Language, Meaning, And Social Cognition

1Language, Meaning, and Social CognitionThomas M. HoltgravesBall State University

Yoshihisa KashimaUniversity of Melbourne

in psychology, the role of language has not received thefocal attention that it deserves in social cognition.

There have been earlier reviews of the role of languagein social psychology (Clark, 1985; Holtgraves, 2002;Krauss & Chiu, 1998; Krauss & Fussell, 1996), includingspecific considerations of the relationship between lan-guage and social cognition (Chiu, Krauss, & Lau, 1998;C. Hardin & Banaji, 1993; C. D. Hardin & Higgins,1996; Semin, 2000). Our goal with this review is toupdate these prior reviews and, more importantly, to pro-vide a new framework for a consideration of language byfocusing explicitly on its role as a mechanism for the cre-ation and exchange of meaning. Specifically, we argue thatin social psychology, and especially in social cognition,language is most productively conceptualized as a semiotictool, namely, a tool for meaning making and meaningexchange in imagined or real social interaction. In whatfollows, we first outline our stance by bringing into focushow meaning is critical in human social interaction, howsocial cognition was meant to be an examination of mean-ingful social interaction, and how language is centrallyinvolved as a tool in the process of constructing andexchanging meaning. We then point out that language usehas unintended consequenceswe illustrate this by show-ing how a particular instance of speakers language usecan inadvertently affect the thoughts and actions of thespeakers themselves, the recipients, and the collective thatinclude both. We then consider long-term unintended con-sequences of language use and argue that language usemay play a significant role in the mediation of cultural dif-ferences in a variety of social cognitive phenomena.

Authors Note: This research was supported by a grant from theNational Science Foundation (0131877) to the first author and facili-tated by a grant from the Australian Research Council (DP0450518)to the second author. Correspondence should be addressed to ThomasM. Holtgraves, Department of Psychological Science, Ball StateUniversity, Muncie, IN 47306; e-mail: [email protected].

PSPR, Vol. XX No. X, Month XXXX xx-xxDOI: 10.1177/1088868307309605 2007 by the Society for Personality and Social Psychology, Inc.

Social cognition is meant to examine the process of mean-ingful social interaction. Despite the central involvementof language in this process, language has not received thefocal attention that it deserves. Conceptualizing meaning-ful social interaction as the process of construction andexchange of meaning, the authors argue that languagecan be productively construed as a semiotic tool, a toolfor meaning making and exchange, and that language usecan produce unintended consequences in its users. First,the article shows a particular instance of language use tobe a collaborative process that influences the representa-tion of meaning in the speaker, the listener, and the col-lective that includes both the speaker and listener. It thenargues that language use and social cognition may havereciprocal effects in the long run and may have significantimplications for generating and maintaining cultural dif-ferences in social cognition.

Keywords: language; meaning; social cognition; thought;communication; culture

Many of the processes that are most central to socialcognitionattribution, person perception, stereo-typing, and so oninvolve language in some manner.People use language to communicate to one another(and to researchers) their attributions, perceptions, andstereotypes, for example, with language use sometimesshaping the very products being communicated. At adifferent level, the intrapersonal processes throughwhich these products are generated may be mediated inpart by language. In this sense language is a tool forimplementing social cognition (Semin, 2000), and muchsocial cognition is acquired through the use of language.It is, in fact, difficult to think of any social-cognitiveprocesses that do not involve language in some manner.Clearly, the study of language can contribute greatlyto the understanding of social thought and action.Unfortunately, relevant empirical research on this topic isscattered over a number of different domains, and giventhe diversity (and technicality) of inquiries into language

HUMAN SOCIALITY, MEANING, AND LANGUAGE:LANGUAGE AS SEMIOTIC TOOL

Human sociality is saturated with meaning. Whensocial psychology is said to be about social behavior,what is meant by social behavior is not a sheer bodilymovement but rather a meaningful social action. Whena bodily movement carries meaning, it invites otherhumans and the actors own interpretations of the bod-ily movement. For example, when a school boys righteyelid contracts rapidly, it may be an involuntarytwitch, a wink, or even a mockery of someones clumsyattempt at winking (Ryle, 1971). What people react tois what they think he meant to do, not what he actuallydid. As far as the bodily movement goes, it is all thesamea rapid contraction of an eyelid. Depending onwhether observers think it is a twitch, a wink, or amockery, they would react by ignoring, guessing whatthe winker meant, or hatching a plan to retaliate themockery. When people engage with others, they do soby behaviors, but the others meanings are critical totheir social engagement (Geertz, 1973).

The question of meaning making and meaningexchangehow people engage in meaningful thought andaction in social contextsis a central question of social cog-nition. Early forebears like Bartlett (1932; see Y. Kashima,2000, on Bartletts social psychology) were centrally con-cerned about this. It is related that the construction andcommunication of meaning lies at the heart of symbolicinteractionism (Blumer, 1969; Mead, 1934). More recently,Ostrom (1984) in his manifesto, The Sovereignty of SocialCognition, argued that social cognition conceptualizeshumans as interaction partners who interpret each othersbehaviors in a stream of social interaction. And Bruners(1990) call for cultural psychology is also a more recentadvocacy for a psychology that takes meaning seriously.Put differently, from its inception, social cognition as aresearch tradition was meant to be a scientific examina-tion of the construction and exchange of meaning in socialcontexts. And in this process of meaning making andexchange, language is centrally involved.

Language in its abstract sense is a socially shared toolallowing its users to create and exchange meaning. In sodoing, the users of any particular language are connectedto one another through this shared symbolic system. Inthis way language provides a means for connecting indi-vidual and social levels of analysis, a perspective in linewith that explicated by Wertsch (1991), who integratedVygotskys (1962) and Bakhtins (1981) perspectives inhis theory of cultural psychology (see Fiedler, Bluemke,Freytag, Unkelbach, and Koch, in press, for a more recentsemiotic approach to communication). In Wertchs per-spective, higher mental functioning is derived from sociallife, an internalization of the capacity to engage in dialogue.

Hence, all cognition is essentially social cognition, cog-nition that is in some sense connected to the real orimagined presence of others. Language is the symbolicsystem that makes this possible.

Although language allows for the creation andexchange of meaning, it does not guarantee completeand mutual understanding. This is because meaningdoes not reside solely in the message. Rather, languageis used to create meaning and its users must be activeparticipants in this process. Speakers intend to convey aparticular meaning and have the recipient recognize thatcommunicative intention (Grice, 1957; Holtgraves,2007). This is what speech-act theorists refer to as illo-cutionary force (Searle, 1969). However, the effects of aspeakers utterance (what speech-act theorists term itsperlocutionary effects) may extend beyond what thespeaker intended to communicate (Holtgraves, 2005).Although much research has explored the intended con-sequences of language, it is the unintended conse-quences with which we are most interested.

At this point, let us draw a distinction between two typesof language use as a semiotic tool. One is a specific instanceof language use. Take Lakoffs (1987, p. 77) example andimagine a waitress saying to her colleague, The ham sand-wich just spilled beer all over himself. The use of the nounphrase the ham sandwich in this utterance stands for theperson who is eating the sandwich. A specific, one-offinstance of language use as a semiotic tool like this may becalled a token language use, or what Saussure (1966) calledparole. Over time, however, this type of language use maybecome prevalent at this restaurant, so much so that mostwaitresses use this type of constructionthe use of an orderto refer to the customermost of the time for most cus-tomers. This may be called a type of language use. A tokenis to a type as an exemplar is to a category. If a language usetype occurs prevalently, then, it may be said that it is part oflanguage as an abstract symbolic system, or what Saussure(1966) called langue. Such frequent and widespread use ofa certain feature of language as a semiotic tool has beencalled a linguistic practice (Y. Kashima, Kashima, Kim, &Gelfand, 2006).

Just as any tool can help its users achieve their goals ifused appropriately, language helps people achieve theirgoals. People do things in saying. In this sense, languageuse has intended consequences (e.g., illocutionary force).However, the semiotic tool perspective helps us highlightanother side of language use. When speakers use a lan-guage, they may inadvertently end up doing certainthings but not other things. In other words, language usemay have unintended consequences. There are twoways this can come about. One class of unintended con-sequences has to do with consequences of token lan-guage use. As Chiu et al. (1998) have argued, usinglanguage on any specific occasion can have unintended

2 PERSONALITY AND SOCIAL PSYCHOLOGY REVIEW

social-cognitive consequences. We extend this analysis byarguing that these consequences occur for (a) speakers,(b) recipients, and (c) a collective consisting of the speak-ers and recipients. The second class of unintended conse-quences has to do with how features of a language, orcommon linguistic practices, influence social cognitiveprocesses. We argue that if a type of language use occursprevalently, then specific instances of this language usetype can have long-term effects on social cognition.

