences than the stories about making someone
Each story was divided at the point where the background or intro
duction ended and the narration began to cover the actual incident
itself. The hypothesis that the transgression stories would emphasize
the introduction disproportionately, as compared with the stories
about making someone happy, was supported. The stories about hurt
ing someone had significantly more sentences in their introductions
than the comparison stories, t(121) = 2.65, p < .01, and they had mar
ginally significantly more words, t(121) = 1.71, p < .10. The endings
(including the narrations of the incidents themselves) did not differ
Schütz, Baumeister / LANGUAGE OF DEFENSE 277
significantly. Neither the the number of words northe number of sen
tencesyieldedasignificantdifferencebetweenthe two typesofstories,
t < 1, ns for both.
The results suggest that narrators went to greater lengths to pre-
sentthe backgroundsand mitigatingcircumstances oftheir misdeeds,
as compared to the stories about good deeds. If one wants to describe
somethinggood that onedid, such asmaking someone happy,it is per
haps enough to skip over the background and simply present the com
mendable action itself. In contrast, when describing a transgression,
people may seek to furnish persuasive accounts of why they acted as
they did, even to the extent that their acts may seem reasonable and
justifiable. Simply describing a transgression by itself could leave the
278 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
Mean Frequencies of Linguistic Features in Two Types of Stories
Made Someone Happy Hurt Someone
“I decided to . . . ” or similar 0.54 0.65 ns
“It happened . . . ” or similar 0.10 0.52 ***
Number of words 213.79 231.65 ns
Words in introduction 139.95 165.61 *
Words in end 73.84 66.03 ns
Number of sentences 28.34 35.26 **
Sentences in introduction 16.83 23.02 **
Sentences in end 11.51 12.24 ns
Words per sentence 7.82 6.84 ***
Words per sentence in introduction 9.04 8.50 ns
Words per sentence in end 6.44 5.39 *
Number of adjectives 11.59 10.30 ns
Number of details 3.36 2.21 **
Number of quotations 0.07 0.31 *
Emotions of target 3.80 1.59 *
Positive 2.77 0.18 *
Negative 1.03 1.42 ns
Emotions of narrator 1.21 1.84 *
Positive 0.85 0.31 ***
Negative 0.36 1.53 ***
Total Emotions (combined) 5.02 3.44 ***
Emphasis with target’s emotions 1.18 0.47 ***
Emphasis with narrator’s emotions 0.28 0.53 *
Cognitions by target 0.23 0.06 *
Cognitions by actor/author 0.34 0.34 ns
*p < .05. **p < .01. ***p < .001.
for little reason. By explaining the background circumstances, people
may seek to avoid that impression.
Fragmentation. The pattern of including more sentences but not
more words naturally pointed toward the possibility that sentence
length would vary betweenthe stories. We computed the average sen
tence length for each story and then entered these individual (partici
pant) means into an analysis. Stories about hurting someone were
indeed composed of shorter sentences overall than the happy stories,
t(121) = 3.59, p < .001. We also compared fragmentation in the intro
not differ with respectto sentence length in the introduction, t <1,ns.
Sentence length, however, did differ in the endings of the stories. Sto
ries about hurting someone had shorter sentences in the latter part
than did stories about making someone happy, t(121) = 2.45, p < .05.
implications of one’s actions, such as their harmful consequences.
This pattern suggests that people use a less integrated and more
fragmented or deconstructed form of speech when describing their
transgressors present the reader or listener with small units of infor-
mation instead of broader, more integrative ones. Small units may
leave the job of working out implications to the listener, and in that
way narrators may avoid having to face up to the full moral implica-
tions of their actions. Apparently, transgressors may write long sen-
tences when describing the background of their actions because they
want to justify and explain what they did, but when they come to
describing the transgression itself and its consequences, they shift
toward shorter sentences. The deed and its consequences are what
make them look bad, so they present these aspects in brief, decon
structed units, unlike the background information which they can use
to mitigate their appearance of guilt.
Details. We predicted that there would be less elaboration in the
transgression stories than in the comparison stories. During the cod
ing process, it became apparent that this prediction could be tested
with respect to three different aspects of elaboration: number of adjec
tives, number of specific details, and number of quotations used. Con
sistent with the prediction, there were fewer concrete details in the
transgression stories, t(121) = 2.42, p < .05. The transgression stories
also showed a trend toward having fewer adjectives, but this failed to
reach significance, t(121) = 1.47, ns.
