Discourse-Cognition-Society Current State and Prospects of the Socio- Cognitive Approach to Discourse

This article presents the current state of the art of the socio-cognitive approach in Discourse Studies. The theory relates structures of discourse with social interaction and social structure through a socio-cognitive interface, with the argument that discourse structures and social structures are of a different nature and cannot be related directly. The cognitive theory consists on the one hand, of a personal dimension of personal mental models (i) of the events talked or written about, and (ii) of the communicative situation itself (context models), and on the other hand of a social dimension consisting of socially shared beliefs, knowledge, attitudes and ideologies – which control the personal mental models, and indirectly all discourse. In this way, we are able to explicitly describe how discourses are involved in the reproduction of social problems such as racism or sexism. Moreover, cognitive analysis also describes and explains many aspects of discourse itself, such as all discourse structures based on knowledge (such as local and global semantic coherence, evidentials, etc.), and ideology (such as the polarization of various discourse structures related to Us (ingroup) and Them (outgroup).


Introduction
T he critical study of discourse should be based on a multidisciplinary theory explicitly relating discourse structures with societal structures and thus describe and explain how structures of power and power abuse are discursively enacted and reproduced. The main thesis of a socio-cognitive contribution to this theory is that these relations between discourse and society are cognitively mediated. In this chapter I sketch the current state of this socio-cognitive approach as well as its future prospects, based on our research of the past 30 years (see also Van Dijk, 2008a, b).
Most earlier and contemporary theories in Critical Discourse Studies (CDS), as well as in neighbouring disciplines, such as sociolinguistics and linguistic anthropology, assume a direct link between discourse and society (or culture). It is generally assumed in these disciplines that social variables such as those of social class, power, gender, ethnicity or age directly cause or control language variation and structures of text and talk. The problem is that the nature of these causal or similar direct relationships is not made explicit but taken for granted or reduced to unexplained correlations.
Another fundamental problem is that societal structures and discourse structures are of a very different kind and hence cannot enter in a causal relationship in the fi rst place. A socio-cognitive theory assumes that social structures need to be interpreted and represented cognitively and that such mental representations affect the cognitive processes involved in the production and interpretation of discourse. The same principle holds true for the reverse relationship, namely how discourse is able to affect social structure -namely through the mental representations of language users as social actors.
Similar theoretical limitations characterize contemporary 'interactionist' approaches to talk, for instance in Conversation Analysis (CA) and Discursive Psychology (DP). These approaches directly link the structures of talk with structures of interaction at the micro-order of society. They thereby ignore that also this relationship is cognitively mediated: trivially, language users not only act but also think when they speak.
Denying or ignoring the existence or the relevance of this cognitive dimension of interaction is in many ways a contemporary version of similar arguments defended by behaviourism decades ago, such as the alleged lack of 'observability' or 'sociality' of cognitive representations. Many aspects of conversation, such as the role of implicit or implicated meanings, thus remain without explicit description or explanation. Indeed, even more generally, the very defi nition of the fundamental notion of 'action' without a cognitive basis is thus reduced to a behaviourist concept of 'observable' conduct. In fact, the same is true for the meaning of text or talk, which also hardly can be theoretically accounted for in terms of observability in the common empiricist sense.
From the point of view of contemporary cognitive science these statements may seem trivial although the detailed theoretical and analytical implications of these assumptions are only partly understood. Even the cognitive psychology of discourse until today has no explicit theory of how social and communicative 'environments' affect text and talk. Hence, one of the tasks of this chapter is to sketch the main tenets of just such a theory as the core of the cognitive interface of the relations between discourse and society.

Cognition
The cognitive interface of the relations between discourse and society is as complex as the very structures of text and talk, on the one hand, and those of society, on the other hand, and we are here able only to summarize some of its most relevant notions (see also Introduction, this volume; Hart, this volume). Hence, we ignore many details of the properties of Working Memory and various aspects of Episodic and Long Term Memory, cognitive processing as well as their neuropsychological foundations (of many contemporary studies in these areas, see, e.g., Baddeley, 2007;Tulving and Craik, 2000).
Instead, we focus on those aspects of the structure and role of personal and social cognition that directly account for the most fundamental properties of the production and comprehension of discourse (for details, see Graesser, Gernsbacher and Goldman, 2003). 1

Personal vs. social cognition
A fi rst crucial distinction of the cognitive framework underlying language use and discourse is that between personal and social cognition . This distinction is vaguely refl ected in the division of labour between cognitive and social psychology, although we obviously deal with cognition in both cases.
Personal cognition accounts for the ways individual language users, as members of linguistic, epistemic and social communities, subjectively produce and understand text and talk. Although such an account is framed in terms of the mental and neurological structures and processes of individual language users, it must be based on socially shared representations of individual social actors as members of various social collectivities. At the same time these mental representations and processes are activated, applied and adapted to the properties of ongoing and situated social interaction and communication, through which they are acquired, changed and socially reproduced in the fi rst place. In other words, the personal and the social in discourse processing are inextricably intertwined.
In this chapter we limit our account of personal cognition, and hence the subjective and unique properties of individual text and talk, to a brief summary of the role of different kinds of mental model. On the other hand, the social dimension of the cognitive interface between discourse and society will be described in terms of the structures and the role of knowledge on the one hand, and shared social attitudes and ideologies, on the other hand.

