The smallest unit of understanding, according to Kintsch (1974, p. 32), is the proposition. He bases his assumption on semantics and argues that a proposition is a semantic structure that is made up of one or more arguments (that realise semantic functions such as agent, object, etc.) and a relational concept between such arguments (the logical relationship holding between the arguments concerned). In order for propositions to be understood, the reader must have two kinds of memory: episodic memory, which helps processing what is actually verbalised in the text, and semantic memory, which contains the knowledge of vocabulary items used in the text. The full text, however, will constitute one meaningful unit only if the text is coherent, which, in Kintsch’s (1974) notion, was equated with referential coherence: “referential identity has been suggested as a major test of coherence” (qtd. in Young 1977, p. 20). It can thus be concluded that Kintsch (1974) saw text comprehension as a linear, non-hierarchical process.

Connecting memory and text comprehension, Crothers (1972) takes up the Bartlettian (1932) notion of memory, according to which building on past experiences and remembering are a constructive process during which previous memories influence the understanding and recall of a text actually being read. Thus, Crothers (1972) asserts that as a result of earlier textual experiences, former related semantic representations, which are made up of concepts and their relationships, exist in the memory and are activated when one is processing and understanding a text. Crothers’ (1972) theory already points towards a hierarchical structure of text comprehension with contents of the memory providing a kind of basis for the concepts activated through reading a given text. Furthermore, it is along these mental concepts that the actual comprehension of a text takes place. That being the case, this theory already presupposes hierarchical relationships between different concepts in the mind and the textual input.

Taking this further, Meyer (1975) relies on case grammar and the structure of rhetorical predicates established by Grimes (1975). Grimes’ (1975) rhetorical predicates are practically a hierarchical concept, which entails that the pragmatic purpose of a text will surface in the textual relations between the individual sections of the text in question. In other words, a report or a scientific study, for instance, will predominantly contain question-answer relations, and thus the text will be structured in a way that higher ranking predicates constituting answers always take lower ranking predicates constituting questions as their arguments so that the realisation of the pragmatic purpose referred to above can be ensured. Based on Grimes (1975), Meyer (1977) examines the recall and structure of paragraphs and concludes that the surface of texts signals diverse rhetorical predicates, which seem to be arranged in a hierarchical structure pointing towards the intended pragmatic aim of the text. This conclusion is also underscored by Meyer’s (1975) later findings, namely that any content of the text that is featured higher in the rhetorical structure will be better remembered than those at lower levels in the rhetorical structure, and that contents occupying lower levels will start to fade in memory right after reading the text.

In an attempt to provide further explanation for the hierarchical nature of text processing and text recall, Kintsch and van Dijk (1978) as well as van Dijk (1979) proposed a model describing the comprehension of entire pieces of texts and did not restrict their research to paragraphs as it was done earlier in Kintsch’s (1974) or Meyer’s (1975) research. The starting point of Kintsch and van Dijk’s (1978) hierarchical model, called the Model of Text Comprehension and Production, is Kintsch’s (1974) finding. Kintsch (1974) developed his model for determining propositions as well as for analysing cognitive processes related to text processing with reference to text recall mechanisms of producing gist and reading for gist. This model, in our view, presupposes an ideal Fillmorian reader (Fillmore 1981), who is aware of any implicit and explicit text base necessary for understanding the text in question. Furthermore, the model assumes that text comprehension is in fact a linear process that takes place while listening to or reading a piece of text. To put it very plainly, the model describes the semantic structure of texts and the semantic interrelations contained in the text.

As opposed to Kintsch’s (1974) model, in Kintsch and van Dijk’s (1978) Model, during text comprehension the words of the actual text, by their semantic functions, act as propositions (i.e. statements expressed verbally) at the lowest level of discourse. That is, words in a text, while constituting propositions, function as either arguments (i.e. concepts expressed by the text or, in other words, information) or predicates (interrelation of concepts). A proposition, as a rule, contains a predicate and one or more arguments at the level of microstructure. In order to describe the propositions (containing both arguments and predicates) and the microstructure of a full piece of text, Kintsch and van Dijk (1978) use one paragraph of a newspaper article entitled ‘Bumperstickers’ for demonstrating the application of their theory. The paragraph of the newspaper article reads like this:

 

On the other hand, microstructure is the web and network of propositions and their relations. Propositions in a coherent text can be understood by the receiver if a given proposition shares at least one argument of another proposition: this minimum requirement is argued to be necessary for making a text understandable or coherent (Kintsch and van Dijk, 1978, p. 368). Practically speaking, such common arguments make the text coherent by connecting to the referents, and ensure that readers are thus able to connect the referents to one another in their mental representation of the text in question and consequently interpret the text to themselves. As a result of the linear process of text comprehension, the mental links that are established by semantically interlinking the propositions in one’s mental representation of the text in question can be mapped with the help of so-called coherence graphs. These depict how the propositions with (a) common argument(s) are linked to each other, as shown through an example in Figure 3.1.

