Glenberg & Mathew (1992) have defended the constructivist position and have raised several valuable arguments against minimalism. We identify additional problems with the minimalist hypothesis and provide additional support for the constructivist position.
1.1 The minimalist attack on constructivist theories of language comprehension (McKoon & Ratcliff, 1992) (M&R) has sparked a lively debate about inference generation in text comprehension. Glenberg & Mathew (1992) have defended the constructivist position and have raised several valuable arguments against minimalism. This commentary identifies additional problems with the minimalist hypothesis and provides additional support for the constructivist position.
1.2 The minimalist case rests upon two basic assumptions. First, it is possible to read without a goal. Second, speed of processing is the primary determinant of whether or not a process is automatic. These assumptions are incorrect. In this commentary we will address the first assumption, whereas the second assumption is addressed in our commentary on the Garnham (1992) target article (Zwaan & Graesser, 1993). We will argue that M&R cannot rule out the possibility that their subjects read with the goal to construct a very shallow representation (a minimal representation) because a shallow representation was sufficient for their experimental tasks. This is also one of the reasons why M&R have not adequately tested whether subjects make global inferences. We will also discuss an further reason for this conclusion. Our final point is that spatial models may be one of the easiest types of situation models to test, but that this does not make them the most relevant type of model in story comprehension. Our arguments further support the constructivist position and bolster the position taken by Glenberg & Mathew.
2.1 M&R assume that it is possible for readers (and in particular the subjects in their own experiments) to read without a specific goal. How plausible is this assumption? First of all, it should be pointed out that M&R do not present empirical data to support their assumption. We believe that subjects in reading experiments do have specific goals. Subjects usually have a concept of what their task is and their goal may be to perform optimally on the expected task. For example, subjects may expect to be tested on their memory for the text. In this case, most subjects would generate the goal to construct a representation that is as close to the text as possible. By adopting such a goal, they would be able to provide an accurate recall protocol or summary of the text. Schmalhofer and Glavanov (1986) have demonstrated that subjects form a strong propositional textbase and a relatively weak situation model when given prior information that a summarizing task will follow the reading task.
2.2 Subjects are frequently required to read dozens of texts in psychological experiments and are tested after each text. M&R's experiments follow such a format. These intervening tests presumably influence the goals of the subject. The subjects in M&R's Experiments 1 and 2 were tested with a recognition test word directly after the final sentence of a text. These subjects probably constructed relatively shallow representations of the texts because this was satisfactory for them to perform well on the simple tests. M&R themselves acknowledge that tests may induce reading goals. On p.459 of their paper, they argue that the evidence for situation models obtained by Morrow, Greenspan, and Bower (1987) may be the result of the subjects' adopting a strategy to perform well on Morrow et al.'s test questions. M&R provide no compelling reasons for assuming that subjects adopt test-induced strategies only in experiments that lend support to the constructivist view and not in their own experiments.
2.3 M&R cannot rule out the possibility that their subjects had the goal to read their texts at a shallow level, i.e., the exact wording and surface form. It is therefore not surprising that they found evidence of minimalist processing. We do not have evidence that M&R's subjects actually had a shallow reading goal. However, the onus of demonstrating that their subjects read without a specific goal is on M&R. On the face of it, their tasks lend themselves to shallow processing.
2.4 In assuming that their subjects read without a goal, M&R make their theory immune to criticism. In the absence of a method to determine the goals of subjects in experiments, M&R could dismiss every experiment that demonstrates that subjects construct situation models by claiming that these subjects must have been reading with a specific goal, just as they did with the Morrow et al. experiments.
2.5 M&R correctly point out that there is a lack of research into reading goals and strategies. Glenberg & Mathew (Part I, 1.5.4.) correctly point out that "reading for good comprehension requires mental model construction and is often goal directed and strategic." However, we think this position can be articulated more strongly: The amount and type of inferences readers generate are a function of the reader's goal (Graesser & Kreuz, in press; Zwaan & Van Oostendorp, in press). Therefore, knowledge of these goals is needed before we can conclude whether or not people construct situation models and, if they do, what kind of models.
2.6 We distinguish three general goals in spontaneous reading. First, readers may simply want to explore the stimulus. Second, they may want to be informed about new facts and occurrences. Third, they may want to be entertained. These goals constrain the types of inferences readers will generate. More specific goals (e.g., reading for aesthetic as opposed to hedonistic pleasure) may further constrain inference generation and the construction of situation models. Further research must develop a taxonomy of reading goals before conclusions about situation models can be drawn.
2.7 So, have M&R adequately tested global inference generation? We are not convinced that their Experiment 3 adequately tested whether people make global causal inferences. One reason is that there was no assurance subjects had the goal of reading for deep global comprehension. A second reason is that M&R do not appear to have conducted any pilot work on their test items to make sure a normal reader could have made the inferences they used as their test items. It is possible to collect think aloud protocols (Trabasso & Suh, in press) or gather question answering protocols (Graesser & Clark, 1985) in a norming study to verify that these inferences could be made. It is conceivable that M&R's subjects made global inferences other than the ones M&R were testing for. Hence M&R's conclusion that readers do not make global inferences when the text is locally coherent may be premature. In fact, recent evidence shows that people do make various global inferences even when a text is locally coherent (Albrecht, Hakala & O'Brien, 1992; Long, Golding & Graesser, 1992; Suh & Trabasso, in press; Van den Broek & Lorch, in press).