Situation models
Many studies in the last three decades of the cognitive psychology of discourse have shown the fundamental role of mental models for the production and comprehension of discourse and more generally for interaction and the perception of the environment (Garnham, 1987;Gentner and Stevens, 1983;Johnson-Laird, 1983;Oakhill and Garnham, 1987;Van Dijk and Kintsch, 1983).
Thus, it has generally been assumed that the understanding of discourse involves the ongoing activation, updating or construction of situation models that represent the events or the situation the discourse is about (Van Dijk and Kintsch, 1983). Since such models are the cognitive correlate of what was traditionally called the 'referential' aspect of language use, such models may also be called semantic . They account for what in philosophical terms is called aboutness, that is, the 'intentional' or representational aspect of language use.
Situation models should not be confused with the (intensional) meaning of discourse, which is a specifi c and different level and aspect of discourse processing. Indeed, situation models are independent of language use: our mere experience and observation of, and our participation in, events or situations take place in terms of situation models, whether or not we talk about them. In other words, our ongoing experience and understanding of the events and situations of our environment take place in terms of mental models that segment, interpret and defi ne reality as we 'live' it (Shipley and Zacks, 2008). Although the structures of language use (e.g., sentences and stories) are infl uenced by the more primitive structures of these mental models, the mental models of our everyday experiences are independent of text or talk. Indeed, in order to survive, primates needed such models to interact with their environment before they could speak in the fi rst place (Plotkin, 2007).
Mental models are assumed to be represented in Episodic Memory, that is, the part of Long Term Memory where we represent our autobiographical experiences or personal 'memories' (Baddeley, Conway and Aggleton, 2002;Tulving, 1983Tulving, , 2002. Although as yet no explicit theory of the structures of mental models has been formulated, it is plausible that they consist of hierarchical structures formed by a limited number of fundamental categories that defi ne the basic structure of our experience: a spatio-temporal setting, participants with different identities, roles and relations, aims, and an action or event. Interestingly, these structures also appear in the semantic case structure of the sentences of natural languages (Fillmore, 1968) as well as in stories about such events and situations (De Fina and Georgakopoulou, 2012). We here encounter fi rst evidence for the projection of mental structures on structures of language use, text and talk.
Mental models are multimodal . They represent the complex, embodied experience of events and situations, including visual, auditory, sensorimotor and emotional aspects of an experience (Barsalou, 2008;Zwaan, 2004). As such they are also uniquely personal. Indeed, they not only represent our knowledge of an event, but may also feature our evaluative personal opinion or emotions about an event -which again may be expressed (or not) in many ways in the sentences or stories about such an experience.
In sum, the understanding of text or talk not only involves the construction of its meaning (or intension ) in terms of some semantic representation, but also the construction of its referents (or extension ) in terms of mental models stored in Episodic Memory. Conversely, talking or writing about specifi c events, as is the case for storytelling or news reports, is based on the personal, subjective, situation models language users construe of such events. Obviously, such situation models may also be (partly) expressed and communicated by other semiotic systems, such as drawings, paintings, dance, gestures or music.
Since we deal with mental models as the interface between discourse and the social or natural environment, we shall see below how discourse structures resemble but also differ from the structures of mental models. One crucial difference is that models are much more complete and only partially expressed in text and talk because of the well-known epistemic and pragmatic fact that recipients only need 'half a word' to reconstruct an intended mental model with help of the inferences based on situationally and socioculturally shared generic knowledge -to which we turn below. This also explains the obvious consequence that recipient models may be different from intended speaker models. Recipients construe their understanding of discourse, that is, their mental model, not only with the expressed meaning of the discourse as well as socially shared knowledge and ideologies. They also activate strictly personal old models based on earlier discourse or experiences. In other words, this cognitive approach to discourse in terms of mental models also explains the classical distinction between speaker meaning, discourse meaning and recipient meaning.

Context models
So far strangely ignored in the cognitive psychology of discourse processing is the obvious fact that language users not only construe 'semantic' mental models of the events or situations they talk about , but also 'pragmatic' mental models of the very ongoing communicative experience or situation in which they are currently engaged. These dynamically changing and subjective mental models of the ongoing communicative situation account for what traditionally was called the context of language use and discourse, for instance in sociolinguistics (cf. Fetzer, this volume). They also provide the necessary cognitive interface for contemporary interactionist approaches that lack an explicit theory of context (for a multidisciplinary theory of context, see Van Dijk, 2008a; see also Givón, 2005; for a more interactionist approach to context, see Duranti and Goodwin, 1992).
It is at this point where we arrive at the kernel of the cognitive interface between discourse and society. Context models represent the aspects of the communicative environment, and hence the social parameters of language use, as they are defi ned to be relevant by and for the participants (for a more abstract approach to relevance, see Sperber and Wilson, 1995).
Mental models of communicative situations -like all mental models -feature at least a spatio-temporal setting, participants in various identities, roles and relationships, an ongoing action and its goals. However, specifi c of context models is that the roles, actions and goals of the participants are communicative and not only more generally interactional.
These parameters of context models provide the basis of indexicals , such as deictic expressions referring to the time, place, participants and action of the communicative situation, as well as the appropriate conditions of speech acts (Searle, 1969). Indeed, the main function of context models is to control the ways language users are able to adapt their ongoing discourse and interaction to the current (and ongoingly changing) communicative situation.
As is the case for semantic situation models, also pragmatic context models are multimodal , specifi cally also featuring the very experience of speaking, writing, listening or reading, and also feature evaluative opinions and emotions (happiness, fear, etc.) associated with the communicative situation (see also Royce and Bowcher, 2007).
As is the case for all mental models, also context models are represented in Episodic Memory. And just like (semantic) situation models, we later may recall and tell a story about an earlier communicative event in which we participated, in the same way as we may refer to and speak about properties of the current communicative situation, namely by deictic expressions or other indexicals (Levinson, 1993). In this case the semantics and pragmatics of discourse overlap because situation models and context models overlap.
With the postulation of semantic situation models and pragmatic context models we have defi ned the theoretical core of the cognitive interface between discourse and society: thus, fi rst of all, language users are able to mentally represent social events and situations and talk about them, which is crucial for the survival of the species as well as for interaction in everyday life (Van Dijk, 2014). At the same time their talk is controlled by their subjective context models representing communicative, and hence social, events and situations, in such a way that their talk, and hence their communicative interaction, is adapted to the communicative and social environment. In other words, we thus have defi ned the cognitive basis of the fundamental semantic and pragmatic aspects of language use and discourse through an interface that links the nature, conditions and control of discourse structures to the represented events and situations of the social world, on the one hand, and more specifi cally with the social aspects of the communicative situation on the other hand.

