An introduction.

Teun A. van Dijk
University of Amsterdam
Universitat Pompeu Fabra, Barcelona

Version 1.0. October 25, 2000


There is no such thing as "cognitive analysis". Or is there?

We have discourse analysis, and its many branches (stylistics, rhetoric, narrative or argumentation analysis, as well as syntactic, semantic or pragmatic analysis, and of course conversation analysis), but "cognitive analysis" is not a well-known, standard way of looking at text or talk.

We have a cognitive psychology of discourse processing (production, comprehension), and we have a social psychology of discourse (the Loughborough school) called "discursive psychology", but the latter rejects any mental approach and in fact advocates a more ethnomethodological approach to discourse within social psychology.

So, if we speak about "cognitive analysis" , it is something we have to invent ourselves. And we must show why it is relevant for our understanding of discourse.

The argument behind such an attempt is that text and talk do not exist in isolation. Most obviously, discourse analysis since many years emphasizes the relevance of the study of context for our understanding of many aspects of discourse. Relevant in such contexts are the social domain (e.g., Education, Politics), the global act partially accomplished by text or talk (e.g., legislation, teaching, etc), the participants and their various communicative, social and professional roles, the relations between participants (such as that of power), the setting (time, location) and maybe some other social or interactional properties of communicative event.

Part of the context, however, are also some of the 'cognitive' properties of the participants, such as their aims, beliefs, knowledge and opinions. Without taking into account, we cannot understand why people are speaking or writing at all, or how they show adapt what they say or write to the knowledge or other beliefs of the recipients.

In other words, not only because of a 'mentalist' aim to understand the processes of actual discourse comprehension or production, but also for important contextual reasons, a study of the cognitive aspects of communication is highly relevant.

In this introductory working note, we provide an outline of how to do a 'cognitive' analysis of discourse.

2. Discourse processing

For more than 25 years, cognitive psychology has been actively engaged in the study of the processes of discourse production and comprehension. Emerging from the psycholinguistic study of sentence processing in the 1960s and 1970s, this approach emphasized that mental processes of understanding should not be limited to isolated sentences. People produce and understand whole discourses, and even the processing of words, clauses and sentences needs to be studied as integrated part of the processes involved in the production or comprehension of discourse.

A psychological or cognitive study of discourse is rather different from a more formal, grammatical or (say) stylistic, narrative or argumentative analysis. It does not deal with abstract categories and rules purported to describe 'structures' of discourse, but with the actual mental representations and processes of language users. In that respect, psychology intends to provide a more 'empirically' based understanding of discourse.

Thus, instead of analyzing 'given' structures, such a more 'strategic' approach, studies discourse processing 'on line', at several levels at the same time, as a fast but imperfect sequence of mental acts that are geared towards 'making sense' of the respective words, clauses, sentence, paragraphs etc of discourse. Thus, in a cognitive analysis, interpretation is not static, nor an abstract procedure, as in linguistic semantics, but a dynamic, ongoing process of (at first tentatively) assigning meaning and functions to units of discourse.

In order to account for such processes of production and understanding, psychology makes use of a large number of more or less technical notions describing various aspects of the 'mind':

  • Short Term Memory (STM) vs. Long Term Memory (LTM)
  • Episodic Memory vs. Semantic Memory
  • Situation or Even Models
  • Knowledge (scripts, etc.)
  • The overall process of strategic discourse production may then be summarized by the following ten steps (for details, see Graesser, et al, in Van Dijk, 1997, Vol. 1).
  • Text/talk is read/heard and interpreted on line, unit by unit (e.g., word for word)
  • On the basis of world knowledge, as well as knowledge of words, syntactic structure, overall meaning (topics), discourse structures and aspects of context (goals, etc) such units are assigned provisional meaning
  • Meanings of smaller units are combined (and where necessary corrected) into those of bigger units, such as propositions, and sequences of propositions, until the Short Term Memory buffer is (nearly) full.
  • The thus interpreted bigger chunks are stored in a Textual Representation (TR) in Episodic Memory.
  • Parallel to this understanding of the respective units of the text and the formation of a TR in Episodic Memory, language users activate an old, or construct a new (mental) model of the events or situation the text is about.
  • Understanding a text means the construction of such a model: A text is meaningful or understandable when a recipient is able to construct a model for it.
  • Apart from meaning elements from the (mental representation of the) text, also information from previous (old) models (= earlier experiences), as well as specific instances of general (socially shared) knowledge help to build up such models.
  • Models are both personal (featuring individual knowledge, beliefs, opinions of language users) as well social (applying general, socially shared knowledge), but each model is unique. The same person may construct a different model (=a different interpretation) of the same text tomorrow.
  • The whole process of understanding is coordinated by the model language users have of the communicative situation, namely their context model (or simply: context). The context model tells the language user what the aims of the discourse are, who the participants and their roles are, what they know and do not yet know, in what setting the discourse is being understood, and so on. Such crucial information is necessary to understand such diverse properties of discourse as its intonation, lexical and syntactic style, which meanings are expressed or left implicit and what speech acts are being performed.
  • Once formed --or updated-- a (mental) model of a discourse, language users may generalize such models and thus construct more general, more abstract knowledge structures. It is in this way that language users may learn from their experiences. Of course, they may also make false generalizations, and thus, form prejudices.
  • The process of discourse production may be characterized in a similar way, but in a different direction, starting with a mental model: something you know or have an opinion about, is gradually (strategically, step by step) transformed in the meanings of a discourse, and then expressed word for word, following the constraints of the context model.