Figure 1 summarizes our theoretical assumptionswhile depicting the relationships between a specificinstance of token language use and the totality of lan-guage use types (language as an abstract system). Theunderlying assumptions are that (a) speakers and listen-ers cooperate with each other to construct a particularinstance of language use during their social interaction,(b) this particular instance of language use affects theircognitions, (3) this type of language use may be soprevalent that it can be understood as part of the total-ity of the language system, and (4) at this point the lan-guage system can be said to influence the socialcognitions of its users. The following sections are orga-nized in accordance with the two time frames of unin-tended consequences of language use. We first addressthe microinteractional level and short-term implicationsof (a) and (b) by arguing that different aspects of lan-guage affect speakers and listeners social cognitionthrough its use as a semiotic tool for meaning produc-tion and exchange. We then broaden the time frame andconsider the ways in which language as an abstract sys-tem can be said to influence the social cognitions of itsusers through linguistic practices (c and d).

LANGUAGE USE AND THE CREATION OFMEANING

In the perspective of language as a semiotic tool,the process of meaning making by language in social

interaction can have important cognitive consequences(Chiu et al., 1998; C. Hardin & Banaji, 1993). In this sec-tion we focus on the cognitive consequences of using lan-guage on a particular occasion in which individuals taketurns assuming the roles of speaker and recipient. Both thespeaker and the recipient coordinate their activities toestablish a mutual understanding by using language, andthis process is highly collaborative (e.g., Clark, 1996).

In a particular instance of producing an utterance, aspeaker engages a number of cognitive and social systems(Levelt, 1989). Oftentimes the first step in this processwill involve the creation of a linguistically coded repre-sentation from a nonlinguistically coded representation, aprocess that can influence the nature of the resulting rep-resentation. By linguistic representation we are referringto an internal, conceptual entity with language-basedconstituents. Nonlinguistic representation, on the otherhand, refers to internal, nonconceptual entities that arenot comprised of language-based constituents.

This process may be akin to the conversion of animplicit attitude into an explicit attitude. Gawronskiand Bodenhausen (2006) in a recent theoretical integra-tion suggested that implicit and explicit attitudes mayresult from associative and propositional processes, whichare driven by different principles. Associative processesare based on spatio-temporal contiguity; propositionalprocesses follow syllogistic reasoning. If nonlinguistic rep-resentations engage associative processes that supportimplicit attitudes, linguistic representations are closelyaligned with propositional processes that support explicitattitudes. Whether these involve different processes asGawronski and Bodenhausen claimed they do or have oneunderlying process (Albarracn, Hart, & McCulloch,2006; Kruglanski & Deschesne, 2006), we suggest thatlanguage use frequently involves the recoding of implicit,nonlinguistic representations into explicit, linguistic ones.Note that a person does not need to actually verbalize thesethoughts aloud, simply creating a linguistic representationmay be sufficient (Vygotskys, 1962, inner speech).

The creation of a linguistic representation that is tobe conveyed to a recipient via an utterance will involveadditional social and collaborative processes that fur-ther influence the nature of those representations. Thisis because meaning construction is a collaborative activ-ity as interlocutors mutually orient to each others per-spectives and construct and interpret utterances fromthese perspectives (e.g., Clark, 1996). As interlocutorsconverse, they provide evidence (e.g., nonverbal behav-iors such as a nod and a smile, simple verbal utteranceslike yeah or OK, and more extensive verbal utter-ances that paraphrase the speakers utterance) of mutualunderstanding (Clark, 1996), and when they mutuallyrecognize that they have reached an understanding that issufficiently similar for the current purpose, they are said

Holtgraves, Kashima / LANGUAGE AND SOCIAL COGNITION 3

Figure 1 Proposed relationships between language use and cog-nition over time.

Instance ofLanguage Use at t

Language at t+1 Language at t

SpeakerCognition

RecipientCognition

Interaction Episode at t Time

to have grounded the utterance (Clark, 1996). This collec-tive representation may be conceptualized as each com-munication partner having a representation that weshare our representation (Kashima, Klein, & Clark,2007; Searle, 1990), thus establishing common ground.The collective representation generated through the col-laborative use of language results in a shared reality (C. D.Hardin & Conley, 2001; C. D. Hardin & Higgins, 1996;Higgins, 1992). It is important to note, however, thatalthough interactants strive to establish common ground,a state of mutual understanding is not always achieved(e.g., Holtgraves, 2005). What is crucial from our perspec-tive is that interactants try to achieve common ground andthat they actively collaborate in the creation of meaning.It is this process that plays a crucial role in the unintendedconsequences of language use.

We begin with a discussion of the effects of languageuse on a speakers internal representations, that is, howthe creation of a linguistic message can influence thenature of ones thoughts. This is followed by a consid-eration of how the collaborative and interpersonalnature of language use affects the process of communi-cation with a corresponding impact on the recipientsinternal representations. Finally, we turn to collectiveconsequences of language use, that is, the production ofwhat we call collective representations.

LANGUAGE USESPEAKER EFFECTS

Similar to the communication of an impression to anaudience in the cognitive tuning experiments (Zajonc,1960), language is an essential semiotic tool for convert-ing a speakers inchoate experience into an explicit andcommunicable form. The creation of an explicit linguis-tic representation engages processes that may alter exist-ing representations in the speaker. In other words, anewly created linguistic representation does not simplyoverlay an initial nonlinguistic representation. Rather, itis a new and different representationthe creation ofnew meaningthat may alter the nature of the initialnonlinguistic representation as well as related content.Here we consider three areas of research that have demon-strated this unintended consequence of language use.

Decision Making

Language is a tool with which people explicate theirreasons for decisions. Wilson and colleagues have con-ducted a number of studies demonstrating that verbaliz-ing reasons for a decision can influence various aspectsof the decision-making process, including the quality ofthe decision itself (Wilson & LaFleur, 1995; Wilson,

Lisle, & Kraft, 1990; Wilson et al., 1993; Wilson &Schooler, 1991). The essential comparison in these experi-ments is between participants who make a decision andmust verbalize their reasons for that decision and partici-pants who make a decision but are not required to verbalizetheir reasons for that decision. For example, Wilson andSchooler (1991) asked participants to judge the quality ofdifferent brands of strawberry jam (Experiment 1) or toindicate their preferences for different college courses(Experiment 2). In both experiments, participants whowere asked to verbalize the reasons for their decisionsmade less optimal decisions than did participants who didnot verbalize their reasons.

The general interpretation of these findings is that ask-ing people to articulate their reasons for a decision causesthem to attend to nonoptimal criteria. That is, the creationof a linguistic representation of a reason for a decisionengages processes that alter the nature of that decision.More specifically, reasons that are the easiest to verbalizeare the ones most likely to be articulated, and other rea-sons that may be important, yet are more difficult to ver-balize, will tend not to be included. The reasons that areverbalized drive the decision and lessen its quality. Forexample, in a study conducted by Wilson et al. (1993),participants rated a set of posters that were either humor-ous (easy to verbalize features) or artistic (difficult to ver-balize features). For participants who were asked toverbalize their reasons, the ease with which poster attrib-utes could be articulated affected the reasons that weregiven; more positive attributes were articulated for thehumorous poster than for the artistic poster. Ultimately,this difference affected participants ratings; people whohad generated reasons preferred the humorous posters;control participants preferred the artistic posters.

More recently, the research of Dijksterhuis and col-leagues (Dijksterhuis, 2004; Dijksterhuis & Nordgren,2006) has demonstrated a similar pattern of results interms of conscious versus unconscious thought. Forexample, Dijksterhuis and van Olden (2006) conducteddecision-making experiments in which some partici-pants were kept from engaging in conscious (and hencelanguage-based) thought. These participants made moreoptimal decisions than did participants who wereallowed to engage in conscious thought. One of the rea-sons for this difference is that language-based consciousthought involves a less than optimal weighting scheme,a tendency to weight more heavily those factors thatare accessible and easy to verbalize (similar to Wilsonet al.s, 1993, research). In a way, it is a case of the tailwagging the proverbial dog, or a semiotic tool drivingones decision. A tool is meant to help users achievetheir goals but in this case its use can have additionalunintended collateral consequences.


Emotions

Language is a primary means by which emotions arecommunicated after they have dissipated. However,verbally identifying emotions (in others and in oneself)can create or alter the nature of those emotional repre-sentations. For example, Halberstadt (2003) demon-strated that the process of talking about anothersemotion can bias memory for the expression of thatemotion. In this study, all participants were shown pho-tos of people displaying ambiguous emotional expres-sions. Some of the participants were asked to explainwhy the person was experiencing that particular emo-tion; other participants were not asked to do this.Participants who explained the emotion rememberedthe faces as expressing that emotion more intensely rel-ative to participants who were simply told what emo-tion the person was experiencing. Using language toexplain the emotion forced participants to decomposethe face into specific features that were then reinte-grated in terms of the relevant emotional category.Hence, using language to describe anothers emotionaltered the underlying representation of that emotion.

The use of language not only plays a role in the per-ception of others emotions, it also plays a fundamentalrole in the perceptions of ones own emotions. This is nota new idea. Spinoza (1675/1949) wrote, Emotion,which is a passion, ceases to be a passion as soon as weform a clear and distinct idea, thereof. More recently,Lieberman (in press) has argued that affective labelinghas the consequence of activating the reflective neuralsystem (C-system) and dampening the reflexive neuralsystem (X-system). Hence, when one uses language tolabel an emotional state, the underlying physiology ofthat state is altered. More specifically, at least for nega-tively valenced images, linguistic processing of the imagelessens activation of the amygdala (Hariri, Bookheimer,& Mazziotta, 2000; Lieberman et al., 2007).