 

Understanding a text with an implicit text base is possible because readers are able to provide the missing logical and/or event links (relations or events of the plot not stated expressly by the surface text, which are added as propositions by the reader) relying on their general and contextual knowledge of facts associated with the piece of text in question. Consequently, as part of the process of inferencing (providing missing links or propositions) briefly described above, readers, on the one hand, develop fundamental text-level relations between immediate propositions, and, on the other hand, establish logical relations between the meanings contained in the text as a means of making the actual text coherent for themselves. Coherence, therefore, is realised if “respective sentences and propositions are connected, and if these propositions are organised” (Kintsch and van Dijk, 1878, p. 365) in the mental representation of the readers. (It is to be noted, nonetheless, that this definition has been subject to criticism – e.g. Young 1977 – for featuring the subjective idea of propositions being simply connected, without any specification as to how connection is established.)

When readers make a piece of text coherent for themselves, they establish text bases, which are, in turn, organised into macropropositions. Macropropositions are higher-ranking propositions encapsulating more than one proposition at the level of macrostructure, above the level of microstructure. Macrostructure is a textual feature that relates to the global nature, the gist of discourse as a whole. In short, macrostructure is the semantic structure or semantic content of a text; or, in other words, the hierarchically arranged structure of the propositions that make up the actual text. In the next paragraphs, Macrostructure will be discussed in detail below.

Macropropositions are in fact propositions or the generalisations thereof pointing towards the macrostructure or the gist of the text in question and are selected from among all propositions with the help of so-called macrorules or macro-operations. Such macrorules or macro-operators denote mental processes through which propositions of a text get hierarchically organised along their semantic content. To put it simply, the working of these macrorules is based on common semantic denotations and rules that follow from the semantic content of the propositions themselves. Macropropositions will constitute one interpretable and meaningful unit, as the propositions they are based on at the level of micropropositions are also meaningful.

As a first step in the construction of the macropropositions of the macrostructure, macro-operations will form generalisations of micropropositions (acknowledged to be a somewhat intuitive process by Kintsch and van Dijk [1978] themselves). Once these generalised macropropositions have been formed, another factor, schema comes into play. Schema is the readers’ mental representation of their prior reading experiences and their consequent expectations of texts of the same kind. The schema is based on readers’ former reading experiences of having been exposed to different genres, discourse types or cognitive schematic structures. Through these schematic structures “language users are able to understand a discourse as a story” (Kintsch and van Dijk, 1878, p. 366) or any other text as a specific type of text. Schema then is the formal representation of “the reader’s goals in reading [that] control the application of the macro-operators” (Kintsch and van Dijk, 1878, p. 373). In other words, these former reading experiences will surface as reader expectations concerning a piece of text and will guide the reader in constructing the macrostructure (or the gist) of the text. The construction of macrostructure is executed through several cycles of macroproposition reduction at different levels of the macrostructure. This means that the schema will also influence which macropropositions and which parts of these macropropositions will finally make it to the highest level of the macrostructure (i.e. the shortest possible but still meaningful summary) of a given text. The set of eventually selected macropropositions will constitute the highest level of macrostructure of a given text. It follows from the above that the different levels of the macrostructure of the text are in fact the summary of the text in differing detail and lengths. (Macrostructure as the summary of a text will be further clarified below in this section in the discussion of van Dijk’s [1980] Macrostructure Model.) As far as text recall is concerned, Kintsch and van Dijk (1978) conclude that readers will remember the summaries of texts, referred to as the macrostructure of a text, and note that macrostructures are mentally constructed at the time when the text in question is read.

Apart from propositions and the resulting macrostructure, Kintsch and van Dijk (1978) as well as van Dijk (1977 and 1978) discuss another level of text structure called superstructure above the level of macrostructure. Superstructure (also called schema in Kintsch and van Dijk [1978]), apart from being the linear structure of a text and its parts – such as the introduction, discussion, etc. –, organises the macrostructure of a text as suited to the actual text type: i.e. the superstructure will influence which propositions (potentially both micro and macro) will surface in the macrostructure. This suggests that superstructures, on the one hand, have their own role to play in making a piece of text conform to the referent text type features, and, on the other hand, the superstructure signals the given text type to the reader. The above researchers then concluded that reader expectations about the text type of a piece of discourse plays a fundamental role in the cognitive organisation of any given text. This also seems to suggest that the superstructure also influences the macrostructure of texts. Consequently, Hoeys’s (2001) Problem-Solution Model plays a vital role in the construction of the macrostructure of argumentative texts, cf. Section 3.4 on Tirkkonen-Condit’s (1985) Macrostructure Model.

Kintsch and van Dijk’s (1978) Model of Text Comprehension and Production was introduced, further developed and described in more detail in van Dijk’s (1980) Macrostructure. An Interdisciplinary Study of Global Structure in Discourse, Interaction and Cognition. This work focuses on the hierarchical nature of macrostructure in greater detail and establishes a hierarchical structure within the macrostructure. Nonetheless, the theoretical foundations and terms used in van Dijk (1980) are largely identical with those used in Kintsch and van Dijk (1978) presented above. Below, van Dijk’s (1980) Theory of Macrostructure will be described, and consequently some of the notions already described (e.g. macrorules) will be further discussed in connection with macrostructure.

