Learning From Text Research Paper

1. Definitions

How do we define learning from text? Learning from text is often equated with learning of text, that is, with memory for the text information proper (e.g., the memorization of a poem). In contrast, Kintsch makes a clear distinction and defines learning from text as ‘the ability to use the information acquired from the text productively in novel environments’ (Kintsch 1998, p. 290).

Using concepts from psycholinguistic theories of text comprehension, these definitions can be further sharpened: van Dijk and Kintsch (1983) distinguish three levels of representation. First, the surface level encodes the verbatim wording of the text information. On this level, the syntactic and lexical realization of the text is represented. The second level is a semantic representation, called the textbase, closely corresponding to the propositional, or meaning-based content, but independent of the exact wording. In addition, the textbase contains some information not explicitly mentioned in the text, such as associative elaborations, certain inferences, and macropropositions. Finally, and most importantly, the situation model is an integration of the reader’s prior knowledge and experiences with the information currently encountered. This representation of ‘what the text is all about’ does not need to be in a verbal or propositional format, and it depends both on the text information, and on the individual background the comprehender contributes.

These three levels of representations are differentiated not only by their content, but also by their retention characteristics: The surface level is forgotten quickly, the textbase might be available over several hours, but the situation model is what survives in the long term. Thus, learning of text is located at the surface and textbase levels, whereas learning from text needs to be conceptualized as the construction, retention and utilization of a situation model representation (Ferstl and Kintsch 1999).

2. Empirical Methodology

For dissociating memory for text and learning from text, we need measures that tap these levels of representation differentially. Traditional measures for the textbase are recall and recognition, as well as the factual comprehension questions, that is, questions about information explicitly mentioned at a particular point in the text. Moreover, those inference questions that probe information on presumably automatic inferences (e.g., bridging inferences) are also located on the textbase level.

In contrast, for assessing the situation model, tasks requiring the ‘productive use’ of the acquired information in ‘novel environments’ are needed. When a manual is read, for instance, we can directly test the nonverbal ability to operate the device explained. For testing learning from a physics textbook we can ask students to perform problem-solving tasks requiring the transfer or the application of text information. However, even when the text information cannot be tested nonverbally, we can devise ‘novel situations’ going beyond the explicit text information. From recall protocols, we can score those statements that were not mentioned in the text, but were inferred or elaborated by the learner. An evaluation of their plausibility within the ideal situation model provides a score of the creative, reconstructive use of the information. In recognition tasks we can include words or statements that are consistent with the situation described, although they were not directly included in the text. In designing questions, the situation model is probed when a combination of prior knowledge with text information or the integration of statements from different parts of the text is needed.

Another approach attempts empirically to describe learning from text as the change of the reader’s prior knowledge structure (Ferstl and Kintsch 1999). After identifying key concepts of the text’s domain, each learner’s internal organization of these concepts is estimated based on relatedness ratings, results from sorting tasks, or cued associations. The resulting associative network is considered an approximation of the learner’s knowledge structure. After reading the text, the same task is repeated, and the resulting network is considered an approximation of the learner’s situation model. The changes from before to after reading are taken as an index of learning from text. Moreover, the knowledge networks after reading can be compared with an ideal structure, obtained, for instance, from an analysis of the text content, or from an expert’s assessment of the same domain.

Many inconsistencies in the literature on learning from text might be explained by evaluating in more detail the methods used for assessing learning. However, it is important to note that, although there are empirical dissociations, in most cases building a good textbase representation goes hand in hand with establishing an elaborate situation model. Therefore, in describing learning from text, the vast knowledge about text memory accumulated to date should not be neglected.

3. Text Characteristics

An obvious answer to the question of how we can improve learning from text is to just write better texts. Consequently, there have been numerous suggestions for writers on how to increase the readability of their texts (e.g., Taylor and Taylor 1983), most of which seem to consist of simplifying the style. On the word level, short and familiar words are recommended, and on the sentence level, syntactically complex constructions are to be avoided.

On the text level, the notion of coherence has been shown to be crucial for readability, that is, for comprehensibility. Local coherence refers to the connection between subsequent sentences, while global coherence refers to the connection between parts of the text, such as paragraphs. For instance, when moving back from a concrete example to the general topic of the text, global coherence is preserved, while local coherence might be violated. Coherence facilitates comprehension because no additional inferences are required for deriving the text structure. Situation model construction can be facilitated by adding explicit cohesive ties, that is, lexical information signaling the type of connection between successive sentences.