The knowledge device
One of the crucial parameters of context models is the knowledge language users need to have about the knowledge of the recipients. Virtually all structures of language, from stress and intonation, topic-focus articulation of sentences, word order, foregrounding and backgrounding, evidentials, modalities, local and global discourse coherence and schematic organization, storytelling and argumentation, speech acts and conversational interaction are profoundly and ongoingly infl uenced by a pragmatic knowledge device (K-device) that defi nes the Common Ground (Clark, 1996) of language users in interaction and communication (for detail, see Van Dijk, 2008a, 2014. Thus speakers need not express and hence may presuppose the information or knowledge they know or believe recipients already have or may easily infer themselves from socially shared knowledge (to which we shall turn below).
Similarly, many structures of discourse should be defi ned in terms of the relationship between given, known, presupposed or backgrounded information, on the one hand, and new, renewed, unsuspected or foregrounded knowledge, on the other hand (Lambrecht, 1994). Indeed, besides the many other functions of language and discourse, the communicative function is the basis and the core of the other ones, namely the transmission and the acquisition of new knowledge and relating it to old knowledge. Speech acts such as assertions and questions are thus defi ned by this epistemic aspect of context models.
The knowledge or beliefs of speakers about those of the recipients is a well-known aspect of the classical philosophical, psychological and neurological issue of Other Minds (Givón, 2005). Besides their neurological basis in terms of mirror neurons which enable the fundamentally interactive nature of language, mutual knowledge and Common Ground in talk and text based on generally simulating knowledge and intentions of others by analogy to our own, as well as shared experience (models) of the same or previous communicative situations and the shared, sociocultural knowledge of members of the same linguistic and epistemic communities (among many studies on other minds, shared knowledge and intentions as the basis of human interaction see, e.g., Goldman, 2006;Tomasello, 2008).
The K-device ongoingly manages these different sources of knowledge so as to make sure that all aspects of ongoing text or talk are epistemically appropriate in the current communicative situation.
We see that besides the other structures of semantic and pragmatic mental models underlying discourse production and comprehension, and hence all verbal interaction, mutual and shared knowledge and its ongoing management and expression is a fundamental aspect of the cognitive interface of discourse and the social environment. Without such a cognitively (and socially) based epistemic component, many if not most aspects of text and talk cannot be accounted for.

Social cognition I: sociocultural knowledge
We have seen that and how semantic situation models and pragmatic context models are needed as part of the cognitive interface between discourse structures and the structures of the communicative and social environment in which and about which language users interact and communicate. They specifi cally defi ne the subjective, personal and ongoingly contextual nature of all language use and discourse.
However, language users are not merely individuals but also social actors who are members of linguistic, epistemic and social communities and societal groups, institutions and organizations. As members of linguistic communities, they share a natural language. As members of epistemic communities they share various kinds of sociocultural knowledge about public events as well as generic structures of the natural and social world. As members of social groups and communities they share norms and values and the attitudes and ideologies based on them.
Indeed, the K-device of context models would not be able to operate without such shared sociocultural knowledge of the members of epistemic communities, allowing the necessary inferences about the nature of the current communicative situationsuch as the general norms and conditions of speech acts, conversations or various discourse genres.
Similarly, such socially shared knowledge is crucial in the very construction of situation models about specifi c events and situations, namely as instantiations or 'applications' of more generic or abstract knowledge and its inferences. Thus, we only are able to construe a specifi c personal mental situation model of a story about a bank robbery, if we have and apply more general knowledge about banks, money, thieves and their actions.
Despite the fundamental nature of socially shared knowledge, we as yet barely understand the details of their representation in memory, besides the assumption of its representation in the 'semantic' part of Long Term Memory, as well as its conceptual and prototypical structures, its organization in scripts and other schemas as well as epistemic domains (e.g., plants, animals, human beings, social groups, etc.), and its possibly multimodal foundation in various regions of the brain (among a vast number of studies on the cognitive psychology of knowledge, see, e.g., Anderson, 1980;Barsalou, 2008;Collins and Quillian, 1972;Rosch and Lloyd, 1978;Schank and Abelson, 1978).
Whatever the details of the cognitive and neurological structures of the representation of socioculturally shared knowledge in our minds/memory and brains, relevant for our discussion is fi rst of all their role in the construction of mental models of communicative and other social situations -and hence as the socially shared basis of all individual text and talk about specifi c events as well as all interaction in general.
The same mental models are involved in the generation of inferences derived from general knowledge, for instance as a basis of local and global discourse coherence. In other words, mental models on the one hand need general knowledge for their construction, and general knowledge may in turn be produced by the generalization of situation models. Indeed, most of the general knowledge we have about the world beyond our daily experiences, such as about natural catastrophes, wars, social confl icts, countries and famous people, is derived from the generalization and abstraction of mental models of specifi c instances of public (mostly media) discourse.
Secondly, such general knowledge may be the cognitive basis of discourse genres and structures that express such knowledge directly, as is typically the case for expository and pedagogical genres, as well as the (often) implicit arguments of argumentative discourse and interaction. In the same way as some structures of mental models are correlated with the structure of sentences and discourse, it may be that some structures of expository discourse exhibit underlying generic knowledge structures, as is the case for conceptual categories and relations, as well as the fundamental parameters of reality, such as those of time and causality of events, and the size, weight, form, appearance, actions or functions of landscapes, things, animals and persons.