This is not necessarily a bad thing. For negative expe-riences, especially those that are traumatic, verbalizingor writing ones thoughts about the experience can havea beneficial effect on overall well-being (Lyubomirsky,Sousa, & Dickerhoof, 2006). According to Pennebaker(Pennebaker & Greybeal, 2001; Pennebaker & Seagal,1999), talking and writing about ones experiencesallows one to organize and structure those experiences,thereby allowing one to complete and hence bring thosenegative experiences to an end. Recently, Lyubomirskyet al. (2006) pitted the effects of the writing and talk-ing about an experience versus only thinking about theexperience. They found enhanced satisfaction with lifefor those who wrote or talked about their experiencesand diminished satisfaction with life for those instructedonly to think about their past experiences. It appears

that the use of language (affective labeling) to expressreactions to negative experiences has the effect of less-ening the automatic rumination that might prolong neg-ative affectivity (Lieberman, in press). Hence, journalwriting is an activity that has been recommended bytherapists (Pennebaker, 1997).

Using language as a semiotic tool enables people tocommunicate emotions but it has the perlocutionaryeffects of affecting the communicators representations andaltering the meaning of the emotional experience. Althoughit may distort the communicators memories in someinstances, it may help them regulate their emotionsespecially those related to traumatic experiences.

Story Telling

Story telling is a universal method by which humansmake their inchoate experiences intelligible to others andto themselves. Language is centrally involved in this activ-ity that researchers have argued has important cognitiveconsequences (e.g., Marsh, 2007; McAdams, 1993;Pennington & Hastie, 1988; Schank & Abelson, 1977,1995). Consider first the manner in which an initial repre-sentation is transformed into a story. Many times the con-struction of a story will involve the glossing over of details(everything cant be included in a story) and the forcing ofa sequence of events into a story-like structure. In otherwords, the construction of a story is influenced by a story-telling schema (Rumelhart, 1975). Consequently, the rep-resentation that remains is a function of the act ofverbalizing the story (Schank & Abelson, 1995). A studyconducted by McGregor and Holmes (1999) illustrates thiswell. They had participants tell a biased story about a rela-tionship conflict as if they were a lawyer for one of the twocharacters. Later, participants memory for and judgmentsof the targets were assessed. In all studies there was clearevidence of a storytelling effect; participants judgmentswere biased in the direction of the story that they told.

It is perhaps more important that story telling illus-trates the collaborative nature of language use and howinternal representations may be jointly constructed.That is, stories are told to other people in a social contextand frequently story tellers will tailor their tales for theirspecific audience. Clark and Murphy (1982) called thisaudience design (also see Fussell & Krauss, 1989). Thisrecoding of the originally linguistic stimuli into anotherform of linguistic representation can then influence sub-sequent memory. Because people talk about the past,and because they frequently do so in the context of others,the act of talking about the past can influence onesrepresentation of the past. Pasupathi (2001; Pasupathi,Alderman, & Shaw, 2007) has made this point explicit,arguing that what people know is based, in part, on what


they tell. Two principlescoconstruction (interactantsmutually influence the construction of stories about thepast) and consistency (conversational reconstructioninfluences later memory)are invoked as mechanismsexplaining how the act of talking about the past willinfluence later memory for those past events. In thisway, the act of telling stories can play a critical role inthe development of the self as well as play a role in thestability of the self (McLean, Pasupathi, & Pals, 2007).Along somewhat similar lines, social constructionists(e.g., Gergen & Gergen, 1988) have also argued thattalk can influence the nature of mental representations.

An example of this can be seen in the classic study con-ducted by Higgins and Rholes (1978). They had partici-pants describe a target person whose portrait wasevaluatively ambiguous (e.g., Donald who is adventurousor foolhardy) for an audience who presumably liked ordisliked the target being described. Participants alteredtheir descriptions based on their knowledge of the audi-ences preferences. When their memory and impressionsof the target were subsequently accessed, those who pro-vided more positive descriptions had more positiveimpressions and memory of the target. This was dubbedsaying is believing (SIB), and subsequent research hasreplicated the SIB effect (see McCann & Higgins, 1990,for a review). Similarly, Tversky and her colleagues(Dudukovic, Marsh, & Tversky, 2004; Marsh &Tversky, 2004; Tversky & Marsh, 2000) also showed theeffects of story telling on subsequent memory, pointing tothe robustness of the phenomenon.

In sum, the act of talking about ones decisions or emo-tions, or telling a story to others, can alter the speakersrepresentation of what has been talked about. Using lan-guage as a semiotic tool, communicators can convertinchoate psychological experiences into an explicit andcommunicable form to explicate ones reasons for a deci-sion, to communicate ones feelings, and to make intelligi-ble ones complex experience in narrative form. Theconstruction of a linguistic representationwhat previ-ously was implicit is made explicitenables speakers tocommunicate the referent of the representation to othersand to turn their cognitive capacities to those explicit rep-resentations as an object of cognitive activities. These lin-guistic representations may be altered further in order tofine-tune them for recipients in the service of communica-tive goals, and this tuning can have long-term cognitiveconsequences as well. We return to the collaborative con-structions of meaning below.

LANGUAGE USERECIPIENT EFFECTS

Language is a critical semiotic tool that speakers useto have their meaning and intention recognized (Austin,

1962). Recipients are oriented toward recovering thespeakers intention (Sperber & Wilson, 1995). Hence,much of the recipients cognitive activity is directedtoward uncovering this intended meaning (or illocution-ary force). In the process of doing this, however, manyunintended consequences (or perlocutionary effects)can occur. For social cognition, as we will see, what issignificant is how the recipients go beyond the informa-tion given, and different types of linguistic informationall invite recipients to construct different meanings.These effects can occur at the semantic, syntactic, andpragmatic levels.

Semantics

Recipients use their semantic knowledgewhat wordsmeanto construct mental representations of the objectsand events referenced in an utterance. In this profoundlyimportant, although apparently trivial, sense, languageuse affects the recipients mental representations. AsAllport (1954) noted about the role of language in preju-dice, calling someone a fag instead of gay makes a differ-ence in the recipients representation of that person(Carnaghi & Maass, in press; Greenberg and Pyszczynski,1985; Kirkland, Greenberg & Pyszczynsky, 1987; Simon& Greenberg, 1996). Traditionally, it is grammaticallycorrect for the speaker to use words such as men and hisin the generic sense of people in general; however, recipi-ents appear to understand them as males (Ng, 1990). Evencommon first names such as John and Joan can create dif-ferent impressions about a person (Kasof, 1993).

At a different level, one of the more subtle semanticaspects of language use is the meaning of a verb; differentverbs imply different kinds of causalitya phenomenontermed implicit causality. It was in the late 1960s and early1970s (Abelson & Kanouse, 1966; Garvey & Caramazza,1974; Kanouse, 1972; McArthur, 1972) that implicitcausality was noted in the literature. Brown and Fish(1983), however, are probably most responsible for bring-ing the phenomenon to the attention of psychologists. Ingeneral, implicit causality refers to a tendency for (actionand state) interpersonal verbs to imply a particular causalfocus. For action verbs (e.g., help) people are more likelyto assign greater causal weight to the agent (person per-forming the action) than to the patient (person who is therecipient of the action) (Brown & Fish, 1983; Corrigan,1988). For example, upon hearing Bob helped Tom,they are more likely to believe that Bob is the cause of thisevent rather than Tom and that Bob is a helpful personrather than that Tom is a needy person. Note that this isnot a grammatical effect; the agent is assigned greatercausal weight regardless of whether it is the grammaticalsubject or object. For state verbs (e.g., like), people tendto assign greater causal weight to the person who brings


about the state (stimulus) rather than the person whoexperiences the state (experiencer) (Au, 1986; VanKleek, Hillger, & Brown, 1988). Upon hearing thatBob likes Tom, people are more likely to judge Tomas more responsible than Bob and that Tom is a likeableperson rather than that Bob tends to like people. Again,this is not a grammatical effect as it occurs for stateverbs with the experiencer in the subject role (termedexperiencer-stimulus verbs; e.g., Bob likes Tom) andthose with the stimulus in the in the subject role (termedstimulus-experiencer verbs; e.g., Bob impresses Tom).

Why does implicit causality occur? One possibility isthat verbs activate associations that then influencecausal judgments. All interpersonal verbs have associ-ated with them dispositional terms referencing the sen-tence subject, object, or both. However, these deriveddispositional terms are not symmetrical but rather par-allel quite closely the causal reasoning tendencies forthese verbs. Specifically, for action verbs (of the agent-patient variety) most of the dispositional terms refer-ence the agent rather than the patient (e.g., help helpful), thereby mirroring the tendency to perceive theagent rather than patient as the causal locus. And bothstimulus-experiencer verbs (e.g., impress impressive)and experiencer-stimulus verbs (e.g., like likeable)have dispositional terms referencing the stimulus ratherthan the experiencer. In an extensive analysis of approx-imately 900 English interpersonal verbs, Hoffman andTchir (1990) found that more than 90% of the actionverbs had dispositional terms referencing the agent andonly 25% of the terms referenced the patient. For stateverbs, more than 75% had terms referencing the stimu-lus and fewer than 50% referencing the experiencer.