A second important factor is the macrostructure of a text, corresponding roughly to the outline or a summary of the text. The macrostructure has to be constructed by using macrorules, such as marking some propositions as particularly important or subsuming others under a more general one. Writers can facilitate the application of macrorules, for instance, by using a canonical, expected structure. The macrostructure of an existing text can be clarified by adding explicit information in the form of subheaders, introductory or structuring statements (e.g., ‘the following three examples … ’), or underlining of important information. Using these macrostructure cues helps to set up the textbase, and in turn frees resources for processing on a situation model level (for an early approach to text structure, see Meyer 1975).

4. Adjunct Information

Another way of improving learning from text is to provide additional materials that extend the text information. Very influential suggestions, in particular in educational research, are advance organizers and illustrations.

Outlines or summaries given before reading explicitly provide the macrostructure of the text to be read. Similar to headers or explicit organizing statements within the text, this additional information facilitates comprehension. An advance organizer, as proposed by Ausubel (1968), goes beyond a simple outline or summary. Rather, it uses a different level of abstraction and has the function to ‘provide ideational scaffolding for the stable incorporation and retention of the more detailed and differentiated material that follows in the learning passage’ (Ausubel 1968, p. 148). Although this concept seems compelling, there was considerable debate about the efficacy of advance organizers. While some studies showed a beneficial effect, others failed to replicate. One reason for these inconsistencies was that early studies did not have available the tools for analyzing the content and structure of both text and advance organizer, and that the levels of representation were not carefully distinguished. As an example of a more recent study controlling these factors, Mannes and Kintsch (1987) studied the interaction between organizer structure and representational level. When the organizer and the text were of a similar structure, readers formed a better textbase representation, because the advance organizer provided the macrostructure for the text read later. Different structures, on the other hand, improved the verification performance on situation model statements. In this case, the macrostructure of the target text had to be derived and the availability of the information from the advance organizer made a more elaborate situation model likely.

One of the most intriguing demonstrations of the effect of illustrations on text comprehension is the study by Bransford and Johnson (1972). They demonstrated that looking at a picture showing an unusual serenade substantially improved comprehension and recall of a vaguely worded text describing the scene on the picture. Since that early study, numerous experiments attempted to delineate the possibilities and limits of illustrations for improving learning processes (Mandl and Levin 1989). The kind of picture used depends on the text information to be illustrated. For instance, a narrative can be embellished by a portrait of the protagonist, a manual might contain a schematic depiction of the device, and the description of a hiking trail is facilitated by a map. Illustrations aid learning from text when they are available before or during processing, and when the information provided by the picture is relevant to and redundant with text information. Purely decorative illustrations do not have a beneficial effect. One theory for explaining the facilitative effect of illustrations was proposed by Paivio (1986), who postulated that dual coding of the information, in a verbal and in a visual representation, increases the likelihood of the information becoming available through one of the codes. However, illustrations tend to have a smaller effect on recall of text, but often facilitate situation-based inferences and elaborations. Rather than merely doubling the amount of information, as the dual coding theory suggests, the illustrations provide background knowledge, and they encourage the mapping of the text information to the visually presented information. Thus, illustrations and other nonverbal sources, such as the demonstration of a procedure, are utilized for constructing the situation model.

5. The Learning Process

As we have seen, the external information not only enhances the reader’s background knowledge, but also has a direct impact on the type of processing. How a text is being read determines the kind of representation being built up. More careful, effortful reading, with more time spent on the task, increases performance. In addition, there is no doubt that instructions can be used to induce effectively certain reading goals. For instance, when told to memorize a text for later recall, readers focus more on the construction of a good textbase representation than when told to read for comprehension. When instructed to attend to a certain type of information, encoding takes place selectively and irrelevant information gets neglected. When provided with an illustration, the reader processes the text information overlapping with the picture more thoroughly.