Social cognition II: attitudes and ideologies
So far, much of the cognitive interface between discourse and society, with the exception of context models, is standard theory of the cognitive psychology of discourse processing. Yet, we have seen that mental models may feature personal opinions of language users, and these are not just based on generic sociocultural knowledge, but also on evaluative representations shared by the members of social groups: attitudes , such as our attitudes about immigration, abortion, homosexual marriage, the free market or wars (Eagly and Chaiken, 1993;Pratkanis, Breckler and Greenwald, 1989).
As forms of socially shared cognition, attitudes are probably also represented in semantic or 'social' memory, in which we store all general, generic and socially shared beliefs that are necessary as a basis to form specifi c, local and personal mental models and their opinions. As is the case for generic knowledge, at present little is known about the cognitive structures of attitudes, beyond hypotheses about their schematic organization (e.g., as an abstraction of mental models, for instance about immigration), the role of underlying norms and values defi ning what is good or bad, permitted or prohibited.
Like socioculturally shared knowledge, attitudes are essentially social (Jaspars and Fraser, 1984). They should not be confounded with personal opinions as stored in mental models, as is quite common in traditional attitude research. They are shared by members of social groups, each with their own identity, actions, norms and values, relations to other groups and resources (as the basis of their power and reproduction). Thus, whereas the K-device of context models is the specifi c interface between the epistemically controlled structures of discourse and the shared social knowledge of communities, attitudes represent the relationship between social groups and their members and the ways members as language users express opinions about social events, situations, people or groups. More generally, such attitudes are at the basis of all social practices of group members, as is the case for ethnic prejudices as a basis for specifi c forms of discrimination and exclusion in general, as well as for racist text and talk in particular.
Crucial for our theoretical framework is that the relationship between social structure, such as domination relations between groups, as exercised for instance in discriminatory social practices, is mediated by mental representations of such attitudes. It is also in this way that -conversely -discriminatory discourse and other social practices are involved in the daily reproduction of social structures of domination and resistance. Obviously, we need this component as a theoretical basis for all critical studies of discourse. Thus, racism, sexism and other forms of social inequality do not directly infl uence discourse, nor vice versa does discriminatory discourse infl uence societal structures of domination. This is only possible through the cognitive interface of socially shared attitudes and the personal mental models (in turn infl uencing personal actions and discourse) based on them.
It is also through such socially shared attitudes that group members are able to cooperate in the attainment of personal and social goals, because they allow individual social speakers and actors to infer the opinions and current goals of other group members. Thus, racism and hence white group power is not reproduced by individually bigoted people, but by the joint or separate daily cooperation of (white) group members to exclude, marginalize and problematize members of other ethnic groups in many different everyday situations.

Ideologies
Many social attitudes have a broader and more general sociocognitive basis that allows different attitudes to be formed, acquired, and applied, namely by underlying ideologies . Thus, there are many different sexist or racist attitudes about different social issues, but they may at a more abstract level be based on underlying sexist or racist ideologies. Such ideologies are the basic cognitive self-schema of a group and its interests, and defi ned by such general categories as the identities, actions, goals, norms and values, relations with other groups and the (power) resources of a group. In other words, together with socially shared group knowledge, they are the cognitive core of social groupness (Van Dijk, 1998).
Ideologies obviously not only have sociocognitive properties, but also societal ones, for instance in terms of interest groups, their leaders and institutions, and especially how the ideologies of the groups are shared, reproduced, taught and acquired by their members, for instance by schooling and indoctrination and by specifi c discourse genres, such as catechisms, party programmes, lectures, hymns, protest songs among many other genres (see Cap and Okulska, 2013).
The theoretical framework sketched so far shows that there is quite a distance between abstract underlying ideologies, on the one hand, and their expression in discourse, on the other hand. Thus, ideologies are more specifi cally specifi ed in socially shared attitudes about particular 'issues' or group concerns (such as racist ideologies may be applied and specifi ed in the formation of attitudes about immigration or employment quota).
These attitudes are in turn instantiated in the personal opinions of group members in their situation models, and these biased situation models may fi nally be (partly) expressed in discourse and other social practices as controlled by context models.
This may also mean that underlying ideologically based opinions of group members in a specifi c context are not expressed at all, e.g., because they are found inappropriate or otherwise counterproductive (against the goals) in the current communicative situation, as represented in the context model of the speaker. More generally the difference between discourse structures and model structures is crucial, and hence cannot be mutually reduced to each other. Thus, we may explain many aspects of discourse structures, such as local and global coherence, implicitness, indexicals, speech act conditions, as well as ideological structures, in terms of underlying situation and context models, but do not reduce such structures to mental models.
We now have reviewed the major components of the cognitive interface between discourse structures and social structures, by relating personal, subjective semantic and pragmatic models as well as their underlying forms of sociocultural knowledge, attitudes and ideologies, with represented social events, communicative situations, and members of epistemic communities and social interest groups, institutions and organizations. This complex framework also explains how communities, social groups and power relations are reproduced by discourse and other social practices based in turn on underlying personal mental models as well as socially shared forms of cognition.

Society
The minds of language users are concretely embodied in real persons who besides being unique individuals are members of social groups, institutions and organizations, and who interact and communicate with other members through text and talk. Thus, in the same way as we need a cognitive interface in order to describe and explain many properties of discourse, we also need a societal basis for both cognition and discursive interaction. Above we have summarized the sociocognitive dimension of this societal basis, namely in terms of socially and culturally shared mental representations (knowledge, attitudes, ideologies, norms, values) of groups and communities. Below, we shall see that part of the social order, at the micro level, is constituted by social interaction in general, and communicative interaction and discourse in particular. Crucially, though, the societal basis needs further analysis that goes beyond that of discursive social interaction. Accounting for institutional talk, for instance, presupposes a theory of institutions and their roles in society. Shared social knowledge presupposes epistemic communities, whereas ideologies presuppose specifi c social groups.

Power
Specifi cally relevant for the critical study of discourse is the analysis of power and power abuse. Of the vast number of characteristics of societal structure, which are obviously beyond the scope of a single chapter, we shall therefore briefl y summarize some of the properties of power and how they are related to discourse (for detail, see Van Dijk, 2008b).
Although power relations between individuals are also discursively created, expressed and reproduced by discourse, CDS is specifi cally interested in social power relations between groups, organizations or institutions. One of the ways to defi ne such social power is in terms of control. One group has power over another group if it is able to control (specifi c) actions of (the members of) another group, thereby limiting the freedom of the other group (see, e.g., Lukes, 2004;Stewart, 2001).
Since discourse is a form of action, power may be exercised by controlling discourse, that is, In order to be able to exercise such power, groups need a power basis, which may be material or symbolic. Relevant of symbolic power resources is preferential access to public discourse, as is the case for the symbolic elites, such as politicians, journalists and professors. Thus, each social group is not only characterized by its structures, relations to other groups, the characteristics of its members, but also the presence or absence of power resources. More specifi cally, a group may be defi ned in terms of the nature of its access to and control of public discourse. Thus, journalists have active access to the construction of news, politicians may have active access to parliamentary debates, and professors to the production of scholarly discourse, whereas most common citizens have only passive access, as recipients to such forms of discourse, or only as participants in the representation of discourse, for instance as news actors or citizens talked about in political or educational discourse.