This asymmetry suggests that implicit causality ismediated by these derived dispositional terms. That is,the use of a certain verb activates the related disposi-tional term which then influences (often unintentionallyfrom the speakers perspective) judgments of causality.Consistent with this, Hoffman and Tchir (1990) founda larger implicit causality effect for verbs with terms ref-erencing the expected causal locus than for those thatdid not have a dispositional term (but see Rudolph &Fosterling, 1997, for an alternative interpretation). Aseparate experimental test of this hypothesis by theseauthors yielded conceptually similar results, as did aseries of memory studies conducted by Holtgraves andRaymond (1995) in which participants were more likelyto recall the names of agents (rather than patients) andstimuli (rather than experiencers).

An alternative view of implicit causality is that it reflectsprimitive schemas for thinking about interpersonal causal-ity (Rudolph & Fosterling, 1997).1 Because of an agent-patient schema people tend to assign causality to agentsrather than patients, and because of a stimulus-experiencer

schema they tend to assign causality to stimuli ratherthan to experiencers. These schemas are assumed tooperate independent of language; in fact, their existenceis assumed to influence language (rather than the otherway around). Thus, a tendency to view agents asresponsible has resulted in the development of disposi-tional terms referencing agents rather than patients.

To a certain extent, the two mechanisms are not mutu-ally exclusive, and which one operates may be a functionof the task. Semin and Marsman (1994) demonstratedthat if participants are asked to make dispositional infer-ences, then the lexical hypothesis receives support; partic-ipants are more likely to make dispositional inferences asa function of whether there is a derived dispositional termfor the verb. In contrast, if participants are asked to makejudgments of causality, then the existence of the derived dis-positional term has relatively little effect on these judgments.More generally, these two views of implicit causality reflect thereciprocal relationship between language useand social cognition, a topic that we elaborate on below. Atendency to perceive agents as causally responsible mayinfluence the development of the lexicon (a diachroniceffect) such that there are more derived dispositionalterms referencing agents rather than patients. But thenthe existence of this linguistic asymmetry may influencethe thought processes of language users (i.e., produceunintended consequences) when they use language on aparticular occasion (a synchronic effect).

Syntax

In addition to the implicit semantics of verbs, predi-cate forms can vary in terms of their syntactic properties.This may encourage recipients to construct differentmental representations and give rise to additional unin-tended consequences. A case in point is the researchbased on Semin and Fiedlers (1988, 1991) linguistic cat-egory model. Their approach points to the role of differ-ent linguistic categories, including verbs and adjectives, inthe recipients construction of meaning. Their schemeincludes action verbs (amaze, anger) and state verbs (like,hate), both identical to the stimulus-experiencer andexperiencer-stimulus verbs, respectively, in the Brownand Fish (1983) scheme. Action verbs are subdivided intodescriptive action verbs (concrete and objective descrip-tions of observable events such as call, meet, visit, etc.)and interpretive action verbs (less objective and moreinterpretive verbs such as help, cheat, imitate, etc.). Theirmodel also includes adjectives, a category that representsthe endpoint of the abstractness continuum.

These categories differ in linguistic abstractness.Descriptive action verbs (e.g., talk) are the least abstractand revealing of the actor and the easiest to verify.Interpretive action verbs (e.g., help) are more abstract


and revealing but are less easy to verify. As a result, theyare subject to greater observer disagreement; it is lesseasy to identify a behavior as constituting help than abehavior as constituting talk. The continuum then con-tinues with state action verbs (e.g., surprise), followedby state verbs (e.g., like), and lastly adjectives (e.g., hon-est). As internal states, state verbs are less observableand hence less verifiable and agreed uponthan areaction verbs. State verbs are also more enduring thanare state action verbs and hence more abstract andrevealing about the person (an internal reaction thatpersists reveals more of an actor than one that is amomentary reaction to an event). Relative to verbs,adjectives (e.g., honest) are more abstract (they are nottied to a specific behavior) and reveal more about theperson being described. It is related that they are lessobjective and verifiable than verbs. Almost any particu-lar action can be described at different levels, and thepredicate types chosen to describe that action can haveimportant consequences. For example, Tom could bedescribed as hitting Bob (a descriptive action verb),or more abstractly as hurting Bob (an interpretiveaction verb), or more abstractly yet as hating Bob(a state verb). Or, rather than describing the action,Tom might simply be referred to as an aggressiveperson. Linguistic abstraction can play an importantrole in social perception. When speakers describe behav-iors in more abstract terms, their listeners form moreabstract representations of these behaviors as well (Maass,Salvi, Arcuri & Semin, 1989). In this way, variations inlinguistic abstractness can contribute to the developmentand maintenance of stereotypes about outgroups, a phe-nomenon that we discuss in more detail below.

Pragmatics

A speakers intended meaning cannot be derived fromsyntax and semantics alone; additional interpretiveprocesses are required. Traditionally, pragmatics repre-sents the domain of language that considers the mannerin which utterances are interpreted within a social andcultural context. This interpretive process can produceunintended consequences because interlocutors may dif-fer in their assumptions regarding the operation of cer-tain pragmatic rules (Holtgraves, 2005). There is researchdemonstrating that this can occur in psychology experi-ments because experimenters and participants sometimesmake differing pragmatic assumptions (Hilton, 1995;Schwarz, 1996), a phenomenon that occurs with somefrequency, as we describe below.

According to Grice (1975; see also Sperber &Wilson, 1986, 1995), all communication (including thecommunication of an experimenter in a psychologyexperiment) comes with a presumption of relevance. As

a result, participants in psychology experiments will gen-erally interpret the communications of an experimenter(including the stimulus materials used in the experiment)as if they were relevant, an assumption that may not havebeen intended by the experimenter. In other words, exper-imenters may provide participants with information to seeif they use it. But participants, following standard prag-matic rules, assume that all information is relevant and sothey use it. For example, in the classic Kahneman andTversky (1973) studies, participants were provided withdescriptions of a target person consistent with either alawyer or engineer stereotype. In addition, the number(base rate) of lawyers and engineers in the sample wasmanipulated. Participants were far more likely to predictthat the target was an engineer (or lawyer) when thedescription matched the engineer (or lawyer) stereotypeeven when the underlying base rate supported the otheroccupation. The usual interpretation of this finding is thatparticipants judgments were based almost exclusively onthe degree of similarity between the sample and a popula-tion and that participants failed to use appropriately thebase-rate information that had been provided.

What seems to be crucial here are the instructions.Participants were told the description was based on per-sonality tests administered by a panel of psychologists.This clearly represents individuating information ratherthan base-rate information, and because the experi-menter gives this information to participants as part ofthe experimental procedure it seems reasonable thatparticipants will assume this information is relevant (inGrices, 1975, sense) and should be used. Several stud-ies have demonstrated that if the presumed relevance ofthe individuating information is undermined in variousways (e.g., by telling participants the descriptions wererandomly generated), people are far less likely to use it(Igou & Bless, 2003; Krosnick, Li, & Lehman, 1990;Schwarz, Strack, Hilton, & Naderer, 1991).

Other judgmental biases can be interpreted in thisway as well. For example, in the classic Jones andHarris (1967) study, participants read an essay presum-ably written by another student and were then asked tojudge the essay writers true opinion regarding the issue.Some participants were told the essay writer had beeninstructed by the experimenter to adopt the positionconveyed (forced-choice condition); other participantswere told the writers position had been freely chosen(free-choice condition). Participants inferred the authorsposition to be in line with the position advocated in theessay even when that position had been assigned andhence was not diagnostic of the writers true opinion.Participants are given information to use in performing aspecific task (judging the essay writers opinion), and it isreasonable for them to assume that such informationshould be used (why else was it provided?). When the


relevance of the information is lessened, people aremuch less likely to use it. Wright and Wells (1988) repli-cated the Jones and Harris (1967) study but told someparticipants that the information given to them hadbeen randomly selected and, hence, might not be rele-vant for the judgment task they were to perform. In thiscondition the fundamental attribution error was dra-matically reduced.

Unintended pragmatic consequences can occur in psy-chology experiments in more subtle ways. Research par-ticipants may sometimes generate inferences to maximizethe relevance of the experimenters communications, andbecause of this, pragmatic processes may play a role inhow people respond to virtually any self-report measure.For example, Strack, Schwarz, and Wanke (1991) askedrespondents to rate their happiness and their satisfactionwith life, two questions that obviously are highly related.For some participants the two items were presentedtogether at the end of a questionnaire; for others the lifesatisfaction item was presented as the first item in anostensibly unrelated questionnaire. In the first situationthe second question would appear to be redundant (itviolates Grices, 1975, maxim of relevance) and so partic-ipants might reason that it must mean something differ-ent from the first. Why else would it be asked? Responsesto the two questions differed significantly when they werepart of the same questionnaire. Such a concern does notarise when the two items are part of two separate ques-tionnaires, and responses to the two items were almostidentical in this situation.

