The minimalist position offers a good summary of much, but not all, of the data on inferencing. It is, however, to be faulted for being virtually impossible to falsify. Although minimalism provides a good summary of the data that was available at the time of its formulation, it should be noted that many new studies suggest that occurrence of inferences may not be as minimal as indicated by initial data.
1.1 In the old days of gender stereotyping, there was a popular children's riddle that went like this. A man and his son were in a terrible auto accident. The man was killed instantly; the boy was rushed to the hospital for emergency surgery. Upon seeing the child, the surgeon exclaims, "Oh my God, this is my child." How can that be? The answer: the surgeon is the child's mother. People who struggle with the riddle seem to infer that the surgeon is a male; and since the child's father is dead, there seems to be no easy answer. The inference that the surgeon is male is not required for coherence, and in fact, interferes with one's ability to solve the riddle; nonetheless, many people in the old days made this inference.
1.2 Elaborative inferences such as these are thought to be a normal part of skilled reading. Indeed, reading comprehension tests often focus on the reader's ability to draw inferences. These tests assume that a competent reader must draw a variety of inferences, such as inferences about the consequences of actions, the author's point of view, the characters' moods and intentions. Not only is inferencing thought to characterize a skilled reader, but it is also thought to be a sign of general intelligence. Smart people go beyond the literal: they foresee the consequences of actions, they perceive motives, they understand puns, and so forth.
1.3 These informal notions of an intelligent language processor contrast sharply with much of the recent research in reading comprehension. The data show that many of the most obvious inferences do not seem to be drawn. For example, most instrument inference studies find that when college students read "he stirred his coffee," the concept spoon is no more active than some unrelated concept, such as moon. And when they read that "someone fell from the roof of a 14-story building," the concept dead is no more active than it is after reading a sentence that has nothing to do with injury or possible death.
1.4 I think it is this striking contrast between our intuitions about the reading process and the actual data that motivated McKoon and Ratcliff (1992) (hereafter, M&R) to propose their theory of minimalism. It is a call to researchers to revise our intuitions about reading so as to be more in line with the data. In other words, quit talking about all the inferences the skilled reader draws until the data are there to support such claims. At the same time, it is a call to perhaps revise the data base. As M&R say, minimalism is a research strategy: given the current data, let's just assume that only the most minimal inferences are drawn and then design studies that might disprove this notion. Perhaps with better inference assessment techniques, more interesting materials, or broader classes of inferences, we will come up with data that correspond more to our intuitions about inferencing. In these respects, I think the minimalist proposal can be useful, although, as I shall show below, M&R's particular formulation of it is not testable.
2.1 Garnham (1992) takes issue with the utility of M&R's proposal. First, he argues that minimalism is not a new position. He's right. But I think the striking contrast between the way we talk about inferencing and the data itself required the kind of reminder that M&R's minimalism theory provides.
2.2.1 Garnham also argues that in defining minimalism by contrasting it with constructionism, M&R have created a false dichotomy. In my view, there are a number of dimensions that can be used to contrast the two positions. The dichotomy can be made to seem false or true depending on the dimensions selected for contrast. Let me illustrate.
2.2.2 At least four different dimensions have been used to contrast minimalism with constructionism. They are: (1) quantity of inferences, (2) type of inferences, (3) type of representation, and (4) amount of background knowledge used. The positions can thus be characterized in the following way:
MINIMALISM CONSTRUCTIONISM
1. few inferences 1. many inferences
2. local inferences only 2. local plus global inferences
3. representation of the linguistic 3. representation of the situation
form of the text referred to by the text
4. little use of background knowledge 4. extensive use of background
knowledge
2.2.3 When one considers only a single dimension, it is hard to argue that the dichotomy between the positions is false. The way in which Garnham makes his case that the dichotomy is false is to mix dimensions. Inferencing is minimal in the sense of number and type of inferences (local); but it is constructive in terms of the nature of the representation (situational) and the generous use of background knowledge. I agree with this characterization. More important, I think M&R would too, because I think they use minimalism to refer only to the first two dimensions. What this shows is that in discussing theoretical positions, it is necessary to clarify the underlying assumptions.
3.0.1 Minimalism strikes me as OK as a loose summary of inference data or as a research strategy; but as a formal theory, it has serious problems. One is a lack of clear definitions of terms like "local coherence" and "easily available information." Another problem is that the theory cannot be falsified. A third problem is that in formulating the theory, M&R ignored data showing that some global inferences occur under conditions that would not typically be characterized as involving special strategies.
3.1 DEFINITIONAL PROBLEMS
3.1.1 The lack of a clear definition of local coherence is not unique to M&R. It is a problem that exists for the field, and it is a difficult one. However, if one is going to posit a theory that only two classes of inferences are drawn: (1) those necessary to establish local coherence, and (2) those based on easily available information, then it is incumbent upon the theorist to say what is meant by these terms. M&R state that because there is no formal definition of local coherence in the field, they will define it as "making sense" with the previous context. But by whose standards should it make sense? One need only examine Garnham's discussion of Noordman and Vonk (1992) to see how there can be varying degrees of sense or coherence between sentences. And M&R's discussion of the causal inferences shown to be drawn in Keenan, Baillet & Brown's (1984) study shows that M&R have not worked out one of the most basic questions about cohesion -- the extent to which it involves causal as well as referential connections.
3.1.2 The definition of inferences based on easily available information is equally problematic; it is circular. How do you know which inferences will be drawn? Answer: those for which information is easily available. How do you know which information is easily available? Answer: see which information is easily activated inferentially.