The number of (supposedly) exact quotations was relatively small
overall, but they were significantly morecommon in the transgression
stories, t(121) = 2.42, p < .05. In general, the quotations referred to a
decisivepointin an argumentorto theexchangein which thenarrator
Schütz, Baumeister / LANGUAGE OF DEFENSE 279
said something that hurt the other person. Often the quotations
sation that led to the hurtful remark, or they presented the narrator’s
own words in a way that made them seem less offensive than the tar
get’s subsequent reaction would imply.
The reliance on quotations in the transgression stories was not pre
dicted, but it may serve the same end as the observed fact that trans
gressions had longer introductions and consisted of shorter sentences.
By relating precisely what was said, the narrator can appear to be
offeringthelistenerallthefactssoastopermit the listener tomakeup
his or her own mind. The quotations often seemed to have the advan
son. Moreover, when the transgression involved saying something
hurtful or upsetting to the target, the narrator can minimize blame by
itself did not generally seem all that blameworthy. Thus, again, the
defensive narrator presents small units of information and ostensibly
allowsthe listenerto makeup his or her ownmind. Asexpected, inclu-
sion of specific details was not common in the transgression stories;
indeed, these stories had fewer such details (e.g., names, dates, loca-
tions, ages, distances) than the comparison stories.
This suggests that transgressors have multiple strategies for deal-
ingwith facts. Onone hand, theypresent crucial factsin a fragmented
their misdeeds and make them seem all the more real and concrete.
avoid letting the reader know precisely what they have done. A trans
help the reader put himself or herself in the transgressor’s place and
regard the transgressor’s actions as understandable.
Emotions and Thoughts. We coded four categories of emotions in
each story: the positive and the negative emotions of the target, and
then of the narrator. Adding these all together, there were fewer total
emotionsin thetransgression stories than in thestories aboutmaking
someone happy, t(121) = 3.39, p = .001.This globaldifference conceals
tallies, however. An ANOVA with three independent variables (story
type × valence of emotion × target vs. narrator) yielded a three-way
interaction among the variables, F(1,121) = 9.90, p = .002.
The obvious and predictable difference between the stories was
12.07, p < .001, and for the narrator, t(121) = 3.46, p = .001. The narra
tor’s negative emotions were more common in the stories about
280 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
hurting, than in the stories about making someone happy, t(121) =
4.66, p < .001. This was, however, not true for the target’s emotions,
t(121) = 1.55, ns. Inspection of the means (see Table 1) suggests that
this is due to an unusually high frequency of negative target emotions
in the stories about making someone happy. This probably signifies
that subjects described the target as feeling bad before they made him
or her happy.
Probably the most important comparisons are within story type,
across person—that is, the comparisons referring to whose emotions
t(60) = 9.04, p < .001, and for negative emotions, t(60) = 3.35, p = .001.
In view of this pair of differences, it is not surprising that the stories
about making someone happy also showed significantly more total
emotions of the target than of the narrator, t(60) = 8.72, p < .001.
In contrast, the transgression stories failed to show a preponder
narrator’s own emotions. The preponderance of narrator emotions
was almost significant with positive emotions, t(60) = 1.93, p = .059,
although those were few in general. There was no significant differ-
encewith negativeemotions, and likewise the totaldifference failedto
approach significance, t < 1, ns.
Thedifference between the preponderance oftarget emotions in the
VAs. These yielded significant interactions between story type and
person(i.e., whowas feelingthe emotion)on totalemotions, F(1,121)=
pattern that the happy stories featured the target’s emotions more
than the narrator’s, whereas this difference was eliminated or
reversed in the transgression stories.
The instructions for both types of stories pointed toward the theme
of how the narrator’s actions had affected the feelings of another per
Based merely on the instructions, therefore, both stories should have
focused about equally on the target’s feelings. Instead, we found that
the narrators of transgression stories devoted ample coverage to their
own feelings, and particularly their unpleasant emotions.
We followed up the emotion codings witha count of the use of words
such as “very” or “extremely” to emphasize the extremity of the emo
tion. With regard to the target’s emotions, such words of emphasis
4.52, p < .001. With regard to the narrator’s emotions, however, the
reverse was found: Narrators used adverbs (and similar words) to
emphasize their own emotionsmore often in the stories about hurting
Schütz, Baumeister / LANGUAGE OF DEFENSE 281
thaninthestories aboutmakingsomeone happy, t(121)=2.19, p <.05.