There are similar implications for the construction ofresponse scales. Respondents assume that responsescales are relevant in the Gricean (1975) sense, that is,they are a meaningful component of the questionnaire.Hence the numerical value of rating scales may be usedby respondents to interpret the meaning of the scalesresponse options (Schwarz, 1996; Schwarz, Hippler,Deutsch, & Strack, 1985; Schwarz et al., 1991).

In short, language use involves the production andcomprehension of meaning within a web of pragmaticrules. Communication can be less than perfect andresult in unintended consequences because of differingpragmatic assumptions. In the setting of a psychologyexperiment, this can result in unwarranted inferencesregarding the relevance of the experimenters communi-cation as well as the stimulus materials (for moredetailed discussions of laboratory effects that can beinterpreted in this way see Hilton, 1995; Hilton &Slugoski, 2001; Holtgraves & Anderson, in press; Schwarz,1996). Of course such unintended consequences are notlimited to experiments (see Holtgraves, 2005, for ademonstration of how speakers and recipients can differsystematically in their interpretations).

LANGUAGE USECOLLECTIVE EFFECTS

The acts of both producing and comprehending lan-guage can alter, in significant ways, the mental repre-sentations of speakers (via production) and recipients(via comprehension). Nonetheless, the social cognitiveimplications of language use go beyond the speakersand recipients individual cognitions. For example, thetelling of a story (what gets told and how it is expressed)is heavily influenced by the audience, an influence thatcontinues during the course of the story-telling episode.In this way stories are jointly constructed and the inter-nal representation of the story teller and the audienceare collectively altered by the story-telling episode. It isnot just story telling that is interactive in this sense;almost all talk can be viewed as being jointly con-structed. For example, many real-life social judgmentsare elicited in a conversational context and these judg-ments are a result of this collaborative process. People areasked their opinions and whether they believe somethingis likely to occur, and so on. The manner in which thequestion is asked and the speakers beliefs about the stateof the questioners knowledge can influence the answersthat are given, thereby altering the representations of theinteractants. For example, the attributions that peoplegive in a conversational setting are sometimes based onassumptions regarding the recipients knowledge(Slugoski, Lalljee, Lamb, & Ginsburg, 1993); people tendto focus on and identify those causes of which the recip-ient is probably not aware. Similarly, people are morelikely to convey stereotype-consistent information if theybelieve the recipient does not possess that particularinformation (Lyons & Kashima, 2003). Judgments, then,are not simply the result of intrapersonal cognitiveprocesses but frequently the result of an interactionbetween those processes and certain relevant, pragmaticprinciples regarding communication. One consequence ofthis collaborative activity is that speakers and recipientscoconstruct collective representations (Y. Kashima et al.,2007). Collective representations in this context are rep-resentations (a) that are held by the individual languageusers involved in the communication, (b) that representthe content of the mutual understanding, and (c) that areassociated with collective identity, namely, representa-tions about the holders of the mutual understanding. Thespeakers and recipients are likely to believe this collectiverepresentation to be reliable and valid to the extent thatthe communicators trust each other as a source of accu-rate information and therefore trust the holders of thecollective representation (C. D. Hardin & Conley, 2001;C. D. Hardin & Higgins, 1996).

Echterhoff, Higgins, and Grolls (2005) recent workspeaks to this. As noted earlier, the SIB effect (Higgins


& Rholes, 1978) represents a social cognitive conse-quence of language use, in which speakers bias their mes-sages about a target person in line with their perceptionsof their audiences attitudes toward the target, and thebiased messages tend to further bias their impressionsand memories about the target. Echterhoff et al. showedthat the SIB effect depends significantly on the extent towhich the speakers trust their audience and expect to beable to and are motivated to establish a mutual under-standing with their audience. In particular, the SIB effectwas stronger when the recipient of the message was thespeakers ingroup member than when the recipient wastheir outgroup member, and this was explained by theextent to which the speakers trusted their audience. Theytrusted their ingroup audience more than their outgroupaudience, and the more they trusted, the stronger was theSIB effect. To put it differently, although language usemay have cognitive implications in and of itself, part ofthe social cognitive consequences of language use is likelydue to the collective implications of language usenamely, an establishment of a mutual understanding andthe production of collective meaning.

Intergroup Stereotypes

Nowhere is a collective consequence of language usebetter illustrated than in the area of intergroup stereo-types. Stereotypes about ingroups and outgroups are oftencommunicated and speakers must make choices about thewords and phrases used to describe different targets, aprocess that can mutually influence the interactantsstereotypes. These linguistic choices can play a role in theperpetuation of intergroup stereotypes, a phenomenonreferred to as the linguistic intergroup bias (LIB), or ten-dency to use differing levels of abstraction when describ-ing the positive and negative actions of ingroup andoutgroup members (Fiedler, Semin, & Finkenauer, 1993;Maass, 1999; Maass & Arcuri, 1996; Maass, Milesi,Zabbini, & Stahlberg, 1995; Maass et al., 1989).

An early demonstration of the LIB was provided byMaass et al. (1989). They showed members of differentriding groups in Italy pictures of cartoons (one halfsocially desirable and one half socially undesirable)involving members of these different riding groups,including each participants own club (and hence onesingroup) and other clubs (outgroups). Participants wereasked to choose an alternative that described the scene(Experiment 1) or to provide a free description of thescene (Experiment 2). When the cartoon depictedingroup members engaged in positive behaviors, partic-ipants were more likely to use adjectives, words thatreflect relatively enduring and stable qualities. Wheningroup members were depicted as engaging in negativebehaviors, participants described the behavior more

concretely (with descriptive action verbs), thereby impli-cating that the behaviors were unique and unstable andnot revealing of the personal qualities of the actor. Thistendency was reversed for cartoons depicting outgroupmembers; their positive behaviors were described con-cretely (descriptive action verbs) and their negativebehaviors more abstractly (although the results werestronger for the positive than for the negative behaviors).Note that this bias is not restricted to groups but occursat the individual level as well (Maass et al., 1995).

The LIB may be driven, in part, by a motivationaltendency to protect ones social identity. Evidence forthis comes from the tendency for the LIB to increasewhen ones ingroup is threatened (Maass, Ceccarelli, &Rudin, 1996). Also consistent with this motivationalview is the demonstration that people scoring high onthe need for cognitive closure (a desire to possess adefinitive answer on some topic) are more likely thanthose low on this dimension to use a high level of abstrac-tion when describing positive ingroup behavior and nega-tive outgroup behavior (Webster, Kruglanski, & Pattison,1997; see also Maass, Cadinu, Boni, & Borini, 2005).The LIB may also reflect a cognitive tendency to useabstract language for behaviors that conform to expecta-tions (linguistic expectancy effect; Maass et al., 1989;Wigboldus, Semin, & Spears, 2000). Because positivebehaviors are generally expected for ingroup members,there is a tendency to use more abstract language whenreferring to them. Evidence for this cognitive explanationhas been provided by studies demonstrating the use ofmore abstract language for expected behaviors regardlessof behavioral valence (Maass et al., 1995) and by Seminand Smiths (1999) finding of a relationship between lin-guistic abstractness and memory; the more temporallydistant an event, the more likely the event will be repre-sented with an abstract term. Both motivational andcognitive factors are likely in operation in the productionof the LIB (e.g., Maass et al., 2005). Although the litera-ture draws a clear distinction between LIB and linguis-tic expectancy effect, in this review we will use the labelof LIB throughout to avoid further complexityin exposition.

The LIB emerges in an instance of speakers languageuse. According to Wenneker and Wigboldus (in press),the LIB may occur either during encoding or during utter-ance production. Regarding the former, when people per-ceive outgroup members stereotype-consistent (typicallynegative) behaviors, they may construct more abstractlinguistic representations (spontaneous trait inferences;Uleman, Newman, & Moskowitz, 1996; for demon-strations of this effect for stereotypes, see Stewart,Weeks, & Lupfer, 2003; Wigboldus, Dijksterhuis, &van Knippenberg, 2003). When given a chance, thesegenerated abstract linguistic representations may be


used to describe groups, thus producing an LIB. Consistentwith this, Wenneker, Wigboldus, and Spears (2005)showed that an LIB occurred when the group member-ship of a target was known at the time of encoding; itdid not occur when group membership was not known.It is also possible for the LIB to occur during the courseof utterance production, in the absence of linguistic rep-resentations about ingroups and outgroups. However,in this case the language users communicative intent isnecessary for an LIB to emerge. Semin, de Montes, andValencia (2003) have demonstrated that the LIB occursonly when a communicator has a clear communicativepurpose; it fails to emerge when a communication goalis lacking.