    Although the theory just summarized is more or less the standard-theory (with the exception of the role of context models, which is not yet generally known and accepted, but consistent with current theory), there are many other cognitive properties of the mind involved in the processing of discourse. Thus, discourses are also produced and understood as a function of socially shared attitudes and ideologies, norms and values, and possibly other forms of 'social cognition'. So far there is not get a generally accepted theory of these forms of socially shared mental representations, nor of their role in discourse processing.

    3. What is "cognitive analysis"?


    Against this (highly simplified) background of a theory of discourse processing, we may try to develop a type of discourse analysis that takes into account the mental representations and processes described above. For traditional linguists and discourse analysts, as well as for many social/critical discourse analysts as well as 'discursive psychologist', such an approach may be too 'mentalistic' -- either because it is not about text structure but about mental structure, or because it ignores the important interactional and social aspects of discourse.

    Nothing is less true, however. First of all, the most important properties of discourse, namely those that define its meaningfulness (overall meanings or topics, local meanings, coherence, etc), are no less mental than the type of mental representations introduced above: A meaning of a text is not 'in' the text, or on paper, or in the air, but assigned to a text by language users, and as such represented in their minds. The vast majority of psychologists and many linguists accept this as part of standard theory. The question is rather which of the (other) mental representations (models, knowledge, etc) are necessary to account for discourse processing, and how such processing takes place exactly.

    A cognitive analysis as intended does not at all exclude a further social analysis. Indeed, many aspects of cognitive representations and processing are themselves social -- such as the socially shared knowledge and other beliefs, as well as the jointly constructed social aspects of the context. Indeed, discourse processing and understanding is studied at all levels as part of a communicative event, as a form of social interaction, for which in fact it provides a further cognitive basis: also action and interaction derives its meanings, functions and coordination from cognitive representations that ongoingly monitor it. In that respect a theoretically adequate account of discourse as social interaction is unthinkable without a cognitive component, in the same way as a cognitive account is incomplete without a social component that explains structures of context, as well as the acquisition, change and uses of socially shared representations such as knowledge and other beliefs.

    Psychological study vs. cognitive analysis of discourse

    "Cognitive analysis" of discourse, however, is NOT the same as a psychological study of discourse processing. Psychology focuses on the structures and processes of mental representations, and does so, for instance, with experiments, or using other forms of evidence of what actually goed on in the mind. So, cognitive analysis is not going to measure reading or reaction times, or any other method psychologists use to test their hypotheses.

    Cognitive analysis is focused on discourse and its structures, but derives its terms from the theory of discourse processing. For instance, in order to be able to specify the conditions of discourse coherence -- that is, an essential property of (semantic) discourse structure -- we need cognitive notions such as mental models and knowledge. In this case we could not care less whether such an analysis is rather linguistic or psychological -- because meaning and its properties are simply the object of both, and we are thus working in a boundary area where linguistics and psychology overlap.

    A cognitive analysis, thus, is an analysis of those properties of discourse that are accounted for in terms of cognitive concepts, such as various types of mental representation.

    Indeed, there are structures of discourse that are rather generally accepted as being properties of discourse, but which in fact are usually defined in cognitive terms, such as metaphors, overall topics or themes, coherence, presupposition, relevance, and so on.

