What is Discourse? 
Discourse refers to the goal-oriented use of language through spoken words and written text. We use discourse in all of our social interactions, and depending on context, our discourse styles - or strategies - may change to achieve our original goal. There is an on-going debate of whether discourse should be studied through a product or process view. The product view focuses on the structural aspects of speech or text, suggesting that discourse comprehension comes from the effective use of linguistic rules (van Dijk, 1972). However, some researchers are more interested in the communication process that happens between people engaged in discourse (H.H. Clark, 1996; Sacks, Schegloff, & Jefferson, 1974). This view looks at the pragmatics of discourse and how people evaluate the situation in which they communicate. Regardless of how one approaches the subject, whether through the product or process view, discourse has three structural levels: microstructure, macrostructure, and mental model or mental text representation. In the following chapter, I will look at how elements of both the product and process views are needed to facilitate and understand discourse at every level. This chapter is by no means an all-encompassing analysis of discourse, but rather an introduction to the fundamentals of the subject.
We will first consider the microstructure of discourse; the individual sentences in the text and how they relate to each other (Glowalla, U., & Colonius, H., 1982). Connections between words and sentences at this structural become the basis for the general meaning of the discourse. We employ linguistic rules to make connections, though the specific connections we make can depend on the pragmatic context of the discourse. The goal at the microstructure level is to achieve coherence; the ability to understand statement in reference to the what has previously been said (van Dijk, & Kintsch, 1983). A coherent microstructure facilitates a clear macrostructure and subsequent mental model. In the following section, we will look at principles and rules that help form coherence.
Cohesion describes the connections between the sentences of text and conversation. It helps achieve the goals intended to be realized through discourse. Cohesion in discourse is attained through rules of grammar and linguistic devices. Anaphora encompass some of the grammatical rules that facilitate cohesion in discourse.
An anaphor is a word in a sentence that refers to another word or statement that occurred earlier in the discourse. There are several categories of anaphora, and we will look at pronouns, demonstratives, comparatives, and substitutions.
Pronouns are words that replace a noun that has already been used in the discourse. For example, if the text reads, Henry went to the store to buy ice cream, the next sentence could read, When he got to there, he could not decide what flavour he wanted. In the second sentence, there is no explicit mention of Henry or the store. But readers understand that he refers to Henry and that there refers to the store. By using pronouns, we avoid repeating nouns every time we refer to them. Think of how tedious and redundant it would seem to read, Henry went to the store to buy ice cream. When Henry got to the store, Henry could not decide what flavour Henry wanted.
Demonstratives are another type of anaphor that contributes to the process and comprehension of discourse. These anaphora point to previous experiences in text or conversation. For example, after walking out of a movie theatre, one might say, That was the worst movie I have ever seen. Because of the context under which the statement is uttered (walking out of a movie theatre) we understand that that refers to the movie.
Comparatives are anaphora that contrast one referent in the discourse with another by using words such as like, different, same. For example, The trees in Nova Scotia are different from those in Costa Rica.
Substitutions are anaphora that replace entire previous statements. For example, if a text reads, Jen writes her last exam tomorrow. After that, she will celebrate with friends. In the second sentence that refers to the entire first sentence, Jen writes her last exam tomorrow.
Anaphora help create cohesive discourse by drawing connections between statements. There is also evidence to suggest that when readers come across anaphora in text, they slow down to connect the anaphor with to the original word it refers, also known as the referent. When the connection is made between anaphor and referent, coherence is established. However, resolving anaphora is not the only way to establish coherent discourse, and it is important to make the distinction between coherence and cohesion. Although cohesive discourse usually makes for more coherent discourse, it is possible to have one without the other. For example, you could have a technically cohesive text, but if pronouns are used extensively without repeating the original noun, readers might start getting confused about which noun the pronouns are referring to; the referent must be repeated frequently enough in order to achieve coherence. On the other hand, a reader may understand some text that contain no anaphora at all. For instance, the text could read, Marcus went to the casino with 20 dollars. He came home empty handed. Even though there is no reference to the 20 dollars in the second sentence, readers understand that Marcus gambled the 20 dollars away. This understanding comes from our ability to make inferences about missing information.