Example: The reproduction of racism
Whereas this offers one dimension of the complex framework that relates discourse and a specifi c dimension of society, namely social power defi ned as control, obviously detailed empirical research is required to spell out exactly what social group or institution controls what discourse structures in what communicative situations. Thus, if we want to analyse how racism is reproduced in society as a form of power abuse, or domination, of one (European) group over other (non-European) groups, we obviously need much further analysis of all dimensions. However, even in summary, the Discourse-Cognition-Society triangle fi nds an excellent illustration in an analysis of racism (for detail, see Van Dijk, 1984, 1991, 1998, 2009b. Racism as a social system of domination has two major dimensions, that of socially shared representations (prejudiced attitudes), on the one hand, and specifi c social practices of illegitimate treatment (discrimination). Crucial for the reproduction of the system is the reproduction of the social representations on which it is based. This happens through forms of racist text and talk, one of the forms of discriminatory social practice. Although all dominant group members may have access to specifi c discriminatory practices and discourse in everyday interaction, most infl uential are the discriminatory public discourses of the symbolic elites in politics, the media, education and scholarship. If these elites control the topics, lexicon, arguments, pictures, metaphors of the discourse about immigrants and minorities, and if discourse structures may affect, as explained, the formation by the recipients of mental models of ethnic events, these models may also be ideologically biased, and when generalized form or confi rm the prejudices shared in the dominant group. Finally, specifi c negative attitudes may be further abstracted from and form underlying racist ideologies that sustain the sociocognitive basis of the system of racism. Racist ideologies, as is the case for many ideologies, are generally organized by a bipolar schema of Positive Self-Presentation and Negative Other-Presentation (Derogation), a schema that also infl uences the structure of specifi c racist attitudes (e.g. on immigration or quotas), and these may fi nally infl uence the concrete mental models group members form of specifi c ethnic events they participate in or read or hear about. Ideologically based polarized racist mental models, depending on context, may be expressed in racist practices such as discourse that is similarly organized between US vs. THEM, at all levels (pictures, topics, lexicon, metaphors, and so on).
We see that the analysis of a form of social domination, or power abuse, such as racism, requires analysis of the three fundamental dimensions of Discourse, Cognition and Society. Racism as a system is based on shared social representations (prejudices), which however can only be acquired by specifi c structures of public discourse infl uencing the formation of racist mental models. But socially speaking the production of such discourses is controlled by the symbolic elites. Such an analysis is not only relevant at the global level of societal reproduction of racism, but also for the analysis of specifi c, individual communicative events, at the micro-level.
It hardly needs repeating that this summary of a general theory, intended to introduce and explain the relevance of a sociocognitive approach in Critical Discourse Studies, is unable to specify the vast amount of details required for the analysis of specifi c cases. For an analysis of racist news in the press, for instance, one would need previous analysis of the social representations and prejudices of the journalists, the everyday practices of news gathering, who are interviewed or otherwise have access to the journalists, the cognitive processes and the contextual constraints (size, deadline, style, preferred contents, etc.) of newswriting, the many structures of news reports and the ways these are read, stored and recalled by the readers, among many other and more detailed structures. But also in the study of specifi c cases, we always fi nd a combination of social structures and interaction, discourse structures and underlying cognitive structures such as models, knowledge and ideologies.