    Thus, a topic or macroproposition is not necessarily expressed in a text (although it might, for instance in a Headline or Conclusion category of some schematic structure), but something assigned to discourse by language users. In many respects it is thus part of the (global) meaning of the text, but its definition is given in terms of propositions that are assigned or derived from discourse by language users in processes of understanding or production. A presupposition can only be properly defined in terms of knowledge that is assumed to be shared with recipients. Similarly, a metaphor cannot be accounted for only in semantic terms, but needs to be described and explained in terms of cognitive processes, representations or the structures of knowledge. And finally, all discourse meanings and discourse processing are based on knowledge, but knowledge is not a linguistic but a cognitive category. In other words, many of the interesting properties of discourse require a cognitive analysis, whether or not they traditionally were (also) accounted for in linguistic semantics.

    Thus, modern discourse semantics is both linguistic, as well as cognitive (as well as social, for that matter).

    Although ALL of semantics is thus also cognitive, a cognitive analysis focuses in particular on those aspects of discourse meaning that are accounted for in cognitive terms. This means that we do not so much focus on the meanings of words and sentences, which can also be accounted for in classical linguistic semantics, but rather on those properties of text and talk that can only be accounted for in cognitive terms. In the rest of this introductory note, we shall discuss some of these 'cognitive' aspects of discourse meaning.

    4. Elements of cognitive analysis

    As suggested, a cognitive analysis may be given of all semantic structures (and indeed other structures) of discourse. In what follows, however, a limited number of typical discourse structures are mentioned which specifically (also) need a cognitive analysis. We begin by global meaning structures and then pay attention to more local meanings of discourses and sentences.


    Topics are formally defined as (macro) propositions that can be derived from sequences of (macro) propositions of a text. However, an empirically more adequate definition of topics is given in terms of the global meaning assigned to or inferred from (fragments of) discourse by language users. Such a global meaning may be subjective or biased -- it need not be the same for the speaker and the recipient, nor for various recipients. Indeed, topics may thus be defined as the top (macro) propositions of the Text Representation or of the mental model assigned to the text by speaker/writer or recipient.

    How do we know this? Well, language users often provide their (own, subjective) summaries of a text, and such a summary in many respects expresses the macrostructure of the text. This global inference of topics by the language user is not ONLY based on word or sentence meanings of discourse, but also needs vast amounts of knowledge that is NOT provided by the text, but by the knowledge representations of the language users. A cognitive analysis of topics, then, involves spelling out those knowledge items (or information) that were used or need to be used by participants in order to derive (their) topics from a given text -- for instance as expressed in headlines, titles or conclusions.

    Implications and implicatures Much semantic analysis of discourse casually uses the notion of 'implication', in the same way as pragmatic analysis uses the related notion of 'implicature'. However, implications and implicatures are strictly speaking not part of the semantic representation of a text, as traditionally defined, but propositions inferred from (the meaning of) actually expressed words, phrases, clauses or sentences of discourse. Again, there is a formal definition in terms of semantic entailment (implication based on the meaning of terms, e.g., 'dead' entails 'not alive', etc.), but in most discourse analysis that would neither be sufficient, nor indeed necessary. Implication in discourse is used in a much less strict way, and we may say that if two states are at war, this usually implies that they fight and use arms and armies to do so. This is not a logical implication, nor merely part of the meaning of 'war', but rather propositions derived from our socially shared knowledge about wars. Thus, spelling out the (cognitive) implications of a proposition expressed in the text, means to list (some of) the propositions that may be derived from this propositions given a relevant domain of knowledge. Such a list may be very long, and hence impractical, so that we usually spell out only those implications that are relevant in the current context. One of the elements of this context (NB: a mental model of the reader or the analyst !!!) may be our assumptions about the intentions of the speaker/writer. In that case we say: By saying A, the speaking probably implies that B, C and D. Or we can specify our own model of the context, and say: For me, A implies E, F or G.

    The same is true for (pragmatic) implicatures, which are inferences based not only on meaning, but on 'meaning' of speech acts or other communicative acts in specific contexts. Thus, not responding to a question may 'implicate' that the recipient does not want to continue speaking about a topic.

    We see that this (as well as all other) cognitive operations are always formulated in relation to a specific context, for instance the intentions, goals or beliefs of speakers/writers and recipients.