We conducted a repeated measures ANOVA to examine these differ
ences together. There was a main effect for type of story, F(1,121) =
4.48,p <.05, reflecting lessemphasis onemotions inthe transgression
stories. There was also a main effect for person, F(1,121) = 23.89, p <
.001, indicating overall greater emphasis on the target’s emotions.
F(1,121) = 31.82, p < .001. The happy stories emphasized the target’s
emotions, whereas the hurt stories showed a slight reversal.
References to thoughts or cognitive processes were also coded sepa
rately for the narrator and the target. Stories about hurting someone
made marginally significantly fewer references to the target’s
thoughts than stories about making someone happy, t(121) = 4.31, p <
.10. The narrator’s thoughts were covered about equally in the two
typesofstories, t<1,ns.As withemotions,the importantcomparisons
may be the ones within story type rather than across type. Stories
about hurting someone were significantly less likely to include any
description of the target’s thoughts than to include the narrator’s own
thoughts, t(60) = 3.29, p < .01. In contrast, the stories about making
someone happy devoted roughly equal coverage to the two characters’
thoughts, t(60) = 1.02, ns.
Whatdo allthese patterns mean?The tendencyfor transgressorsto
emphasize their own emotions emerged as a notable and unusual fea-
ture of these results. The transgressors’ discussion of their own emo-
tions,andespecially theirnegative emotions, maycontribute to reduc-
ing guilt and avoiding unfavorable impressions. The more the story
dwells on the distress and suffering of the victim, the worse the narra-
torlooks. Therefore itis notsurprising thattransgressors devotedless
space and less emphasis to the victims’ feelings. That seems to be gen
erally what transgressors do, in order to reduce their guilt (Baumeis
owndistress,one canfurthermoremake aclaimthat whateveronedid
or said was done in the heat of passion and hence should not be con
demned as severely as a similar act that had been done in a calm, con
siderate state of mind (e.g., Averill, 1982).
Similarly, by suggesting that he orshe suffered during the episode,
thetransgressor canperhaps portray himselfor herselfas afellow vic
tim, and as deserving of sympathy instead of punishment. Orbuch,
Harvey, Russell, and Sorenson (1992) found that people who read
accounts of romantic breakups liked the storyteller more when he or
she expressed feelings of distress over the breakup, rather than the
ing may indeed be an effective technique for obtaining a positive reac
tion from the reader or listener. Transgressors may be able to reduce
theappearanceof being evilbytelling howbadlytheyfelt. Thegeneral
stereotype of evil, from religious stories to modern movies and
282 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
cartoons, involves villains who derive pleasure and satisfaction from
inflicting harm on others (Baumeister, 1997). Insisting that one did
not getpleasure and in fact suffered distress and pain may help trans
gressors distance themselves and their actions from that stereotype.
Limitations. Several limitations of the present work must be noted.
First, the participants were college students, and as such they may
have more facility with language than less well-educated people. One
should therefore be cautious about generalizing from these findings to
the linguistic and speech patterns that less well-educated people
We had asked people to describe their most important transgres
sions to get ahomogeneous and well-remembered sample of stories. It
is certainly possible that less extreme transgressions would be told
usingfewer of thelinguistic devices observedhere. In asimilar vein, it
cally relative. Concern over protecting one’s image and self-esteem is
ple in the United States (many of whom even had programs and
courses in school aimed at boosting their self-esteem). It is plausible
that cultures or circumstances that would support self-criticism or
humility would not elicit such defensive maneuvers.
Furthermore, personality traits may moderate how defensive peo-
tive self-views may readily accept wrongdoing on their part (e.g.,
Blaine & Crocker, 1993). On the other hand, people with high self-
esteemor even inflatedself-views, as they are measuredin narcissism
scales, may be very reluctant to admit that they acted wrongly (e.g.,
Baumeister, Smart & Boden, 1996; Schütz, 1997). Effects of mood or
gender were not observed with this sample, however. Another poten
tial boundary condition is that our research participants provided
written stories. Stories told to others orally may follow different rules
(see Schütz & DePaulo, 1996). Subjects may, for example, anticipate
the audience’s reactions and want to avoid appearing defensive. They
may, therefore, employ fewer defensive strategies.
dent in which the participant had hurt someone. It is conceivable that
other types of transgressionswould yield different results. To be sure,
hurting someone is one prototype of a transgression. Still, victimless
sors downplayed the victim’s thoughts and feelings would have no
direct parallelin victimless transgressions. Converging evidence from
studies about other kinds of transgressions or, indeed, about other
incidents that could reflect badly on the narrator’s image would be
valuable to extend and corroborate the present findings.