A somewhat related issue is whether the LIB occursspontaneously or strategically. People who can controltheir responses to explicit measures of prejudice do notcontrol manifestations of the LIB, and measures of the LIBcorrelate with implicit but not explicit measures of preju-dice (von Hippel, Sekaquaptewa, & Vargas, 1997).Wigboldus et al.s results (2003) suggest the LIB derivingfrom the encoding process may be spontaneous. Nonethe-less, the LIB may be used strategically as a means of influ-encing the opinions of others. For example, prosecutionlawyers tend to describe a defendants actions with abstractlanguage (to imply dispositionality); defense lawyers tend touse more concrete language (as a means of implying situ-ationality) (Schmid & Fiedler, 1996, 1998; Schmid,Fiedler, Englich, Ehrenberger, & Semin, 1996). Likewise,Fiedler, Bluemke, Friese, and Hofmann (2003) showedthe effect of audience identity on LIB, and Douglas andSutton (2006) have demonstrated the effect of languageabstractness on perception of the speakers relationshipwith the recipient. Perhaps the most dramatic demonstra-tion of strategic language use in the LIB is by Douglas andSutton (2003), who showed that the linguistic intergroupbias can be overridden when participants are asked toprovide descriptions that are designed to accomplish aparticular goal (e.g., derogate a target).

However, the LIB is generated in an instance of lan-guage use; once it occurs, its effects can be observed inboth recipients and speakers, illustrating the formationof collective representations among them. First, inreceiving more abstract descriptions of the negativebehaviors of an outgroup, the recipient of such commu-nications is likely to form more stereotypical views ofthe outgroup; the abstract language serves to induce dis-positional attributions regarding the outgroup (e.g.,Wigboldus et al., 2000). Similar effects have beendemonstrated to occur for the speaker, suggesting thatthe LIB can help to maintain ones beliefs as well(Karpinsky & von Hippel, 1996). Although to the bestof our knowledge no direct demonstration exists, a par-ticular instance of LIB in language use may produce or

strengthen the speakers and recipients collective repre-sentations about the intergroup stereotypes and may inthe long run result in the formation and maintenance ofthe intergroup stereotypes culturally shared in the com-munity at large (Maass, 1999).

LANGUAGE AS ABSTRACTSYMBOLIC SYSTEM

As we have seen, the use of language in a particularinteraction episode has unintended social cognitive con-sequences for the speaker, the listener, and the collec-tive. In these instances it is the effects of using alanguageany languagethat is of critical importance.Language use is patterned, however, and those patternsmay become linguistic practices (Y. Kashima et al.,2006). Examples of linguistic practices include droppinga pronoun (E. Kashima & Kashima, 1998; Y. Kashima& Kashima, 2003), using adjectives as opposed to verbs(Y. Kashima et al., 2006; Maass, Karasawa, Politi, &Suga, 2006), using contextual qualifiers (e.g., addingwhen I am with my friends in saying someone is socia-ble; Y. Kashima et al., 2006), and so on. If a certain lin-guistic practice is adopted repeatedly by a majority of thepeople in a linguistic community across a wide range ofsocial contexts, this linguistic practice could have long-term and wide-spread social cognitive consequences. Thatis, a particular practice may have the unintended conse-quence of influencing the nature of the linguistic represen-tations of its users. Furthermore, if this linguistic practiceis used widely in a linguistic community for a longperiod of time, eventually it may be considered to bepart of the language itself (a type rather than a token).At that point, the language system itself may be consid-ered to have an influence on users social cognitiveprocesses. As a collection of language users use theirlanguage system as a semiotic tool to construct andexchange their meaning in everyday activities, theyengage in their preferred linguistic practice as a matter ofhabit; in so doing, they may end up affecting their ownsocial cognitive processes, thus setting themselves apartfrom others who use a different linguistic practice andcontributing to the formation and maintenance of culturaldifferences between the collectives.

In short, just as the use of language can have unin-tended consequences, so too can various features of alanguage have unintended consequences for social cog-nition and culture. The most well-known approach tothis type of languagethought relationship is what isgenerally known as the Whorf-Sapir hypothesis (ormore commonly, the Whorfian hypothesis), a cluster ofideas based on the writings of Benjamin Whorf (1956)and Edward Sapir (1921).2 The essence of the Whorfian


hypothesis is that the structure of the language one speaksserves to condition particular ways of thinking; in effect,language conditions thought. This hypothesis has been alightening rod for debates regarding the nature of lan-guage, the nature of thought, and the relationship betweenthe two. Although this hypothesis was largely discreditedby the 1970s (at least among psychologists), there has beena clear resurgence of interest in these ideas (for reviews seeGentner & Goldin-Meadow, 2003; Gumperz & Levinson,1996; Hunt & Agnoli, 1991). There is, for example, recentresearch demonstrating the effects of language on percep-tions of color (Ozgen, 2004; Roberson, Davidoff, Davies,& Shapiro, 2004; Roberson, Davies, & Davidoff, 2000),time (Boroditsky, 2001), space (Levinson, 1997; Levinson,Kita, Haun, & Rasch, 2002; but see Li & Gleitman, 2002),and numerical cognition (Gordon, 2004).

There have also been demonstrations of the effects oflanguage on social-cognitive processes. Perhaps the bestknown example in this regard is Blooms (1981) researchon counterfactual reasoning. He argued that the Chineselanguage does not contain constructions (e.g., the sub-junctive) that facilitate counterfactual thinking and thatbecause of this English speakers more accurately andmore easily engage in counterfactual thinking than dospeakers of Chinese. Although his research provided sup-port for this reasoning, his conclusions have proven con-troversial on several grounds (e.g., Au, 1983, 1984;Cheng, 1985; Chen, Chiu, Roese, Tam, & Lau, 2006;Liu, 1985), a controversy that illustrates some of theproblems inherent in testing hypotheses regarding theeffects of a specific language on thought.

In our view, there are two ways in which thelanguagethought relationship can be demonstrated. Toillustrate, suppose that a cognitive task such as memoryis used to examine the effects of language on cognition.This may involve the presentation of original informa-tion as stimuli (e.g., perceptual or linguistic) and thesubsequent response to those stimuli (e.g., recall orrecognition). To demonstrate a Whorfian effect onewould need to demonstrate (a) that people use a particu-lar linguistic practice to recode the stimuli into linguisticcodes and (b) that the linguistically coded representationsare used to produce the subsequent response to thecognitive task instead of the original information.Regarding cross-language comparisons (the typical pro-cedure for detecting Whorfian effects), it is possible thata certain linguistic feature is present in Language A butabsent in Language B. In this case, neither Condition Anor B can be satisfied for the users of Language B.However, this does not guarantee the detection of aWhorfian effect; Conditions A and B need to be satis-fied for users of Language A. That is, for the detectionof a Whorfian effect, the linguistic practice to use theparticular feature of the language should be prevalent

among users of Language A and the task should requirethe use of conceptual meaning rather than sensoryperception.

A second possibility is that a certain linguistic featureis present in both languages but more prevalent in onegroup than in the other (i.e., more frequently used bymore people in a greater number of social contexts). Ifmore prevalent in one linguistic community than theother, this practice is more likely to be used to recode theoriginal stimuli into linguistic representations. However,this is not sufficient to observe a Whorfian effect. The lin-guistic representations, rather than the original informa-tion, must be used to perform the cognitive task. This islikely when the original information is lost, less accessi-ble, or less relied on for the later use. Consequently, aWhorfian effect may be more prominent in domains thatrequire cognitive processes involving conceptual meaning(such as social cognition) than in those that are more basedon sensory perception. To the extent that the responsesrequire the use of representations that are more based onmeaning and less based on sensory perceptions, people aremore likely to use the linguistic representations than theoriginal information.

In our view, then, the effects of language on thoughtshould be stronger for domains that are less dependent onsensory experience and more reliant on conceptual meaning(for a summary of effects in the perceptual domain, seeGentner & Goldin-Meadow, 2003; Gumperz & Levinson,1996). Social cognitive domains are good examples, and wereview research in the realm of person perception that isconsistent with these implications.

Person Perception

Although some of our perceptions of people have a sen-sory basis (e.g., Hassin & Trope, 2000; Montepare &Zebrowitz, 1998; Willis & Todorov, 2006; Zebrowitz,1999), once we begin to contemplate our reaction, to our-selves or to others using language, then language can enterand play an important role in this process. An early andinfluential stage in the process of person perception is theassignment of people to categories, including dispositionalcategories. Obviously this cannot occur without the sym-bolic means to represent those dispositions. Language is asemiotic tool that facilitates this cognitive operation.However, languages differ in the lexical items that areavailable for categorizing others, and this creates the pos-sibility that speakers of different languages may differ intheir impressions of others.

A classic study conducted by Hoffman, Lau, andJohnson (1986) demonstrated this quite clearly. Theypresented Chinese-English bilinguals with personalitydescriptors. Two of the descriptions were consistentwith a one-word label in English (but not in Chinese;


e.g., artistic type), and two were consistent with a one-word label in Chinese (but not in English; e.g., shi gu, aperson who is worldly, experienced, reserved, sociallyskilled, and devoted to his or her family). Participantswere randomly assigned to read the descriptions ineither Chinese or in English, and then 5 days later theirimpressions of and memory for the target were assessed.Participants demonstrated schematic processing of thedescriptions when the language they used provided alabel for the description; their impressions were morecongruent with the label and they were more likely toendorse nonpresented traits consistent with the label. Inother words, participants formed a more stereotypicalrepresentation of the description when the languagethey used contained a relevant personality constructthan when it did not. Hence, differences in the lexiconwere associated with differences in the internal repre-sentation of the target.