    Presuppositions Presuppositions are a specific type of implications. In formal terms, a proposition is a presupposition of p when it is implied by p as well as by non-p. Thus, the proposition 'John used to beat his wife' is a presupposition of 'John stopped beating is wife' because it is implied by the latter proposition, as well as by the proposition 'John did not stop beating his wife'.

    As in the other examples discussed here, however, we favor a more realistic, psychological account of presuppositions. In such an account, a presupposition of a proposition p is any proposition that must be true (or rather: accepted by the participants) for p to be meaningful (understandable) for the language users. In other words, presuppositions are simply the set of meaning conditions of a sentence: what we must know in order to understand a sentence (or sequence of sentences). Thus, whereas the proposition that 'John is dead' generally implies that (we believe that) John is no longer among the living, the presupposition set of such a proposition would include such propositions as 'John died', 'John probably died because of an illness, accident or homicide', etc.

    The "linguistic" nature of presuppositions is given by the fact that presuppositions are not just any odd piece of knowledge we may have to understand a part of discourse, but the fact that there are expressions in the text that express or otherwise signal such presupposed propositions. Thus, if we say that 'even the terrorists took pity of the victims,' then the use of "even" presupposes that terrorists are usually pitiless. The same is true for many other expressions (the use of the definite article 'the' often presupposes the existence of something), as well as preposed that-clauses, as in 'That the elections were rigged, was regretted by all politicians' which presupposes that the elections were indeed rigged.

    In discourse analysis, and especially in critical discourse analysis, presupposition analysis especially focuses on those presuppositions that suggest that some proposition is (accepted to be) true, but in fact is not true at all, or at least controversial. Thus, if police or media report that energetic action is being undertaken against the "rising crime among minorities", such an expression may falsely presuppose (or indirectly assert) that the crime rate among minorities is indeed rising.

    Local coherence Discourse is coherent not only when it is globally coherent (has a topic), but also when its respective sentences (propositions) are locally or sequentially coherent. Such coherence was originally often accounted for in terms of meaning relations between subsequent propositions. However, we now usually define coherence simply in terms of mental models: A text is coherent if its has a mental model; or more psychologically: a text is coherent for A is A is able to assign a mental model to it. In other words, A is able to imagine a situation in/for which the text could be true. In other words, when cognitively analyzing the coherence of the text, we examine the relations between its subsequent propositions, and establish relative to what mental model the text makes sense. This kind of coherence may also be called 'referential' or 'extensional' because it is defined in terms of the (mental models of the) events the text is about.

    There is however also a form of coherence based on meaning, propositions and their (functional) relations, for instance when a sentences B is a Generalization, Specifiation, Example, Consequence, Presupposition, etc of sentence A. Again, such functional relations are not (only) based on conceptual knowledge (as in the fact that a chair is a piece of furniture), but also on broader world knowledge (what chairs look like, are used for etc.). Again, such relationships between propositions may be critically studied, e.g., when someone says that all Moroccans are fanatic as a generalization of the specific sentence that Mohamed is a fanatic. Thus, both for the study of referential coherence as for the study of this kind of 'intensional' coherence, we need to spell out the knowledge or beliefs needed for participants to be able to establish this coherence.

    Lexical meanings; connotations
    In the same way as propositions may have implications and presuppositions, also word meanings may have specific 'implications', often called 'connotations'. These connotations are not always -- or seldom -- in the dictionary, but often assigned on the basis of the cultural knowledge of the participants. Thus, we may describe the same group from different political perspectives and thus wind up being its ally, or its enemy, e.g. when we describe them as 'terrorists' or freedomfighters'. In this case, the first word has rather negative connotations, whereas the second is rather negative. In both cases, we activate our social knowledge about these groups, and then depen on point of view (and perspective).
    5. Conclusion

    We have only given a short list of examples for cognitive analysis. However, we see that a cognitive analysis generally applies to (semantic) structures such as the following:

    - Defining the situation; defining as overall meanings (topics)

  • Examine relevant implied meanings of words or sentences
  • What is being presupposed (as knowledge)?
  • How does the text cohere? etc.
  • Most of these semantic structures, as well as many others, can only be accounted for in terms of personal or socially shared knowledge, and require listing relevant knowledge or other beliefs.

    In future notes, I hope to be able to further detail this kind of cognitive analysis. This is only a first step.

    Please send comments to vandijk?  (replace ? by @)