Schütz, Baumeister / LANGUAGE OF DEFENSE 283
asked to describe their “most important” transgression, than when
asked to describe their most important act of helping someone could
conceivably contribute to the present results, although this concern is
less pressing with the present focus on linguistic patterns than it
would be with other instructions. For example, studies that compare
victim and perpetrator narratives (e.g., Baumeister et al., 1990) must
grapple with the possibility that victims and perpetrators systemati
same events in different terms. Then again, a selection bias would
probably have worked against the present results, because if people
could choose events that did not make them look bad, they would not
have had to resort to various linguistic devices to protect the self from
unsavory implications. The patterns of results we found might well
have been substantially stronger if people could not use any choice as
to what incident to describe, because they would have had to rely all
the more on linguistic devices. In any case, it would be desirable for
some future researchers to find a circumstance in which people are
constrained to describe a particular incident (chosen by the experi-
menterrather than theparticipant), inorder toeliminate anypossible
role of personal selection (cf. Stillwell & Baumeister, 1997).
A last issue is the effectiveness of the defensive strategies we found
(see Baumeister, Dale, & Sommer, 1998, on defense mechanisms).
Although the linguistic shifts appear to be motivated because they
able image, they may or may not be successful. Success could be meas-
linguistic devices to avoid facing up to his or her own guilt, or in terms
of impressions made on others. Future research is needed to address
the question of how effective the various linguistic devices observed in
thisstudy are inmanipulating observers’ impressionsand thus reduc
ing the attribution of blame to the narrator.
Concluding Remarks. The present investigation provided prelimi
nary evidence about how linguistic patterns may be altered in response
to the predicament of having to describe an incident that could make
one look bad. We found that people employed a series of linguistic
devices when telling a story in which they hurt someone, and that
the self from the most damaging implications of the story being told.
We assume that differences in sentence length, suppression versus
inclusion of identifying details, relative coverage of the emotional
states of self and other, and similar patterns do not reflect conscious
strategies. Rather, our participants were probably conscious of the
predicament of having to describe a transgression that could make
them look bad, and the changes in sentence length and so forth
284 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999
emerged as a result of their concern overlooking bad.Still, thedegree
to which the changes we observed were automatic processes and
byproducts, as opposed to deliberately cultivated ploys, remains for
further study. If people were indeed unaware of how their speech pat
ent findings constitute a dramatic illustration of the skill with which
people can manipulate language to protect their favorable images of
self, even while revealing their misdeeds.
Arcuri,L.,Maass,A., &Portelli,G. (1993).Linguisticintergroupbiasinimplicitattribu
tions. British Journal of Social Psychology, 32, 277-285.
Averill, J. (1982). Anger and aggression: An essay on emotion. New York: Springer-
Baumeister, R. F. (1997). Evil: Inside human violence and cruelty. New York: W.H.
Baumeister, R. F. (1998). The self. In D. T. Gilbert, S. T. Fiske, & G. Lindzey (Eds.),
Handbook of social psychology (4th ed.; pp. 680-740). New York: McGraw-Hill.
Baumeister, R.F., & Cairns,K. J. (1992).Repression and self-presentation:When audi-
ences interferewith self-deceptive strategies.Journal of Personalityand SocialPsy-
chology, 62, 851-862.
Baumeister, R. F., Dale, K., & Sommer, K. L. (1998). Freudian defense mechanisms
displacement, undoing, isolation, sublimation, and denial. Journal of Personality,
Baumeister, R. F., Smart, L., & Boden, J. M. (1996). Relations of threatened egotism to
violence and aggression: The dark side of high self-esteem. Psychological Review,
of interpersonal conflict: Autobiographical narratives about anger. Journal of Per
sonality and Social Psychology, 59, 994-1005.
Berry,D. S.,Pennebaker, J.W., Mueller,J. S.,& Hiller,W. S.(1997). Linguisticbasesof
social perception. Personality and Social Psychology Bulletin, 23, 526-537.
