Discourse

The field of discourse processing investigates the structures, patterns, mental representations, and processes that underlie written and spoken discourse. It is a multidisciplinary field that includes psychology, rhetoric, sociolinguistics, conversation analysis, education, sociology, anthropology, computational linguistics, and computer science.

Researchers in discourse processing have identified a number of mechanisms that promote learning. The practical mission of the field is to improve the comprehension and production of discourse in textbooks, tutoring sessions, classrooms, computer-based training, and other learning environments. While focused primarily on cognitive mechanisms, it is clear that cognitive, social, emotional, and cultural foundations are tightly intertwined in contemporary theories of discourse processing.

Levels of Discourse Processing

Discourse researchers have identified five levels of cognitive representation that are constructed during comprehension. These include the surface code, the textbase, the situation model, pragmatic communication, and the discourse genre. In order to illustrate these five levels, suppose that a high school student had a broken door lock and was reading the following excerpt from the book The Way Things Work:

Inserting the key raises the pins and frees the cylinder. When the key is turned, the cylinder rotates, making the cam draw back the bolt against the spring. (Macaulay, p.17)

The surface code is a record of the exact wording and syntax of the sentences. This code is preserved in memory for only a few seconds when technical text is read. The textbase contains explicit propositions in the text in a stripped-down form that captures the semantic meaning but loses details of the surface code. For example, the textbase of the first part of the second sentence includes the following: (1) someone turns a key, and (2) the cylinder rotates when the key is turned. The textbase is preserved in memory for several minutes or longer.

The situation model (sometimes called the mental model) is the referential mental world of what the text is about. In the above example, the situation model contains causal chains of events that unfold as the key unlocks the door, a visual spatial image of the parts of the lock, and the goals of the person who uses the lock. The construction of an adequate situation model requires a sufficient amount of relevant world knowledge, such as general knowledge about locks and mechanical equipment. Deep comprehension consists of the construction of this referential situation model, whereas shallow comprehension is limited to the surface code and textbase. The situation model is retained in memory much longer than the textbase and the surface code, assuming that the comprehender has adequate world knowledge to build a situation model.

The pragmatic communication level refers to the information exchange between speech participants. In a two-party oral conversation, the two speech participants take turns speaking while pursuing conversational goals. There may be additional participants in a conversation, such as side participants in the circle of conversation and bystanders who are outside of the circle. Speech acts are crafted in a fashion that is sensitive to the common ground (shared knowledge) between speech participants, and linguistic cues differentiate given (old) information in the dialog history from new information. The cognitive representation of a spoken utterance can be quite complex when there are several communication channels between multiple participants (sometimes called agents) in a conversation. When printed text is read and comprehended, the pragmatic communication is somewhat simplified, although there are vestiges of oral communication and multiple communication channels in textual matter. For example, there is communication between the reader and writer, between the narrator and audience, and between agents in embedded dialogues within the text content. Text comprehension improves when readers are sensitive to the communication channel between author and reader.

Discourse genre is the type of discourse–such as narration (stories), exposition, persuasion, and so on. Discourse analysts have proposed several different discourse classification schemes, which are organized in a multilevel hierarchical taxonomy or in a multidimensional space (a set of features or levels of representation that are potentially uncorrelated). The Macaulay excerpt above would be classified as expository text. Narrative text is normally much easier to comprehend than expository text because narrative has a closer affinity to everyday experiences.

Deep comprehenders construct rich representations at the levels of the situation model, pragmatic communication, and discourse genre, whereas the textbase and surface code have a secondary status. Paradoxically, the examinations that students normally receive tap the surface code and textbase rather than the deeper levels. Teachers generally ask students to recall explicit content or to answer multiple-choice questions that tap word recognition, definitions, or attributes of concepts. One way of promoting deep comprehension is to compose exams with questions that emphasize the situation model, inferences, reasoning, and other aspects of the deeper levels. Since the late 1980s, researchers have advocated a shift in assessment standards to encourage deep comprehension.

Discourse Coherence

Coherence is achieved both within and between the levels of representation when comprehension occurs. This means that there should be no serious coherence gaps within a particular level and there should be harmony between the levels of representation. A coherence gap occurs within the situation model, for example, when an incoming clause (the clause currently being read) in the text cannot be linked to the previous content on any conceptual dimension, such as causality, temporality, spatiality, or motives of characters. Simply put, a coherence gap occurs when information is mentioned out of the blue. Similarly, there may be coherence gaps at the levels of the surface code, textbase, pragmatic communication, and discourse genre. Regarding coherence between levels, the elements of the representation at one level need to be systematically related to the elements at another level. Comprehension suffers, for example, when there is a clash between the textbase and situation model. If the text stated, "The key is turned after the cylinder rotates," there would be a discrepancy between the order of events in the correct situation model (the key is, in fact, turned before the cylinder rotates) and the explicit textbase, which reverses the correct order.

Comprehension breaks down when there are deficits in world knowledge or processing skills at particular levels of representation. When there is a deficit at one level of representation, the problems can propagate to other levels. For example, nonnative speakers of English may have trouble processing the words and syntax of English, so they may also have trouble constructing the deeper levels of representation. Readers have trouble comprehending technical texts on arcane topics because they lack world knowledge about the topic. A barrier in constructing the situation model ends up confining the processing to the surface code and textbase levels, so the material will soon be forgotten.