Inferences fill in missing information that is not explicitly stated in the discourse. They are activated from previous information of the current discourse, or from stored knowledge of past experiences. Most of the time, inferences are needed to make sense of the text or conversation. Other times, however, we make inferences that are not necessary for comprehension. We call these elaborative inferences. Take the sentence, He mixed the perfect colour for his still-life painting of a pumpkin. Here, the reader activates the word orange even though it is not needed to comprehend the text. In contrast, we use bridging and causal inferences to make the necessary connections that facilitate comprehension.
Bridging inferences connect statements through commonly held world knowledge, and are essential for comprehensive discourse (Michael Halldorson & Murray Singer, 2002). Consider the following text: The spy quickly threw his report in the fire. The ashes floated up the chimney. Here the readers must draw the inference that the report burned in the fire. (Singer & Ferreira, 1983). If the reader did not know that fire burns paper into ashes, they would not comprehend the text.
Causal inferences are made when the reason for one statement is drawn from information from an other. Narratives are full of causal relationships which allow readers to form connections and make sense of the text. The use of causal inferences is explained through the validation model, which says readers compare the sequence of cause and effect in text with general knowledge to determine whether it is true (Singer, 1993). A study by Singer et. al found that participants responded faster to the question, Does water extinguish fire? after they are primed with information that makes them validate the idea that water does in fact cause fire to extinguish. For example, when presented with the text, Mary poured the water on the bonfire. The fire went out, readers infer that water extinguishes fire. In contrast, when readers are presented with a control sequence, Mary placed the water by the bonfire. The fire went out, they do not infer a cause and effect relationship between the two statements, and therefore, they are not primed with the answer to the question, Does water extinguish fire? (Singer et. al, 1992).
We have just discussed how our ability to make inferences comes from our own general knowledge and from information presented in the discourse. Our inability to speak or write exactly what we mean to communicate often conflicts with our desire to understand text in an organized structure. Inferences resolve this by filling in gaps a of missing information and by connection individual events with others in the text. Causal relationships also help readers conceptualize the text as a whole by bridging how actions and events change the state of things and people in a story (Trabasso, T., & Sperry, L., 1985). Trabasso et al. describe actions as things the protagonist in the narrative does. States represent the physical world or physical condition of a character. And events represent changes in states or when states occur. The sequence of events, actions, and changes in states in a narrative is called a causal chain. Since the causal is from the protagonist’s point of view, what might seem like actions of non-protagonist characters are actually events (Trabasso et al., 1985).
By developing a well-structured causal chain, readers are able to construct a mental representation of the text. However, it is important to note that events in stories do not always unfold in a linear fashion where one statement necessarily leads to the next. In narratives (as in real-life conversations) there are multiple sources of causality (Jay, 2003). Text is therefore thought to be produced and understood as a causal network. The network view takes into account the interconnections between causal chains, which all help to validate the cause of events. The network view also allows for alternate ways of arriving at the same outcome. Furthermore, not all actions and events carry the same degree of significance in a text, so readers tend to construct a mental hierarchy of significance within the causal network. As one might assume, memory is better for events at the top of the hierarchy (Kintsch, et al., 1990; van Dijk, T. A.,1972; Thorndyke, P. W., 1975) . Text memory also appears better for statements with many connections, and these highly connected statements dominate how the reader represents the text as a whole in their memory (Trabasso, T., & Sperry, L. L. (1985).
The Macrostructure 
We now turn to the strategies readers use to generate the macrostructure of a text. The macrostructure takes all of the information given and inferences made in a text, and reduces it to its essential components, leaving the reader with the “gist” of the discourse (van Dijk, & Kintsch, 1983). There is ample evidence to suggest that condensing information into a macrostructure is fundamental to discourse comprehension and retention (Kintsch, & van Dijk, 1978). And much research has explored how we draw from microstructural information to form macrostructures.