The differential availability of words can also influencethe perceptions of emotions in others. For example,Lindquist, Barrett, Bliss-Moreau, and Russell (in press)used an emotion word satiation procedure to investigatethe effects of temporarily disabling an emotion word onemotion perception. When participants had an emotionterm disabled (by repeating the word 30 times) they wereslower and less accurate in making perceptual judgmentsabout faces depicting emotions. It is important that thiseffect occurred when participants made comparison judg-ments that did not explicitly require the use of emotionwords. In short, the speed and accuracy with which oth-ers emotions are perceived appears to be influenced bythe temporary accessibility of the lexicon. There are, ofcourse, clear and relatively dramatic cross-linguistic differ-ences in the emotional lexicon (Goddard, 2002;Wierzbicka, 1999). For example, Polish lacks an exactequivalent for the English term disgust, Tahitian lacks anequivalent word for sadness, and so on. Based on theresults of Lindquist et al., it would be reasonable to pre-dict differences in the speed and accuracy of the recogni-tion of emotions as a function of the language that onespeaks (although see Breugelmans & Poortinga, 2006).

In addition to perceiving others, the language onespeaks may also influence how one thinks about oneself.Numerous studies have demonstrated substantial andconsistent differences between North Americans and EastAsians regarding the reported self, with the former morelikely to describe a private (rather than public) self and toreport higher self-esteem than the latter (Bond, 1996;Heine, Lehman, Markus, & Kitayama, 1999; Trafimow,Triandis, & Goto, 1991). This difference, however,appears to be partially mediated by language. For example,Ross, Xun, and Wilson (2002) randomly assigned Chinese-English bilinguals to complete measures of self andmood in either English or Chinese. Those individuals

responding in Chinese reported lower self-esteem, com-parable levels of positive and negative mood, and morecollective self-statements relative to participants whoresponded in English. Trafimow, Silverman, Fan, andLaw (1997) found that when responding to I am . . .items in English, participants produced significantlymore idiocentric cognitions than they responded inChinese (in which case they produced more groupcognitions). More recently, Marian and Kaushanskaya(2004) examined autobiographical memories recalledby bicultural Russian-English bilinguals. The memoriesrecalled in English were more self-oriented (both interms of content and in terms of using personalpronouns) and those recalled in Russian were moreother-oriented.

LANGUAGE, CULTURE,AND SOCIAL COGNITION

As we argue above, differences among languages andlinguistic practices can result in differing social cogni-tions. But what is the source of these linguistic differ-ences? Taking a diachronic perspective, language andlinguistic practices reflect the culture within which theydevelop (Lucy, 1992), with language use being themechanism through which broad cultural orientationsinfluence the development of language. Recall Figure 1where token language use influences language changeover time. More specifically, the collaborative and situ-ated nature of language use results in language usersbeing sensitive to cultural practices. Because of this, theuse of language on any particular occasion reflects thesecultural orientations, and over time consistent ways ofrepresenting culture linguistically become linguisticpractices and eventually part of the language. In thisregard, we have identified a broad cultural difference interms of linguistic contextualization. In what followswe describe this pattern and its various manifestationsand argue that this pattern both reflects cultural differ-ences and helps to maintain them.

Linguistic Contextualization VersusDecontextualization

For illustration, it is useful to start with E. Kashimaand Kashimas (1998, Y. Kashima & Kashima, 2003)research on pronoun use and individualism and collec-tivism (Hofstede, 1980). Speakers of some languagessuch as English usually reference themselves and otherswhen reporting on their actions (e.g., I went to theshow last night); speakers of other languages often donot do this (e.g., Went to the show last night). In fact,Japanese speakers, as an example of the latter, drop the


subject of a sentence in a majority of occasions(E. Kashima & Kashima, 1998). The linguistic practice ofpronoun drop (or pronoun nondrop) may be regarded aspart of the grammar of such languages; namely, Englishmay be regarded as a pronoun nondrop language,whereas Japanese is a pronoun drop language (seeShibatani, 1990). In pronoun nondrop languages, theexplicit mentioning of a pronoun highlights the personthat the pronoun refers to so that the target is decontextu-alized, whereas dropping a pronoun contextualizes thetarget in the context of speech. Requiring a pronoun (andhence making the target person salient) is a linguistic prac-tice that is consistent with individualisms emphasis on theperson. Not requiring a pronoun (and hence not focusingon the individual) is a linguistic practice consistent withcollectivism. Consistent with this reasoning, E. Kashimaand Kashima (1998) reported that countries in whichpeople speak a pronoun nondrop language score higheron Hofstedes (1980) individualism and other correlatedcultural variables than countries in which people do speaka pronoun drop language.

Their subsequent research (Y. Kashima & Kashima,2003), using multiple regression analysis, found thatpronoun variability, gross national product (GNP), andclimate were all significant predictors of a countryslevel of individualism and that pronoun variability hada significant effect on individualism even when GNPand climate were controlled. Furthermore, they found asignificant GNP Pronoun Drop interaction, suggest-ing that the effect of economic affluence on individual-ism may be moderated by pronoun use. Hence, as levelof affluence rises within a culture, the extent to whichthis rising affluence may result in increased individual-ism may depend, in part, on whether members of thatculture use a pronoun nondrop language, a languagethat forces them to identify explicitly the individual per-former of action.

The linguistic practice of pronoun drop is a specificinstance of a broader cultural pattern and tends to cooc-cur with other contextualizing linguistic practices. Inparticular, East Asian (pronoun drop) and someEuropean (pronoun nondrop) languages appear to havedifferent linguistic practices in describing social objects.Y. Kashima et al. (2006) reported that Korean speakerstend to use verbs but English speakers tend to use adjec-tives to describe a variety of social objects including theself and other, ones and a friends relationships, andones family and a friends family. Likewise, Maass et al.(2006) showed that Italians were more likely to useadjectives when describing other individuals and groupsand Japanese speakers were more likely to use verbs.Provided that adjectives are more abstract and devoid ofcontexts in which the target persons behavior wasobserved, whereas verbs retain contextual information

more, adjective use is a decontextualizing linguisticpractice. In addition, Y. Kashima et al. reported thatanother contextualizing linguistic practice was alsomore frequently observed in Korean descriptions ofsocial objects than English equivalents; Koreans weremore likely to provide contextual qualifiers to a descrip-tion of a social object (e.g., sociable when I am withfriends) than were Australian English speakers.

These differences in contextualizing linguistic practicesshow up in other ways as well. Holtgraves and Yang(1992) demonstrated that the politeness of Koreans ismore responsive to the social context (especially the inter-actants relative status) than is the politeness of NorthAmericans. That is, Koreans were more likely to vary theirlevels of politeness as a function of the social context thanwere North Americans. More generally, there is a ten-dency for Koreans to speak more indirectly (to both pro-duce indirect utterances and look for indirect meanings)than North Americans (Holtgraves, 1997). Speaking indi-rectly requires relatively greater attention to the contextthan does speaking directly (which is more context free).This difference is consistent with Halls (1976, 1983)description of the difference between high-context cul-tures (generally collectivist cultures) and low-context cul-tures (generally individualist cultures).

Taken together, in East Asian languages, contextual-izing linguistic practices are more prevalent and decon-textualizing ones less prevalent, than in Europeanlanguages. What are the social cognitive correlates ofsuch linguistic practices? First of all, decontextualizinglinguistic practices such as the preferential uses of adjec-tives as opposed to verbs may contribute to memoryprocesses. Maass et al. (2006) showed that there arememory effects paralleling the practice of lexical choice(adjective vs. verb) in person and group perceptions. Just asItalians tended to use adjectives and Japanese tend to useverbs, Italians were more likely to transform behavioralinformation (verbs) into traits (adjectives) and Japaneseparticipants tended to falsely recall verbs when they hadseen traits. In this case cultural differences in cognitivestyles and linguistic preferences showed clear paralleltrends. Second, different linguistic practices may encouragedifferent attentional processes. In particular, Stapel andSemin (in press) showed that priming of a decontextualiz-ing linguistic practice can result in a greater ability to dis-regard the context and to pay closer attention to the focalobject. In their Experiment 3, they primed a decontextual-izing or contextualizing linguistic practice (adjective vs.action verb use) by having their Dutch-speaking partici-pants complete scrambled sentence tasks, so that whenthey reorganized the words to construct meaningful sen-tences, they would end up using either adjectives or verbs.They were then presented with Kitayama, Duffy,Kawamura, and Larsens (2003) frame-and-line test:


The task was to copy a line framed within a square intoa larger or smaller square so that the drawn line had thesame length as the original. After being primed with thedecontextualizing adjective use, the participants coulddraw the line more accurately than when primed withthe contextualizing verb use, thus showing the adjectiveusers greater ability to disregard the context (differ-ently sized squares).