A more direct way to influence processing is to ask the right questions (Graesser et al. 1994). Questions are asked by the learner to clarify points that are not readily comprehensible. Questions are asked by the teacher to test comprehension, or lead students’ processing into a specific direction. When the effect of this type of question is to be evaluated, the questions are interspersed in the text or added afterwards. These so-called adjunct questions have two functions. First, they can induce comprehension monitoring, an Activity that readers often neglect. An important strategy during learning from text is to amend comprehension failures, for example, by rereading passages, or by activating additional background knowledge. Difficulties with answering an adjunct question, in particular, if feedback is provided, can signal problems that might otherwise go unnoticed.

The second function of adjunct questions is to support directly situation model construction. Questions can be formulated so that they point to particularly relevant factual information (e.g., definitions), or so that they promote inferences needed for deep understanding. Most studies document a benefit of adjunct questions mainly for the information probed, but little transfer to other information in the text. When the questions are not provided, but are to be generated, learners often face considerable difficulties. Without extensive training, effective questioning is more likely to be the result of an already established situation model, rather than an aid to constructing it.

6. Individual Differences

In his account of meaningful learning—which encompasses learning from text—Ausubel (1968) stresses the role of the learner. ‘Meaningful learning presupposes that the learner manifest a meaningful learning set, that is, a disposition to relate the new material nonarbitrarily and substantively to his cognitive structure’ (Ausubel 1968, p. 37). In this section we look at research on individual differences, and consider some specific factors influencing whether a ‘meaningful learning set’ is available.

Learning studies encompass groups as varied as children just barely able to read, high-school students, and adults. Hence it is important to take into account the learner’s age. One important finding is that younger children tend to rely more on verbatim encoding of text, and that strategies needed for situation model construction, such as structure building, analogical reasoning, or active integration with previous knowledge, become available only at a later stage during development.

Studies grouping adult learners according to their reading skills find that it is exactly those strategies that less skilled readers tend to use for compensating their weaknesses on the textbase level. For instance, poor readers rely more on contextual constraints and causal structures. Poor readers have lower recall scores, and benefit from good writing, but these textbase differences are less noticeable on the situation model level—at least as long as the situation model of the particular text is consistent with expectations based on general world knowledge.

Another learner characteristic concerns motivational factors, such as the interest the learner brings to the task at hand. Within a text, interesting information is more likely to be recalled, but embellishing a text with interesting tidbits does not necessarily improve comprehension for the remainder of the text. Thus, few ‘seductive’ details do not increase the overall motivation to become more involved in text processing. Because it is unavoidable that some content areas are not inherently interesting for each student, strategies for increasing the learners’ motivation are important (Wade 1992).

One of the most crucial individual differences is the learner’s background knowledge. In the serenade example, for instance, the illustration provided the necessary background knowledge for constructing a situation model, and in this case, even for constructing a coherent text representation. Some advance organizers also have the function of providing necessary background knowledge. Often, however, it does not suffice to look at a picture or read a second text. Instead, the reader needs more extensive domain knowledge. Experts have better vocabulary, they know more facts, but most importantly, these facts are closely interrelated and well structured. When readers are experts in the text’s domain, they recall more text information, they can distinguish important information from details, they can bridge coherence gaps, and they are more likely to derive goal structures.

The disadvantage caused by not being able to relate the current information to already existing cognitive structures (Ausubel 1968) or to ‘hang it on the right hooks’ in general world knowledge (Kintsch 1998) can be overcome by the better writing. As we have seen before, explicit coherence cues facilitate understanding. And indeed, when learners with high or low knowledge read an explicitly coherent text, their problem-solving performance and card-sorting results are comparable. Only for less coherent text do these situation model measures show a clear benefit for the high knowledge participants (McNamara et al. 1996). A particularly interesting result in the same study was an interaction between knowledge and coherence. High-knowledge readers developed better situation models when they read the less coherent text than when they read the better, more explicit text. An explanation for this somewhat counterintuitive result is that the coherence gaps elicited more active processing. Each coherence gap asks an adjunct question, ‘how does this fit together?’ The high-knowledge readers generated the necessary inferences on their own, but they needed some processing effort for doing so, which in turn helped them in developing deep understanding (for replications see Kintsch (1998)). Thus, just by providing the right text for the right learner, the learning success was increased for both high- and low-knowledge groups.