As mentioned above, we organize material in discourse literature based on our prior experience to the things or situations being described. Our previous knowledge for particular situations are organized into categories known as schemata. Take, for example, the general knowledge of taking a bus-the ‘bus’ schema, if you will. There are certain elements that are always constant in the bus scheme. The driver and passengers play character roles, passengers pay fare, etc (van Dijk & Kintsch). Although schemata are “prepackaged” reserves of knowledge which we apply to discourse comprehension, readers and listeners maintain flexibility within their schemata and can update them depending on the context of the discourse. Different factors, such as route, number of passengers, weather, etc, cause subtle changes in the bus schema, and depending on the specific context, different information becomes more relevant (van Dijk & Kintsch).
Scripts are similar to schemata, though they are more specific and are not as flexible based on context. Scripts carry detailed information on how events unfold and in what particular order. In the restaurant script, for example, the participants wait to be seated, a host or waiter takes them to their table and gives them menus. The participants are then left to choose something off the menu, they order food, wait for food, then eat it. Then they wait for the bill, pay the bill, and leave (Jay, 2003). Knowing the script guides people on how to behave in a restaurant situation, and provides an outline for how to write stories involving restaurants.
Mental Text Representation 
When a particular script or schema is activated, it becomes the basis of the macrostructure. By binding this schematic knowledge with the semantic units (the microstructure) of the text, the reader moves closer to a clear mental representation (van Dijk & Kintsch). Researchers have found evidence that there are three levels of mental text representation (Flethcher & Chrysler, 1990); surface structure, textbase, and situation model. The surface structure refers to the exact words and phrases of the discourse. The textbase encompasses the semantic content of the discourse; the meaning of the sentences and how they connect. At the textbase, readers draw knowledge from stored memory to make sense of the surface structure. Elements of both the microstructure and macrostructure are involved in this level. The third level, the situation model, does not represent any aspects of the text itself. Rather, it represents what the text is about - what it describes (Kitsch et al., 1990). Quite simply, the situation model is the “gist” of the text.
In order to form a coherent representation of the text, readers must process information at all three levels. However, rates of memory decay differ between each level (Vellutino et al., 2000). In a study by Kitsch et at., memory tests were conducted for text information with delays ranging from immediately following the task to four days later (Kitsch et al., 1990). Subjects read a story, then completed a recognition task where they were given 1) verbatim sentences from the story, 2) paraphrases of the story, 3) sentences that could be inferred from the story’s content, 4) novel sentences that were semantically consistent with the story, and 5) novel sentences that had nothing to do with the story. Performance on the recognition tasks was compared to measure differences in memory decay between the levels of mental text representation. Verbatim sentences versus paraphrases measured memory for surface structure. Comparisons between paraphrases and inferences looked at the strength of the textbase. And differences between inferences and novel consistent sentences measured memory strength of the situation model. Results of the study show that surface structure memory is strong immediately following the story reading, but is quickly lost. Textbase memory started high, but delayed gradually over the four days. Memory for the situation model started high, and remained strong for the entire test interval (Jay, 2003).
The Story Grammar 
From our previous discussion it is already evident that memory plays a central role in discourse production and comprehension. We now look at how the structure of a story can influence the efficiently with which we commit the story’s material to memory. Much of the literature examines memory for narrative text through the story grammar framework. Story grammars refer to the typical structure that readers expect a story to follow (Thorndyke, 1975, 1977).