3.1 Glenberg & Mathew (and also Garnham, 1992) appear to work under the assumption that spatial models are either the most important or the only type of situation model people construct while reading a narrative text. Spatial models may be an easy type of model to conduct experiments on because the inferences are often clearcut (e.g., an object can either be between two other objects or not). However, this does not make spatial information the most important or relevant aspect of situation models constructed during naturalistic text comprehension. There is evidence that the construction of detailed spatial models is not an important goal of readers reading a naturalistic story (Zwaan & Van Oostendorp, in press). Instead, readers construct causal chains, goal structures, and emotions (Gernsbacher, Goldsmith & Robertson, 1992; Graesser & Kreuz, in press; Lehnert, 1981).
3.2 Spatial models are relatively difficult to construct from text. Texts are linearly organized. Whereas spatial information is not (primarily) linearly organized (Denis, 1991; Levelt, 1989), temporal, causal, goal-oriented, and emotion-based information is usually articulated in an ordering that directly corresponds to the chronological ordering of the episodes in the situation models. Hence models based on these types of information are often isomorphic to the textual surface structure, whereas spatial models usually are not. This is presumably one of the reasons spatial models are so popular in reading experiments -- it is relatively easy to assess whether people have constructed a spatial situation model as opposed to a textbase representation exclusively. However, the lack of isomorphy between text and model is also the reason it is intrinsically difficult to construct spatial models from text. Ohtsuka and Brewer (1992) have recently shown that people have problems processing temporal information when there is a lack of isomorphism between the surface sequence of episodes in the text and the underlying chronological sequence (for example, in the case of flash forwards).
3.3 Spatial information is surface information in everyday situations when people can directly perceive the environment; that is, most spatial relations do not have to be inferred. Goals, motives, intentions, and emotions, however, do have to be inferred. People spend 16 hours a day explaining actions and anomalous events in their lives. People may be better equipped to draw inferences about goals, actions, and emotions than about spatial relations during text comprehension because we do so during normal everyday interaction.
3.4 To summarize, more convincing empirical support for the constructivist view on text comprehension could be obtained by examining models that people are naturally inclined to construct during text comprehension. Causal, goal-oriented, and emotional information rather than spatial information may form the backbone of these models.
Albrecht, J.E., Hakala, C.E. & O'Brien, E.J. (1992) Updating a mental models: maintaining both local and global coherence. Poster presented at the 33rd Annual Meeting of The Psychonomic Society, November, 1992, St. Louis, MO
Denis, M. (1991) Image and cognition. New York: Harvester Wheatsheaf.
Garnham, A. (1992) Minimalism versus constructionism: A false dichotomy in theories of inference during reading. PSYCOLOQUY 3(63) reading-inference-1.1
Gernsbacher, M.A., Goldsmith, H.H. & Robertson, R. (1992) Do readers represent characters emotional states? Cognition and Emotion, 6, 89-111.
Glenberg, A. M. & Mathew, S. (1992) When minimalism is not enough: Mental models in reading. PSYCOLOQUY 3(64) reading-inference-2.1
Graesser, A.C. & Clark, L.F. (1985) Structures and procedures of implicit knowledge. Norwood, NJ: Ablex.
Graesser, A.C. & Kreuz, R.J. (in press) A theory of inference generation during reading. Discourse Processes.
Lehnert, W.G. (1981) Plot units and narrative summarization. Cognitive Science, 5, 293-331.
Levelt, W. (1989) Speaking: From intention to articulation. Cambridge, MA: MIT Press.
Long, D.L., Golding, J.M. & Graesser, A.C. (1992) A test of the on-line status of goal-related inferences. Journal of Memory and Language, 31, 634-647.
McKoon, G. & Ratcliff, R. (1992) Inferences during reading. Psychological Review, 99, 440-466.
Morrow, D.G., Greenspan, S.L. & Bower, G.H. (1987) Accessibility and situation models in narrative comprehension. Journal of Memory and Language, 26, 165-187.
Ohtsuka, K. & Brewer, W.F. (1992) Discourse organization in the comprehension of narrative order in texts. Discourse Processes, 15, 317-336.
Schmalhofer, F. & Glavanov, D. (1986) Three components of understanding a programmers manual: verbatim, propositional, and situational representations. Journal of Memory and Language, 25, 279-294.
Suh, S.Y. & Trabasso, T. (in press) Global inferences in on-line processing of texts: converging evidence from discourse analysis, talk-aloud protocols, and recognition priming. Journal of Memory and Language.
Trabasso, T. & Suh, S.Y. (in press) Understanding text: Achieving explanatory coherence through on-line inferences and mental operations in working memory. Discourse Processes.
Van den Broek, P. & Lorch, R.F., Jr.(in press) Network representation of causal relations in memory for narrative texts: Evidence from primed recognition. Discourse Processes.
Zwaan, R.A. & Graesser, A.C. (1993) There is no empirical evidence that some inferences are automatically or partially encoded in text comprehension. PSYCOLOQUY 4(4) reading-inference.5
Zwaan, R.A. & Van Oostendorp, H. (in press, a) Do people construct spatial models during naturalistic story comprehension? Discourse Processes.