7. Teaching Strategies

An important driving factor of this area of research is the possibility of developing teaching programs incorporating the educational and psycholinguistic findings. The facilitating factors mentioned before can be provided by the teacher or within the text. Alternatively, the use of any comprehension strategy can be taught to students. For instance, underlining helps the reader distinguish important information from details, summarization puts a focus on the macrostructure, providing background knowledge is helpful, and prior activation of a relevant domain can foster integration of text information with available background knowledge. Some programs incorporate several of these techniques. For instance, the well-known PQRST technique recommends an elaborate sequence of processing: preview, question, review, state, and test (Robinson 1970). Of course, such a technique can only succeed when the learners are able to take advantage of it. As we have seen before, though, it is a skill to ask the appropriate questions for facilitating deep processing. On the other hand, explicitly teaching learning techniques provides a benefit to the learners only if they do not yet employ self-initiated processing strategies. Therefore, empirical confirmations of the efficacy of such programs, that almost everybody would intuitively agree upon, are difficult to obtain. Not just the right text for a given learner, but also just the right teaching technique for that learner is what is needed.

8. Conclusions And Outlook

While it may seem that empirical studies too often produce the answer ‘it depends’ to the question of how we can improve learning from text, this should not discourage further experimental research. We now have the theoretical background for analyzing text characteristics and for assessing learner-centered factors. In addition, there are some promising attempts to automate these (Kintsch 1998). Of course, sometimes, the intuitions of good educators are worth more than a multitude of controlled studies. Nevertheless, we should strive to objectify and delineate these intuitions, in order to support the learning process of many student generations to come.

The way printed media have shaped our intellectual world for centuries accounts for the exceptional interest in the issue of learning from text. This influence of books is gradually changing with the advent of new electronic media. Traditional textbooks or newspapers may soon be a relict of the past, but research on learning from text will become even more important. The large majority of information presented will remain verbal. Existing research on the interactions between text information and other, mostly visual, presentation forms can help to understand and improve modern communication. The usual way of writing texts as coherent wholes that are best read in their sequential order has also changed with the advent of technologies such as the Internet and other hypertext media. Using these tools provides the freedom to select, review, or enhance the learning material on a step-by-step basis. Thus, the learners’ active role in shaping their individual learning environment will become more and more important.

Bibliography:

1. Ausubel D P 1968 Educational Psychology: A Cognitive View. Holt, New York
2. Bransford J D, Johnson M K 1972 Contextual prerequisites for understanding: Some investigations of comprehension and recall. Journal of Verbal Learning and Verbal Behavior 11: 717–26
3. Ferstl E C, Kintsch W 1999 Learning from text: Structural knowledge assessment in the study of discourse comprehension. In: van Oostendorp H, Goldman S R (eds.) The Construction of Mental Representations During Reading. Erlbaum, Mahwah, NJ, pp. 247–77
4. Graesser A, McMahen C L, Johnson B K 1994 Question asking and answering. In: Gernsbacher M A (ed.) Handbook of Psycholinguistics. Academic Press, San Diego, CA, pp. 517–38
5. Hynd C R, Stahl S A, Carr M, Glynn S M (eds.) 1998 Learning from Text Across Conceptual Domains. Erlbaum, Mahwah, NJ
6. Kintsch W 1998 Comprehension: A Paradigm for Cognition. Cambridge University Press, Cambridge, UK
7. Mandl H, Levin J R (eds.) 1989 Knowledge Acquisition from Text and Pictures. Elsevier, Amsterdam
8. Mannes S M, Kintsch W 1987 Knowledge organization and text organization. Cognition and Instruction 4: 91–115
9. McNamara D S, Kintsch E, Songer N B, Kintsch W 1996 Are good texts always better? Interactions of text coherence, background knowledge and levels of understanding in learning from text. Cognition and Instruction 14: 1–43
10. Meyer B J F 1975 The Organization of Prose and its Effects on Memory. Elsevier, Amsterdam
11. Paivio A 1986 Mental Representations: A Dual Coding Approach. Oxford University Press, New York
12. Robinson F B 1970 Effective Study, 4th edn. Harper and Row, New York
13. Taylor I, Taylor M M 1983 The Psychology of Reading. Academic Press, New York
14. van Dijk T A, Kintsch W 1983 Strategies of Discourse Comprehension. Academic Press, New York
15. Wade S E 1992 How interest affects learning from text. In: Renninger K A, Hidi S (eds.) The Role of Interest in Learning and Development. Erlbaum, Hillsdale, NJ, pp. 255–77