Researchers have found that recall for stories is highly dependent on the structure of the plot (Thorndyke, 1975,1977; Stein & Glenn, 1979, Mandler, 1984,). Consistent with the idea of multi-level mental text representation, Thorndyke found memory was better for the general meaning of the story as a whole, rather than for specific details or events. He also found that subjects remember the overall structure of the story, and when this structure was manipulated, recall is compromised (Thorndyke, 1975, Mandler, 1984). Integral to every story grammar is a setting, theme, plot, and resolution. The setting is presented at the beginning of the story, and introduces the overall context of the story and the character involved. The setting tells the readers what is plausible based on spatial and temporal conditions of the story. For example, in a fairy tale, readers can expect dragons and witches, but in a news story, one expects information to be consistent with reality as they know it. The theme of the story either states or implies the character(s)’s main goal. And the plot adheres to the theme of the story, and consists of all the episodes. Episodes are segments of text that are organized around attempts to achieve the goal set out at the beginning of the story. Within the story structure, there are initiating events, outcomes, and goals. There are also attempts, internal responses, and reactions which come from the main character in relation to achieving the desired goal; however, mental text representation and memory are much stronger for initiating events, outcomes, and goals than for the later elements (Jay, 2003).
Partial Processing 
Sometimes people think they comprehend discourse when really they do not. When this occurs, partial processing is usually at play. Take, for example, the Moses Illusion. If I ask, How many of each sort of animal did Moses put on the Ark? you will likely respond, two, right? Wrong. On second thought you might realize that it was Noah’s-not Moses’-Ark. There are different theories as to why we partially process discourse and why we fall into traps like the Moses Illusion. One theory claims that our processing of semantic information is very general, making us less likely to pick up on specific details. Another theory suggests that we verify incoming semantic information with only a small portion of what is stored in our memory because verifying information with our entire memory takes too much time and effort (Shafto & MacKay, 2000). Therefore, when incoming information is semantically similar to what we expect, we prematurely conclude that we comprehend what is being said. According to Erickson and Mattson (1981), Moses and Noah-both being biblical figure-share a close enough semantic relationship to make them susceptible to the described illusion. Similarly, Shafto and MacKay suggest that readers and listeners can be tricked by words that are phonetically alike. He illustrates this through the Armstrong Illusion and asks participants, What was the famous line uttered by Louis Armstrong when he first set foot on the moon? He found that people answer the question without questioning its validity. This is because Neil and Louis are phonologically similar. Despite the differences in the Moses and Armstrong illusions, Shafto and MacKay suggest that both occur because the unpresented word (Noah/Neil) is primed more than the presented word (Moses/Louis). In the Moses Illusion, Moses indirectly activates Noah through bottom-up and semantic priming. In Armstrong’s case, Louis activates Neil. Words like moon and famous activate the idea of an astronaut, and given the content of the question, Neil is also semantically primed (Jay, 2003; Shafto & MacKay, 2000).
Discourse as Social Communication 
Most of the literature in psycholinguistics looks at discourse in terms of text structure and comprehension. But as the very purpose of discourse itself is rooted in social communication, it is important to examine the social motivations for using discourse and how these motivations drive how we interact with language. We will explore this through the realm of discourse production as conversation.
Speech Acts 
Keeping in mind that “language is for doing things” (Clark, H.H., 1996), discourse production can be understood as the planning and execution of actions under specific social contexts. Here, actions refer to certain kinds of goal-oriented events that establish or prevent changes in the world (van Dijk & Kintsch, 1983). This planning and execution of a particular action in discourse is called a speech act, i.e. doing something with language. Examples of speech acts are promises, requests, congratulations, accusations, etc.
Before delving into a conversation with someone, the speaker (or actor) must first evaluate the social context of the situation and set out a strategic plan of discourse production accordingly. Say, for example, that someone is being located to Afghanistan for work, and for whatever reason - be it you will miss them too much or are concerned for their safety - you do not want them to go. In such a case, you would use discourse to arrive at a desired outcome where the person does not go to Afghanistan. To arrive at this outcome, you need to plan a strategy and decide what kind of speech act to use. Assuming the person is your friend, commands would seem mean, and requests might come across as selfish. Warnings, advice, or suggestions might show your concern for the person, and would likely be your best plan of action (van Dijk & Kintsch, 1983).