Linking discourse with cognition and society
Despite the fundamental role of the analysis of cognitive and social structures for the multidisciplinary explanation of text and talk, the discourse component is of course the specifi c and central aim of any critical study of discourse. So, in this last section we focus on the relations between discourse structures and their sociocognitive embedding. A more or less complete account of these relationships would require a multi-volume encyclopaedia, so we can only summarize some fundamental principles. Indeed, especially also for our students, it is crucial to present some guidelines of how to do a sociocognitive analysis of discourse (for other approaches, see e.g., Schiffrin, Tannen and Hamilton, 2013;Titscher, Meyer and Wodak, 2000;Van Dijk, 2007; see also the other chapters in this book).
The fi rst principle of a sociocognitive analysis (henceforth CGA) is that such an analysis goes beyond the classical 'autonomous' theories and methods of discourse and conversation analysis that study the grammatical, semantic, pragmatic, rhetorical, stylistic, narrative, argumentative, interactive or other structures of text and talk. On the one hand, a CGA presupposes the results of these earlier theories and methods, but on the other hand corrects and extends them, because many structures (such as those of coherence or manipulation, among many others) can only be formulated in social and/or cognitive terms.
Secondly, as argued above and unlike the presupposition of most theories in CDS, sociolinguistics or linguistic anthropology, there are no direct links between social structures and discourse structures , because all discourse production, comprehension and uses are mediated by the mental representations of the participants. Thus, if in critical studies a link is established between discourse and social power, such an account should be seen as a shortcut, as incomplete or as tacitly presupposing mental structures of members and processes that remain unaccounted for in the analysis.
Discourse structures are essentially different from social structures, such as those of groups, communities, institutions, nation states and their properties and relations. Hence I repeat that the latter can only infl uence or be infl uenced by text or talk by a mental interface that links the mental representation of social structures with the mental structures representing discourse. This is not only the case for societal macrostructures and relations, such as those of the system of racism and its reproduction by the mass media, but also at the microlevel of everyday interaction, such a journalists interviewing news actors. Thus, it is true that instances of talk are at the same time social speech acts and forms of social (inter)action and as such the fundamental order of society -that is, phenomena where the discursive and the social coincide. Yet, the structures of talk as social interaction should not only be described in the autonomous terms of, for instance, turn taking, interruptions and sequential organization, but also in terms of the mental representations of the participants. Thus, it is not suffi ciently accounted for in Conversation Analysis that turns do not directly occasion next turns, but the mental production of next turns is occasioned by the interpretation, and hence the mental representation, of previous turns . In other words, we should account for text and talk at three levels, the level of discourse sequencing, social sequencing at the micro level, both mediated by sequences of mental representations and processes.
For instance, journalists writing a news report or interviewing a news actor are engaging in many organizational social acts, at various levels, but are able to do so only on the basis of vast amounts of social and political knowledge in general, and specifi c knowledge of news events as represented in mental models. Such knowledge at the same time infl uences the structures of the news report or interview in many ways, such as its local and global coherence, overall schematic organization, implications, presuppositions, actor and action descriptions, foregrounding and backgrounding, indexicals, topic-focus articulation, the interactional structures of interviews, and many more.
In a broader framework these combined discourse-cognition-action sequences may play a role in the reproduction of racism or the management of an economic crisis, and then again relate global discourse actions (publication) of newspapers, prejudices and ideologies of journalists, and the reproduction of social inequality in society. In other words, the Discourse-Cognition-Society triangle can be analysed at all micro and macro levels of description.
Take for instance the use of the metaphor of waves routinely used by the mass media and politicians to describe the arrival of a large group of immigrants in the country. Such a metaphor fi rst of all may be described as part of a sequence of negative Other-description in a news report or editorial, and as a form of appraisal, to be accounted for in the semantics of discourse. Secondly, a cognitive analysis describes the metaphor as a way of embodied conceptualization of large groups of foreigners in terms of large amounts of water in which we may drown, and hence as a vital threat. Such a representation is part of a socially shared negative attitude (prejudice) about immigrants, which in turn may be based on a racist ideology. Thirdly, the metaphor used in this way by politicians and newspapers in order to stimulate fear of immigration is socially a way to (re)produce racism and politically to persuade people to vote for anti-immigration parties of candidates. Of course, each of these levels and dimensions of the account of metaphor need much more analytical detail, but it should be clear that adequate critical analysis requires all levels of description. Limiting the analysis of metaphor to a mere semantic-rhetorical fi gure of speech or as a way of thinking, emphasizing large numbers of people obviously would be an inadequate under-analysis of the data.

A sample CGA-analysis: An editorial
To illustrate these general principles, let us examine in some detail how the structures of discourse are to be related to those of cognition and society. As an example, we use a newspaper editorial, a prominent daily discourse genre which however has received relatively little theoretical attention in discourse and genre studies. Socially, this genre is a routine macro-act of public opinion expression and formation of (the editors of) a newspaper, functioning to infl uence public opinion on relevant social and political events and issues. Cognitively, editorials express opinions, attitudes and ideologies, and presuppose vast amounts of social and political knowledge, and hence require cognitive analysis in order to describe their production, reception and reproduction in epistemic communities and ideological groups. Discursively, editorials are a genre of persuasive discourse. We may therefore expect a variety of rhetorical and argumentative structures, appraisals, ideological structures, descriptions of social events, actors and situations, epistemic structures, and pragmatic structures of assertions, recommendations, advice for other social actors and institutions, as well as forms of politeness when criticizing powerful actors or institutions. Crucial is that all these discursive, cognitive and social structures, strategies and processes take place, in combination, and at several levels, at the same time.
The text we use for the analysis is an editorial on immigration published on 13 December 2012 in the British tabloid newspaper The Sun , well known for its stance against immigration. 8 Immigration can be a sign of a dynamic society. The South East in particular 9 would grind to a halt without industrious foreign workers.
10 Controlled immigration of talented newcomers is welcome, and the 11 Olympics showcased the friendly and positive side of the new-look Britain.
12 But the sheer scale of the infl ux, and its pace, raise serious questions.

21
The upshot is that immigrants suffer as badly as anyone else from 22 overwhelmed public services.
23 If anything, the 2011 Census is behind the times already as new migrants 24 arrive. Nor does it include tens of thousands here illegally.
25 Coming decades will see further changes in our national makeup. But there 26 is no turning the clock back.

27
The challenge now is to ensure future immigration is in line with what the 28 country can cope with.

Context analysis
Systematic discourse study not only requires an analysis of discourse structures but also of the structures of context defi ned as the relevant parameters of the communicative situation as it is construed by participants in their context models. Thus, the journalist writing this editorial is assumed to have a context model defi ned by the parameters on the left and a selection of discourse properties on the right of Table 5.1 .
These context parameters are relevant properties of the communicative situation of editorial writing because they systematically control and defi ne the appropriateness of the discourse in that situation. That is, an editorial is appropriate if it published in a newspaper, on a particular date, and in a specifi c city and country, written by a journalist as editorialist, expresses opinions on recent events or situation, and intends to infl uence public opinion and social or political policy. Within the framework of an SCA this means that discourse structures are not only described as such, but also as being controlled by the cognitive parameters of the context model, knowledge, attitudes and ideologies of the writer, as well as assumptions about those of the readers. This discourse-cognition analysis is then fi nally placed in the institutionalorganizational framework of the newspaper, its relation to its readers and with other organizations or institutions, such as the government. Note though that such a description is not merely the result of macrosociological description by the analyst, but represented in the very context model of the writer and readers. In other words, the discourse analysis is totally integrated with a cognitive and a social analysis.

Text structures
Since a complete analysis of all the relevant discourse structures of the editorial would require many hundreds of pages, we focus on those variable structures that are specifi cally controlled by cognitive and social structures. This means that we ignore obligatory grammatical structures of English that do not vary by cognition and context, as well as those of specifi c discourse grammar, such as the interpretation of the pronouns and verbs of minimal sentences It is (3) or They can't (16), following the fi rst sentence of the same lines, and defi ning a specifi c compact style of editorial syntax.