Blaine, B. & Crocker, J. (1993). Self-esteem and self-serving biases in reactions to posi
tive and negative events: An integrative review. In R. F. Baumeister (Ed.), Self-
Esteem: The puzzle of low self-regard (pp. 55-86). New York: Plenum.
taterentlastende und opferfeindliche Überzeugungen im Bereich sexueller Gewalt
[Myths about rape. Social psychological studies on attitudes that diffuse perpetra
tors’ responsibilities and blame victims in the sphere of sexual violence]. Landau:
Fiedler,K.& Semin,G. (1996).Language in AppliedContexts. InG. Semin& K. Fiedler
(Eds.), Applied Social Psychology. London: Sage.
Frey, D. (1986). Recent research on selective exposure to information. In L. Berkowitz
Gonzales, M. H., Pederson, J. H., Manning, D. J., & Wetter, D. W. (1990). Pardon my
gaffe:Effectsofsex, status,andconsequenceseverityon accounts.Journal ofPerson
ality and Social Psychology, 58, 610-621.
Lamb,S. (1991).Actswithoutagents:An analysisoflinguistic avoidancein journalarti
cles on men who batter women. American Journal of Orthopsychiatry, 61, 250-257.
Schütz, Baumeister / LANGUAGE OF DEFENSE 285
Lamb, S. & Keon, S. (1995). Blaming the perpetrator: Language that distorts reality in
newspaper articles on men battering women. Psychology of Women Quarterly, 19,
Maass, A., &Arcuri,L. (1992).The roleof languagein thepersistence ofstereotypes. In
G. R. Semin & K. Fiedler (Eds.), Language, interaction, and social cognition
(pp. 129-143). London: Sage.
Maass, A., Arcuri, L., Salvi, D., & Semin, G. R. (1989). Language use inintergroup con
texts. Journal of Personality and Social Psychology, 57, 981-993.
ferential expectanciesor in-group protection?Journal of Personalityand Social Psy
chology, 68, 116-126.
McNeil, B. J., Pauker, S. G., & Tversky, A. (1988). On the framing of medical decisions.
and prescriptive interactions (pp. 562-568). Cambridge, UK: Cambridge University
Orbuch, T. L., Harvey, J. H., Russell, S., & Sorenson, K. (1992). Person perception
through accounts: Three studies. Journal of Social Behavior and Personality, 7,
Schlenker, B. R. (1980). Impression management: The self-concept, social identity, and
interpersonal relations. Monterey, CA: Brooks/Cole.
Schmid, J., & Fiedler, K. (1996). Language and implicit attributions in the Nuremberg
Trials:Analyzingprosecutors’ anddefense attorneys’closingspeeches.HumanCom-
munication Research, 22, 371-398.
Schönbach, P. (1985). Account episodes: The management and escalation of conflict.
Cambridge, UK: Cambridge University Press.
Schütz, A. (1991). Mehr oder weniger zugeben: Stufen defensiver Selbstdarstellung
[Admitting more or less: Levels of defensive self-presentation]. In H. U. Kohr &
M. Martini (Eds.), Bewusstsein–Macht–Identität: Europäische Beiträge zur Poli-
tischen Psychologie. Weinheim, Germany: Deutscher Studienverlag.
Schütz, A. (1993). Self-presentational tactics used in a German election campaign.
Political Psychology, 14, 471-493.
Schütz, A. (1997). Interpersonelle Aspekte des Selbstwertgefühles. Die Beschreibung
der eigenen Person im sozialen Kontext [Interpersonal aspects of self-esteem.
Describing oneself in social context]. Zeitschrift für Sozialpsychologie, 28, 92-108.
speak for themselves. Journal of Research in Personality, 30, 137-156.
Semin, G. R. & Fiedler, K. (1988). The cognitive functions of linguistic categories in
describingpersons: Socialcognition andlanguage. Journalof Personalityand Social
Psychology, 54, 558-568.
Semin, G. R., Rubini, M., & Fiedler, K. (1995). The answer is the question: The effect of
verb causality on locus of explanation. Personality and Social Psychology Bulletin,
Stillwell, A. M., & Baumeister, R. F. (1997). The construction of victim and perpetrator
memories: Accuracy and distortion in role-based accounts. Personality and Social
Psychology Bulletin, 23, 1157-1172.
Sullivan, H. S. (1953). The interpersonal theory of psychiatry. New York: Norton.
ric: A time-series analysis. Journal of Personality and Social Psychology, 49,
Wyer, R.S., & Frey,D. (1983).Theeffects offeedbackabout selfand others onthe recall
and judgments of feedback-relevant information. Journal of Experimental Social
Psychology, 19, 540-559.
286 JOURNAL OF LANGUAGE AND SOCIAL PSYCHOLOGY / September 1999