It is possible that decontextualizing linguistic practicesmay jointly contribute to the formation and maintenanceof the correspondence biases in attribution (Gilbert &Malone, 1995), a tendency to infer from the behavior ofothers a corresponding internal disposition such as a per-sonality trait. This bias appears to be more prevalent inEnglish speaking cultures than in Asian cultures (Miller,1984, 1987; Morris & Peng, 1994; for reviews, see Choi,Nisbett, & Norenzayan, 1999; Y. Kashima, 2001). Usingopen-ended descriptions of an individual person, researchhas shown that Western person descriptions are more trait-based than are Asian person descriptions (e.g., Bochner,1984; Cousins, 1989; Dhawan, Roseman, Naidu, Thapa,& Rettek, 1995; Rhee, Uleman, Lee, & Roman, 1995;Trafimow et al., 1991). Some research examining attitudeattributions (e.g., Jones & Harris, 1967) has shown thatEnglish speakers are more likely to infer essay writers atti-tudes based on their essays than are East Asians (e.g., Choi& Nisbett, 1998; Y. Kashima, Siegal, Tanaka, & Kashima,1992; Lee, Hallahan, & Herzog, 1996). Indeed, Kashimaet al. (2006) showed that peoples tendency to use adjec-tives rather than verbs could account for English speakersgreater tendency to use personality dispositional terms todescribe themselves and others relative to Korean speakers,a well-known and well-established cultural difference (e.g.,Rhee et al., 1995).

It is also possible that East Asians contextualizinglinguistic practices may correlate with their tendency toconstrue the individual as embedded in a social context.Note that the linguistic practices of using verbs (ratherthan adjectives) and contextual qualifications tend toexpand the scope of causal reasoning. Consistent withthis, Choi, Dalal, Kim-Prieto, and Park (2003) andNorenzayan, Choi, and Nisbett (2002) have shown thatEast Asians regarded both actors dispositions and theircontexts as the relevant causes of the actors behaviors,although North Americans regarded actors disposi-tions as the causes of their behaviors. It should be notedthat this is different from the oft-remarked cultural dif-ference of Western dispositionism versus Eastern situa-tionism. It is that East Asians regard a broader set ofcauses (i.e., both disposition and situation) as responsi-ble for action than do Westerners. This tendency tohave a broader scope for causal reasoning extends tothe consequences of events. Maddux and Yuki (2006)found that Japanese tend to think an event is likely to

have longer and more distal effects than do their NorthAmerican counterparts.

In this connection, Miyamoto and Kitayama (2002)demonstrated that cultural differences (American vs.Japanese) in the correspondence bias are partially due todifferences in the generation of situational inferences,an activity that may reflect contextualizing linguisticpractices. In their studies, participants indicated theirjudgments of an essay writers true attitude towards anessay topic. When the essay was minimally diagnosticregarding the writers true attitude, Americans exhib-ited a substantial correspondence bias (i.e., indicatedthat the essay revealed the writers true attitude) but theJapanese participants did not. In a second study, usinga thought listing procedure, Japanese participants gen-erated significantly more situation inferences than didAmerican participants when making their judgmentsonline (but not when they were memory based). It isimportant that mediational analyses indicated that thenumber of situational inferences generated mediatedthe cultural difference in the correspondent bias. Hence,the failure of Japanese participants to demonstrate the cor-respondence bias was largely due to their tendency to gen-erate situational (rather than dispositional) explanationsfor the writers behavior.

In this section we have described some of the social-cognitive correlates of contextualizing versus decontextu-alizing linguistic practices. Again, we view the relationshipbetween language and social cognition as being reciprocal.That is, linguistic practices prevalent in a culture (e.g.,contextualizing vs. decontextualizing) may enhance corre-sponding social cognitive processes as we described above(see also Stapel & Semin, in press). These social cognitiveprocesses can further facilitate the use of those linguisticpractices. In other words, once established, a particularlinguistic practice and the corresponding social cognitiveprocess form a positive feedback loop, which furtherentrenches the connection between language use andsocial cognition. Over time, this positive feedback loop islikely to influence the development of language and helpmaintain the cultural patterns within which it develops(cf. Ji, Zhang, & Nisbett, 2004).

CONCLUSIONS AND FUTURE DIRECTIONS

Language and social cognition are intimately con-nected because language is often the medium for develop-ing, representing, retrieving, and communicating socialimpressions, judgments, beliefs, and so on. Although thisfact is obvious, our understanding of it is rather prelimi-nary. To be sure, a voluminous literature exists that exam-ines the role of language in psychological processes.However, not all of it is useful for social psychologys


interest in language. Specifically for social cognition, webelieve that language is best viewed as a semiotic tool, atool used for the creation and exchange of meaning. Aswith any tool, its use can have unintended conse-quences, and it is those consequences that have been thefocus of this article (see Klein, Spears, & Reicher, 2007,for a treatment on a more strategic side).

A serious consideration of language can allow for theplacement of many social cognitive processes in theactivities within which they usually occur. Perceptionsof others, judgments, attributions, and so on are oftengenerated while talking with others, often times inresponse to others questions. Hence, these processesfrequently occur within an interpersonal context. Whatappear to be internal, intrapersonal processes mayinstead reflect a complex interplay of intrapersonal andinterpersonal processes, with language playing a criticalrole in both.

Clearly, then, it is important to study social cognitionin its natural contexts, or at least more natural contexts.Of course this is neither easy nor straightforward. Oneapproach is to incorporate language or conversationvariables within a standard experimental procedure andexamine their subsequent effects on recipients or speak-ers. For instance, research examining the influence ofconversational contexts on person perception (e.g.,Wyer, Budesheim & Lambert, 1990) and judgments(e.g., Schwarz et al., 1991) are good examples of thisapproach. Similarly, research on implicit causality andlinguistic abstraction can be viewed as attempts toexperimentally examine the influence of language vari-ables on (primarily) the recipients cognitions, particu-larly in terms of stereotypes.

Experimental studies of the effects of language vari-ables on recipients can be extended in many ways. Thereis, for example, some research examining how languagecan influence the persuasion process (Blankenship &Holtgraves, 2005) but much more could be done. Also,the role played by metaphors in political discourse andpersuasion (Lakoff, 1996) is an area that is ripe forempirical research (see Ottati, Rhoads, and Graesser,1999). Particularly important also would be studiesexamining the effects of sexist (e.g., Ng, 1990) andracist speech (e.g., Leets & Giles, 1997) on recipientsperceptions, attitudes, and other social psychologicalreactions (e.g., Castelli, Vanzetto, Sherman, & Luciano,2001), a topic that has not received the attention that itdeserves. For example, do indirect insults result in morenegative effects (because of required inferential process-ing) than direct insults? What is the role of the speakerscredibility in these effects? Which features of an existingrepresentation are activated with racist speech?

From the speakers perspective, the effects of usinglanguage on the speakers cognitions have been studied

using various versions of the communication game par-adigm (Higgins & Rholes, 1978; Echterhoff, Higgins,Kopietz, & Groll, in press). Typically this involves par-ticipants generating messages for others who have aknown attitude toward the target. Participants clearlydesign their messages with the target in mind, and as wedescribed above, doing so appears to have importantcognitive (memory and judgment) consequences. Thisbasic procedure has been modified in various ways toexamine, for example, the relative effects of differentcommunication goals (e.g., Echterhoff et al., in press),the interplay of communication goals and activatedconstructs (e.g., Sedikides, 1990), whether people cor-rect their attitudes following the communication oftheir impressions (Todorov, 2002), and so on. This pro-cedure can and should continue to be used to examinethe effects of various aspects of language use on cogni-tion. There are also related techniques that can be usedfor examining the influence of anticipatory (rather thanactual) language use on cognition (Zajonc, 1960).

Whats missing from these approaches, of course, isthe collaborative nature of language use. For example,in the communication game procedure, participantsconstruct a message but the give and take of conversa-tional exchanges is missing; it is a one-way communica-tion situation. And this may be critical. Language use isan inherently collaborative activity (Clark, 1996) andhence many social-cognitive products are essentiallyjoint constructions. That is, the mutual influence thatoccurs when people converse may influence their internalrepresentations regarding what they are talking about.Important advances in this regard have been made byresearchers examining the effects of collaborative lan-guage use (especially story telling) on memory and selfrepresentations (McLean et al., 2007; Pasupathi, 2001;Pasupathi, Stallworth, & Murdoch, 1998). There arenumerous additional research possibilities here. Forexample, there are many variables (e.g., status, distance,formality, group size, etc.) that influence how people talkwith one another. Do these variables then have a corre-sponding effect on interactants internal representations?Because people tell stories so frequently, do they acquireand process new information with a story-telling goal inmind? Also, do people who frequently tell self-stories havemore stable identities (via repetition)? Or does telling self-stories to different audiences (and hence possibly alteringthose stories as a function of the context) result in a morefragmented self? Is the self as contextualized and discur-sive as social constructionists argue?

The colla

Language, Meaning, And Social Cognition

Documents

Transcript of Language, Meaning, And Social Cognition