Semantic structures
Given their central role in discourse analysis beyond sentence syntax, and for critical discourse analysis in particular, let us focus on the semantic structures of the editorial.
Semantic Macrostructure . The overall coherence of discourse is defi ned by its semantic macrostructure or discourse topic, dominating the local meanings of the discourse as partly expressed by its constituent sentences. Thus, the whole text is conceptually organized by the macropropositions such as 'Uncontrolled immigration is bad for the country'. Part of that macroproposition is conventionally expressed by the headline as an obligatory category of the conventional schematic structure (its superstructure ) of the editorial: Infl ux impact. But locally it organizes all the semantic information related to the arrival of (allegedly too) many immigrants: foreign-born population (4-5), three million rise (4), white British (6-7), immigration can be a sign of a dynamic society (8), new immigrants arrive , etc. Another main topic, controlling various local propositions, is for instance that Labour's immigration policies were reckless, as well as the topic defi ning the actual recommendation of the editorial, namely that current government policy should limit immigration.
Although semantic macrostructures are assumed to defi ne the notion of the overall discourse topic, they are not always explicitly expressed in the text. Rather, they defi ne the top levels of the mental situation model that represents overall meanings of the author and reader of the discourse. In other words, such discourse structures can only be defi ned in terms of underlying cognitive structures, such as the hierarchical nature of mental models. At the same time, these macrostructures are also those overall meanings that are best recalled by the readers.
It is at this point where also the social and political functions of the editorial and its overall topics play a role. The Sun as an infl uential tabloid with millions of readers has signifi cant infl uence on public opinion and government policy. Thus, its overall opinions on immigration are not only locally relevant for the interpretation of this particular editorial, but play a fundamental role in the national debate. Thus, mental models of specifi c events and the opinions instantiated in them, also infl uence socially shared mental representations, that is, attitudes and prejudices about immigration and immigrants. In other words, our sociocognitive analysis-related discourse structures (topics, macrostructures), with top level structures of mental models and their prominence in episodic memory, and these in turn with socially shared attitudes, and the infl uence on public opinion of citizens and voters, government policies and party politics. The explicitly expressed negative appraisal of Labour in this editorial more specifi cally relates to the powerful role of the mass media, and especially of a tabloid like the Sun , and its relations to the political parties, and hence as part of the power structure in the country.
In brief, discourse topics of editorials are sociopolitically relevant, but this relevance is cognitively mediated by the role of topics in memory, and hence in the process of discourse processing, comprehension, and hence for the formation of public opinion, and its consequences for national policies, as the main persuasive function of editorials.
Local coherence. Discourse is not only globally coherent, but also locally. This local coherence takes two forms, an intensional (meaning) and an extensional (referential) one. Intensional relations are functional and hold between expressed propositions, such as Generalization, Specifi cation, Example, Explanation, and so on.
Thus, in line (1) the brief last sentence It is , is a Repetition and Confi rmation of what it stated in the conditional clause of the previous sentence. And in the next lines, the proof mentioned in line (4) is followed by a proposition about a three million rise that is a Specifi cation of that proof. In order to be able to establish these relationships, however, author and reader need to activate conceptual knowledge of the world, relating Census with quantitative data about the population.
On the other hand, coherence is not only based on meaning relations and conceptual knowledge, but also defi ned in terms of causal, temporal, part-whole relations between the facts denoted by the sentences of the discourse, as they are represented in the mental models of authors and readers. Indeed, in that sense, a discourse is coherent if it has a mental model. Thus, the event of the changing country mentioned in line 1 is related to the cause of a three million rise in line 4. Such a postposed causal proposition at the same time functions intensionally as an Explanation. Thus, most of the propositions of this editorial are locally connected by direct or indirect temporal and causal relations in the mental models of the author, such as: Three million rise of foreign-born population Country is changing Whites in London are a minority Country is changing Without industrious foreign workers South East would halt immigration can be a sign of a dynamic society (But) Unlimited cheap foreign labour Frustrates governments attempts to make work pay better than benefi ts etc.
Note though that the temporal and causal relations defi ning mental models are often only implicit in discourse, and require inferences instantiating socioculturally shared knowledge of the world. In other words, also local discourse coherence is multiply dependent on underlying mental model structures and more general sociocultural knowledge.
But what are the social and political conditions or functions of this kind of local coherence, beyond the obvious fact that sociocultural knowledge is defi ned for epistemic communities? First of all, such local coherence also depends on group attitudes and ideologies. Thus, to describe the fact that white people in London are now a minority in the negative evaluative terms of a ' startling statistic ', presupposes an attitude and ideology according to which London or Britain should be white, or that whites should at least be a majority. Most crucial however is the statement that increasing immigration is bad for the country, e.g., because housing, hospitals and schools would not cope. Such causal reasoning, however, presupposes a mental model in which specifi cally foreigners are a problem, not the fact that there are more people in the country. Indeed, the newspaper would not object to an increase of the autochthonous population and ignores that an increase of foreign workers leads to an increase in revenues to pay for extra schooling, housing and hospitals and hence would mean an increase of personnel and hence have a positive effect on the labour market. In other words, the local coherence of this editorial is premised on a specifi c mental model of the current situation in the UK, in which foreigners are focused upon as a major cause of social breakdown. Such a model is itself an instantiation of a socially shared negative attitude about immigration, based on a racist or xenophobic ideology.
Finally, in the triangular account of sociocognitive analysis, these mental models, attitudes and ideologies expressed in an editorial of an infl uential tabloid have important social conditions and consequences, namely as a contribution to the formation or confi rmation of racist attitudes among the population, on the one hand, and politically as an opponent of Labour and as partisan of the Conservative Party and its antiimmigrant policies. Again, discourse meanings and their underlying mental representations are not innocent, but multiply connected to social and political conditions and consequences. Note though that the discourse-society relation is not direct, but mediated by mental models of journalists and readers, and shared social and political attitudes and ideologies, without which text and talk would not make sense in the communicative situation.