While studying a test about heart function, McNamara et al. (1996) documented an intriguing interaction among readers' knowledge about a topic, coherence of the textbase, and the level of representation that was being tapped in a test. The readers varied in the amount of prior knowledge they had about the topic covered in the text. In the study, half of the readers read a text with a coherent textbase. That is, clauses were linked by appropriate connectives (e.g., therefore, so, and), and the topic sentences, headings, and subheadings were inserted at appropriate locations. The other texts had low coherence due to violations in the insertion of connectives, topic sentences, headings, and subheadings. The tests tapped either the textbase level of representation (which included recall tests) or the situation level (which included tests of inferences and answers to deep-reasoning questions).

The results of the McNamara study were not particularly surprising for low-knowledge readers. For these readers, texts with high coherence consistently produced higher performance scores than texts with low coherence. The results were more complex for the readers with a high amount of prior knowledge about the heart. A coherent textbase slightly enhanced recall, but actually lowered performance on tasks that tapped the situation model. The gaps, or breaks in temporality, spatiality, and causality, in text coherence forced the high-knowledge reader to draw inferences, construct rich elaborations, and compensate by allocating more processing effort to the situation model. In essence, deep comprehension was a positive compensatory result of coherence gaps at the shallow levels of representation.

Comprehension Calibration

One counterintuitive finding in comprehension research is that most children and adult readers have a poor ability to calibrate the success of their comprehension. Comprehension calibration can be measured by asking readers to rate how well they comprehend a text, and then correlating such ratings with comprehension scores on objective tests. These correlations are always either low or modest (r =.2 to.4), which suggests that college students have disappointing comprehension calibration. Another method of calibrating comprehension is to plant contradictions in a text and observe whether readers detect them. Such contradictions are not detected by a surprising number of adult readers. Instead, there is a strong tendency for readers to have an illusion of comprehension by adjusting their expectations at handling the surface code and textbase. Readers need to be trained to adjust their metacognitive expectations and strategies to focus on the deeper levels.

Classroom discourse is too often skewed to the shallow rather than the deep end of the comprehension continuum. Teachers typically follow a curriculum script that covers definitions, facts, concepts, attributes of concepts, and examples. This content is at the lower levels of Benjamin Bloom's taxonomy of cognitive objectives. Teachers rarely attempt to encourage Bloom's higher levels of inference, synthesis, integration, and the application of knowledge to practical problems.

Discourse Mechanisms that Promote Deep Comprehension

There are some methods of improving deep comprehension and learning by invoking discourse processing mechanisms, including: (1) constructing explanations, (2) asking questions, (3) challenging a learner's beliefs and knowledge, and (4) tutoring.

Constructing explanations. Good comprehenders generate explanations as they read text or listen to lectures. These explanations trace the causes and consequences of events, the plans and goals of agents (humans, animals or organizations), and the logical derivations of assertions. The questions that drive explanations are why, how, what-if, and what-if-not questions. For example, a deep comprehender might implicitly ask the following questions while reading the cylinder lock text: Why would the person turn the key to the right? How does the bolt move back? What causes the cam to rotate? What if the pins don't rise? Students learn much more when they construct these explanations on their own (selfe-xplanations) than when they merely read or listen to explanations.

Asking questions. Students should be encouraged to ask and answer deep-reasoning questions to help them construct explanations. Unfortunately, students are not in the habit of asking many questions, and most of their questions are shallow. A typical student asks only .11 questions per hour in a classroom, and less than 10 percent of student questions involve deep reasoning. When students are trained how to ask good questions while reading or listening to lectures, their comprehension scores increase on objective tests. Teachers rarely ask deep-reasoning questions in classroom settings, so it would be prudent to improve the questioning skills of teachers. Challenging the learner's beliefs and knowledge. One of the easiest ways to get students to ask questions is to challenge one of their entrenched beliefs, and thereby put them in cognitive disequilibrium. Suppose, for example, that a teacher expresses the claim that overpopulation is not a significant problem to worry about. This will normally stimulate a large number of student questions and counterarguments. Research on question asking has revealed that genuine information-seeking questions are inspired by contradictions, anomalies, incompatibilities, obstacles to goals, salient contrasts, uncertainty, and obvious gaps in knowledge. Therefore, one secret to eliciting student questions is to create cognitive disequilibrium and then provide useful information when students ask questions.

Tutoring. One-to-one human tutoring is superior to normal learning experiences in traditional classroom environments. This advantage cannot entirely be attributed to the possibility that tutors are more accomplished pedagogical experts than teachers. Peers often do an excellent job serving as tutors. Normal tutors rarely implement sophisticated pedagogical strategies, such as the Socratic method, building on prerequisites, error diagnosis and repair, or modeling-scaffolding-fading. It is the discourse patterns in normal tutoring that explain much of the advantages of tutoring over the classroom. The discourse in tutoring emphasizes collaborative problem solving, question asking and answering, and explanation building in the context of specific problems, cases, and examples. There is a turn-by-turn collaborative exchange (speakers take turns talking) in tutoring that would be impractical to implement in the classroom.

In summary, research in discourse processing can help solve some of the pressing challenges in education. Discourse plays an important role in helping the learner shift from shallow to deep comprehension, and from being a fact collector to being an inquisitive explainer.