3.2.0 NO WAY TO FALSIFY THE THEORY
3.2.1 There seems to be no way to disprove this theory. In principle, one should be able to disprove it either by showing that inferences other than those predicted by the theory are in fact drawn or by showing that inferences said to be drawn by the theory are in fact not drawn. But the theory is formulated so that neither is possible. Let me illustrate.
3.2.2 If an inference is shown to be drawn, the theory of minimalism can account for it by claiming that either: (1) it was necessary for local coherence, or (2) it was based on easily available information, or (3) the subject adopted a special goal or strategy. Even the most global inferences based on the most remote information can be accounted for by resorting to the claim that subjects can adopt their own special strategies! So, there is no way to argue against minimalism with data showing subjects make a certain type of inference.
3.2.3 Similarly, any evidence showing that some type of inference is not drawn can easily be accommodated by M&R's theory. For example, one might think, as Garnham points out, that the evidence on instrument inferences would be damaging to the minimalist position because it shows that what seems like easily available information, e.g., that one uses a spoon to stir coffee, is not inferred. But because there is no a priori definition of easily available information, M&R can just claim that, intuitions to the contrary notwithstanding, spoon is not easily available in a sentence about stirring coffee. (How do they know that? Because spoon is not inferred.) The circularity of "easily available information" thus insulates the theory from any attempts to falsify it.
3.3.3 When I read an early draft of M&R's paper, I commented on the problems of falsifying the theory. M&R responded by adding to the final version a couple of statements regarding how they think the theory can in fact be falsified. They state, "The minimalist position would be contradicted if it could be shown that some inference was encoded even though it was neither quickly available nor necessary for local coherence" (p.445). But, as I have argued above, that is not true; because of the definitional problems with "easily available information" and "local coherence," M&R could always claim post hoc that some inference actually met these conditions, or, alternatively, they could argue for the use of special strategies. They also state, "The minimalist position would also be contradicted if it could be shown that there were kinds of quickly available information that did not support inferences" (p.445). What would that be? Wouldn't any information that was quickly available and not explicit in the text be an inference by definition?
3.3.4 One final comment on this topic concerns M&R's use of data to support the minimalist position. They review a number of studies showing that inferences are made and they show how these results support the minimalist theory. I do not disagree with their conclusion, but I wish to point out that these results also support the constructionist position, and that they are no more support for one view than the other.
3.4.0 IGNORED DATA
3.4.1 M&R carefully review a large number of inference studies that support the minimalist theory (and as I noted above, also support the constructionist position). What would have been more interesting would have been for them to discuss studies in the literature that seem to contradict their position. I'd be curious to know if M&R would consider these studies as evidence against their theory or if they would find the results consistent with the minimalist theory using one of the "theoretical outs" discussed above.
3.4.2 Some of the best-known studies that most people regard as evidence for nonminimalist inferences are: Bransford, Barclay, & Franks, 1972; Potts, 1974; Sanford & Garrod, 1981; and Suh & Trabasso, 1988 (they do claim that this study's inferences are due to local coherence breaks but they do not really explain). The inferences in these studies seem neither to be necessary for local coherence nor to be based on particularly easily available information (at least no more available than "spoon" with "stir the coffee"). An example from Sanford & Garrod serves to illustrate. They used the paragraph:
John was on his way to school.
The bus trundled slowly along the road.
Last week he had trouble controlling the class.
They found that readers tend to make the elaborative inference that John is a school boy, with the result that the last sentence is then difficult to comprehend. How does this result fit with the minimalist theory?
4.1 Despite our intuitions to the contrary, the minimalist position offers a good summary of much, but not all, of the data on inferencing. As a theory, however, it is to be faulted for being virtually impossible to falsify. Although minimalism provides a good summary of the data that were available at the time it was formulated, it should be noted that many new studies suggest that the occurrence of inferences may not be as minimal as the initial data indicated (e.g., Murray, Klin, & Myers, in press; Whitney, Ritchie, & Crane, in press).
Bransford, J.D., Barclay, J.R., & Franks, J.J. (1972). Sentence memory: A constructive versus interpretive approach. Cognitive Psychology, 3, 193-209.
Garnham, A. (1992). Minimalism versus constructionism: A false dichotomy in theories of inference during reading. PSYCOLOQUY, 3(63) reading-inference-1.1
Keenan, J.M., Baillet, S.D., & Brown, P. (1984). The effects of causal cohesion on comprehension and memory. Journal of Verbal Learning and Verbal Behavior, 23, 115-126.
McKoon, G. & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99, 440-466.
Murray, J.D., Klin, C.M., & Myers, J.L. (in press). Forward inferences in narrative text. Journal of Experimental Psychology: Learning, Memory and Cognition.
Noordman, L.G.M. & Vonk, W. (1992). Reader's knowledge and the control of inference in reading. Language and Cognitive Processes, 7, 373-391.
Potts, G.R. (1974). Storing and retrieving information about ordered relationships. Journal of Experimental Psychology, 103, 431-439.
Sanford, A.J., & Garrod, S.C. (1981). Understanding written language: Explorations in comprehension beyond the sentence. New York: Wiley.
Suh, S., & Trabasso, T. (1988). Convergent evidence on inferences during comprehension of text. Paper presented at meeting of the Psychonomics Society in Chicago; November, 1988.
Whitney, P., Ritchie, B.G., & Crane, R.S. (in press). The effect of foregrounding on readers' use of predictive inferences. Memory & Cognition.