Like written text, conversational discourse also has micro and macro components. The macro speech act is the overall intention or goal of the speaker, while the micro speech act refers to the individual exchanges of information between the speaker and listener in the conversation. The strategic planning that was just discussed is driven by the macro speech act (I don’t want you to go to Afghanistan, is the macro speech act in the example above). We use micro speech acts to arrive at the overall goal, of which the macro speech act is the motivation. Previous speech acts and the hearer’s reaction to them also influence what the speaker will say next (van Dijk & Kintsch, 1983). So the challenge of the speaker is to reach their goal, while keeping in mind (and in conversation) the reactions and micro speech act of the hearer.
Although the listener of a conversation take cues from the speaker's utterances (micro speech acts), the pragmatic context of the conversation largely influences how the listener interprets the discourse. The relationship between the listener and speaker, their cultural norms, gender, and social positions all influence the nature of the conversation. Typically, people extract the meaning of a statement of conversation from this pragmatic context more-so than from utterances. The idea is that if the listener has enough pragmatic awareness, they can make inferences about the speaker's conversational intentions, and thus predict where the conversation is going before getting cues from micro speech acts (van Dijk & Kintsch, 1983). ]
This brief discussion on conversation is assuming the interactions are under very controlled conditions. In reality, conversations are usually spontaneous, and micro speech acts are are highly dependent on novel and surface information from the conversation.
Leaning Exercises 
1. Thomas is studying math in the library. He has an exam tomorrow at 10:00 am, and he doesn’t want to fail math for the second time. His friend Chris calls him, insisting he goes with him to a party. Thomas is hesitant, but Chris assures him there will be lots of beer and good-looking girls at the party. The next morning, Thomas wakes up at 10:30 am, and cannot remember how he got home. a) What is the situational model, or the “gist” of the above passage? b) About how old is Thomas? c) What exam is Thomas writing? d) Why is he writing the exam? e) What did Thomas drink? f) What (unstated) problem does Thomas have when he wakes up?
2. Simon is sitting near the tree, counting his gifts. “I have one more than Lucas!” he says delightedly. Lucas rolls his eyes and continues reading his novel. The boys’ mother comes into the room and orders Simon to go to bed. “Santa Clause won’t come if you’re up all night,” she says. a) List the anaphora in the above text b) List all inferences you make while reading the above text.
Clark, H.H. (1996). Using language. Cambridge: Cambridge University Press.
Jay, T. (2003). The psychology of language. Upper Saddle River, N.J: Prentice Hall.
Kintsch, W., Schmalhofer, F., D., & Zimny, S. (1990). Sentence memory: A theoretical analysis. Journal of Memory and Language, 29, 133-159.
McKoon, G., & Ratcliff, R. (January 01, 1989). Semantic associations and elaborative inference. Journal of Experimental Psychology. Learning, Memory, and Cognition, 15, 2, 326-38.
Sacks, H. Schegloff, E.A., & Jefferson, G. (1974). A simplest systematics for the organization of turn-taking for conversation. Language, 50, 696-735.
Singer, M., Halldorson, M., Lear, J. C., & Andrusiak, P. (1992). Validation of causal bridging inferences. Journal of Memory and Language, 31, 507–524. 507–524.
Singer,M. (1993). Causal bridging inferences: Validating consistent and inconsistent sequences. Canadian Journal of Experimental Psychology, 47, 340–359. Canadian Journal of Experimental Psychology, 47, 340–359. 340–359.
Squire, C. (January 01, 2005). Reading Narratives. Group Analysis, 38, 1, 91-107.
Thorndyke, P. W. (1975). Cognitive structures in human story: Comprehension and memory. Santa Monica, Calif: Rand Corp.
Trabasso, T., & Sperry, L. L. (1985). Causal relatedness and importance of story events. Journal of Memory and Language, 24, 612-630.
van Dijk, T. A., & Kintsch, W. (1983). Strategies of discourse comprehension. New York:Academic.
van Dijk, T. A. (1972). Some aspects of text grammars. The Hague, Netherlands:Mouton.
Vellutino, F. R., van, O. H., & Goldman, S. R. (January 01, 2000). Contemporary Theory and Research in the Study of Discourse Processing. The American Journal of Psychology, 113, 2, 306-318.