The existence of macrostructure is suggested by the everyday experience that readers are able to extract main ideas of discourses and make summaries of discourses. This fact, therefore, presupposes that such summaries can be constructed as texts can meaningfully and systematically be shortened. This realisation points towards the existence of a hierarchical text structure. This is so as summaries capsulate the most important information contained in texts, which presupposes that certain discourse elements are more important than others.

In this light, summaries of texts are in fact the macrostructures of the discourses in question. As summaries of differing information compactness can be written, it has been theoretically assumed that macrostructure has a hierarchical structure (van Dijk 1980) and that different levels of macrostructure will yield summaries of different detail and length. The shortest and still acceptable summary embodies the highest level of the macrostructure. Van Dijk’s (1980) Macrostructure Model describes the operations through which surface texts (i.e. verbalised realisations taking the form of a connected discourse) are transformed into summaries.

The different, hierarchically arranged levels of macrostructure are derived by macrorules. Macrorules are semantic information based theoretical rules that produce the macrostructure of texts. These macrorules operate on a certain level of the macrostructure to produce the macrostructure one level above the given macrostructure. Van Dijk (1980) distinguishes four types of macrorules: zero, generalisation, construction and deletion. Each of these macrorules works along common shared semantic features of the macropropositions situated at the same level and will, as output, produce the macrostructure of the given discourse at one level higher in the macrostructure, incorporating fewer details and offering more compactness of information content than their input macrostructure. Let us examine these four macrorules in more detail.

The zero macrorule will leave the actual proposition, sentence or sentence sequence (depending on the level of macrostructure actually dealt with, henceforth called discourse string) untouched, and, as a consequence, such a discourse string will surface in an intact form in the macrostructure one level higher. The generalisation macrorule will have the effect of taking several discourse strings at a time, and produce a more general summary of the strings on the next level of macrostructure. The construction macrorule will relate to the knowledge frame constituted by the text base of the text and will draw on readers’ knowledge to produce a discourse string for a higher level of macrostructure. The deletion macrorule will delete the actual discourse string and as such the string will not surface in any form in any higher ranking macrostructural levels.

The operation of these rules is illustrated below. The discussion that follows uses not more than a few sentences (and not wider contexts) as examples to illustrate the operation of these macrorules. In our examples, the zero macrorule will be disregarded as this preserves the actual discourse string as explained above, leaving not much to illustrate.

The generalisation macrorule operates the following way: a sequence of propositions may be substituted by one single proposition denoting a semantically immediate superconcept of the micropropositions involved. This substitution will result in one unifying proposition and can also involve the use of more general terms such as for instance hypernyms.

 

The construction macrorule operates the following way: a sequence of propositions may be substituted by one single proposition if the propositions involved denote regular cases or components of events, normal conditions or consequences stated in the macroproposition substituting for the actual sequence of propositions at a level higher in the macrostructure. Example 2 below demonstrates the operation of the construction macrorule.

The deletion macrorule operates the following way: a sequence of propositions may be deleted if the propositions involved denote an accidental, contextually irrelevant property or feature. Such a property or feature is not necessary for the interpretation of the propositions to follow.

Proposition 1: The dog started to bark.

Proposition 2: The dog had fur.

Proposition 3: The fur was white.

According to van Dijk (1980), apart from these repetitively operating, strictly semantics-based rules, another factor, the superstructure also has a role to play in which discourse strings surface in the macrostructure. This suggests that discourse-type specific elements (e.g. the description of a problem and its solution in an argumentative type discourse) must be retained in the macrostructure, otherwise genre-specific constraints in the macrostructure are violated. It can therefore be concluded that the superstructure places discourse-type specific constraints on the realised macrostructure and that macrorules, as a consequence, are also discourse-type sensitive.

In order to validate Kintsch and van Dijk’s (1978) Model of Text Comprehension and Production, Kintsch’s fellow researchers tested text comprehension and recall processes in empirical experiments. Fletcher (1981) explored whether comprehension was a linear process and if the understanding of one proposition depended on the comprehension of a previous one. Fletcher asked research subjects to read a narrative text and purposefully interrupted their comprehension. Then he asked them to select from a number of options containing the verbalisation of propositions featured in the text those propositions that the research subjects remembered reading in the previous sentence. By interrupting research subjects in their text recall, Fletcher could identify common arguments in the proposition actually recalled by the research subjects and previously comprehended propositions. Fletcher found that in the case of central propositions (i.e. propositions holding several common arguments with other propositions) recall times were about 200 ms faster than the recall times of other, non-central propositions and that central propositions remained more accessible in subjects’ memory over time. Fletcher thus concluded that Kintsch and van Dijk’s (1978) Model of Text Comprehension and Production was a highly reliable description of text comprehension and recall processes. As van Dijk’s (1980) Macrostructure Model is based on the same theoretical foundations as Kintsch and van Dijk’s (1978) Model of Text Comprehension and Production, the Macrostructure Model can also be deemed as reliable.