Disclaimers and the denial of racism . One prominent feature of discourse about immigrants are disclaimers, which express brief positive self-presentation (e.g., I am not a racist. . .) with extensive and dominant negative other-presentation (but. . .). This text is a characteristic example of such a disclaimer. Thus, even The Sun in 2012 no longer can openly align itself with the extremist racist position that Britain should be or remain white and without any foreign workers. On the contrary, in lines 8-9 it is conceded that a specifi c kind of immigrant (industrious and talented ones) may be positive for a 'dynamic' society and a new-look Britain. Thus, the main argument of the editorial is not against immigration or immigrants in general, but against 'the sheer scale of the infl ux' -threatening that whites may become a minority in their own country or city. This moreover presupposes that most immigrants are not white, and would hardly be consistent with an opposition against immigrants from Romania and Bulgaria the newspaper and other conservative institutions are opposing.
Disclaimers not only have characteristic semantic structures, but especially important contextual and cognitive implications. First of all, they express specifi c negative models of recent immigration events, as well as more general negative attitudes against immigrants. However, such opinions and attitudes may be inconsistent with liberal, modern, non-racist or non-xenophobic norms, values and hence the public image of the speaker or author. Hence the contextual relevance of the fi rst pair part of a disclaimer as a form of positive self-presentation, intended to block negative evaluations by the recipients.
The social and political conditions and consequences of disclaimers are obvious, and can be formulated in terms of the social image, prestige, and hence the symbolic capital of speakers, authors and institutions. At the same time, as part of a persuasive discourse, they may diminish negative opinions of the authors and hence a bigger chance their argument will be accepted -which enhances the social and political infl uence of the newspaper, and hence its power. At the same time, by emphasizing the negative part of the disclaimer, the newspaper enhances ethnic prejudice in society and thus contributes to the reproduction of racism.
That such a concern is quite relevant in the context model of the author becomes obvious in lines 18-19, which we repeat in full: A sensible debate on immigration was silenced by Labour's poisonous charge of 'racism' against anyone who dared raise the issue.
Indeed, in this editorial as well as in many news and opinion articles in the same newspaper as well as other conservative media, the denial of racism, and the attack on antiracists is a standard strategy of negative discourse on immigration. In this example, this is done by several other strategic moves of persuasive discourse featuring the typical bipolar structures of ideological discourse, in which positive qualities of ingroups and negative qualities of out-groups are enhanced: already saw when we discussed the example of the ubiquitous use of the WAVEmetaphor for the arrival of large groups of immigrants. We argued that this metaphor is hardly innocent, because it precisely defi ned situation models in recipients in which immigration is seen as a threat, as a danger of drowning in immigrants, and hence as propagating fear among the population. It is not surprising that also in this editorial similar metaphors are prominent, as is the case in the topic expressing headline: infl ux (repeated in line 12), as well as in this context negative action of opening the doors attributed to Labour. Similar, accusations of racism are conceptualized as poisonous (19), and hence also ideologically express negative Other-presentation and as enhancing the threat to a 'healthy' debate.
The social and political implication of such mental models and their generalization in shared attitudes and ideologies hardly need to be spelled out -as the election campaigns of many parties in Europe have shown, with similar threatening metaphors, e.g. Das Boot ist Voll (The boat is full) in Germany, made explicit in this editorial by such expressions as 'what the country can cope with' (27-8).
Other discourse structures . For brevity's sake we have focused only on some, mostly semantic, discourse structures. The point of this succinct analysis was to show how specifi c discourse structures have cognitive foundations that are crucial for their description and explanation. We also saw that the combined discourse-cognition structures are themselves related to knowledge, attitudes and ideologies shared by groups, communities and organizations, and that social and political aims, status, infl uence and power both condition as well as follow from these discursive practices through cognitive mediation.
In our example, thus, we would need to engage in a detailed analysis of the argumentation of the newspaper, partly made explicit in our analysis of local coherence and its implicit propositions. The main standpoint is that the UK cannot cope with (massive) immigration, and the arguments are especially sought in the assumed ' overwhelming ' of the public services. For these and other arguments to be cogent, they also need to make explicit many implicit assumptions of mental models and derived from sociocultural shared knowledge and attitudes. Socially and politically the same argument is relevant because it specially addresses the concerns of the citizens about the quality of the public services, thus confi rming the prejudice that foreigners are the main cause of their lacking quality.
Part of the argumentation and typical of negative discourse and media reports on immigration is the number game ( three million, 45% , etc.) of the editorial, sustaining the crucial quantity dimension of the main argument: too many immigrants (or in more formal style: the scale of the infl ux , 12). Again, these are not just meaning or rhetorical hyperboles, but confi rm the embodied, threatening nature of the mental model presented by The Sun .
Among many other semantic and rhetorical structures, similar arguments may be developed for Sun -style alliterations ( Infl ux Impact ), colloquial style addressing readers ( if you think . . .), hyperboles and worst case descriptions ( startling statistic, unlimited cheap labour ), many forms of positive self-presentation ( dynamic society, newcomers are welcome, friendly and positive side of the new-look Britain , etc.).

Conclusion
The study of discourse should be multidisciplinary. Language use, text and talk are at the same time linguistic, cognitive and sociocultural and political acts. A coherent theory of discourse should make explicit how these acts are related, both at the micro and the macro levels of analysis. Especially also for Critical Discourse Studies it is crucial that discourse structures are not directly related to social structures of groups, power and domination. Discourse can only have social and political conditions and consequences if we recognize that discourse is produced by language users as social participants who not only speak and act, but also think, know and feel. This means that in the same way as our analysis of text and talk should be systematic and detailed, also the underlying cognitive analysis should be explicit and sophisticated. Only with this missing link made explicit are we able to understand how discourse is able to function in the reproduction of power abuse and the resistance against it. A brief analysis of especially some of the semantic structures of an editorial in The Sun shows how a sociocognitive analysis relates discourse structures with social structures, via an analysis of cognitive structures, and how such an editorial may contribute to the reproduction of racism in society.