Leo G. M. Noordman (1993) A More Parsimonious Version of Minimalism in Inferences. Psycoloquy: 4(08) Reading Inference (9)

Volume: 4 (next, prev) Issue: 08 (next, prev) Article: 9 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 4(08): A More Parsimonious Version of Minimalism in Inferences

Reply to Garnham on Noordam on Reading-Inference

Leo G. M. Noordman
Tilburg University
P.O.Box 90153
NL-5000 LE Tilburg

Wietske Vonk
Max Planck Institute for Psycholinguistics
University of Nijmegen
P.O.Box 310
NL-6500 AH Nijmegen

Noordman@kub.nl Vonk@mpi.nl


We will argue for an even more parsimonious minimalist theory of inferences than the one proposed by McKoon & Ratcliff (1992). At the same time we would like to indicate some points that are unclear with respect to the criteria for minimalism as well as for constructionism.


constructionism, inference, mental models, minimalism, reading, text comprehension.
1. In his target article, Garnham (1992) argues that the contrast between the minimalist and constructionist accounts of inferences, as proposed by McKoon & Ratcliff (1992) (M&R), is false because constructionist processes play a role in minimalist inferences. Garnham argues that inferences required to establish local coherence (accordingly called minimalist inferences) may require constructive processes, in the sense that they require the activation of background knowledge that is not readily available.

2. We would like to argue for a minimalist theory of inferences, but for an even more parsimonious version than the one proposed by M&R. At the same time we would like to indicate some points that are unclear with respect to the criteria for minimalism as well as for constructionism.

3. It is not quite clear whether M&R and Garnham use the same notion of constructionism. Garnham considers the use of background knowledge evidence for constructive processes. He is then right that constructive processes play a role in local coherence inferences. For M&R, constructionism is the generation of a representation of the situation described by the text, including the necessary inferences. A precise definition of constructionism is not given; it is probably most concrete to consider it the counterpart of minimalism.

4. This brings us to the first issue: lack of clarity with respect to the notion of minimalism. The problem is that M&R propose two different criteria for minimalism: "inferences of only two kinds are constructed: those that establish locally coherent representations... and those that rely on information that is quickly and easily available." In some passages (e.g., on page 441) it seems as if the two criteria reduce to one: availability of information. "Local coherence is defined for those propositions of a text that are in working memory at the same time." Inferences that establish local coherence are said to be based on information in short term memory, explicitly stated information or general knowledge. But, as a whole, M&R's article suggests that there are two different criteria for minimalism.

5. One consequence of the fact that local coherence serves as a criterion for minimalism is that inferences that establish local coherence on the basis of global relations in a text are said to be minimal. It is somewhat curious to call these inferences minimalist, because they require information that is not easily available and they are hence in conflict with the second criterion. There is no principled objection to a dual criterion, but as long as it is not clear what the commonality is between availability and local coherence, the minimalist theory is not very coherent. This is in fact the basis of the problem signaled by Garnham when he says that many inferences that do establish local coherence are constructive because they require background knowledge not readily available.

6. We would like to propose a theory of minimal inferences, defining minimal inferences as those that can be made on the basis of readily available knowledge. Our claim is that these inferences are made during reading. Inferences that establish local coherence are not claimed to be minimal because experimental results indicate that these inferences are not always made during reading. Examples are causal inferences such as:

        Chlorine compounds are frequently used as propellants
        because they do not react with other substances.

in a text on spray cans (Noordman, Vonk & Kempff, 1992, Experiment 2). The conjunction "because" signals that the relation between "propellants" and "not reacting with other substances" should be established. This conjunction suggests to the reader to infer that propellants should not react with the material in the spray can. In that sense the inference establishes local coherence, since it relates the two clauses to each other. The texts were chosen so that readers did not have the background knowledge underlying the inferences. The reading times for the "because" sentences under two conditions (the information relevant for the inference was or was not made explicit in a previous sentence) showed that readers did not make the inferences during reading. Verification data showed that readers made the inference when they had to verify the information after reading the text. In another experiment (Vonk & Noordman, 1990, Experiment 5), similar kinds of inferences were investigated in text dealing with familiar topics. Both the reading times and the verification times indicated that in this case the inferences were made during reading. The results suggested that the reader's knowledge controls the inferences during reading.

7. To investigate this more systematically, a paradigm was used in which experts and novices in the field of economics had to read texts in that domain (Noordman & Vonk, 1992). First the knowledge structures of experts and novices in the domain of economics were investigated using a number of elicitation experiments. On the basis of these experiments triplets of concepts were selected that were causally related for experts but not for novices, although the concepts were familiar to the novices. On the basis of these, triples sentences such as the following were constructed :

        American exports have been suffering a decline in the last
        few months because rising inflation has produced a harmful
        effect on the competitive position of the U.S.A.

These sentences required the causal inference: deterioration in competitive position leads to decline in exports.

8. The comparison of the reading times for the "because" sentences under two conditions (information underlying the inference either was or was not made explicit in the preceding text) indicated that experts did make the inferences during reading but novices did not. The results in this experiment differentiated between the two aspects of the minimalist theory of M&R. Inferences that establish local coherence need not be made during reading: although novices can derive the inferences on the basis of the information in the sentence, they do not actually make them. The results support the minimalist prediction that availability of knowledge controls the inferences during reading and falsifies the other minimalist prediction that inferences necessary to establish local coherence are always made during reading. Consequently, minimalism in a more parsimonious version than the one proposed by M&R is supported.

9. The causal inferences discussed so far are characterized as the derivation of a proposition that relates two concepts to each other, for example, "propellants" and "not reacting with other substances." The causal conjunction is then considered a signal to activate this relation. One might argue, however, that understanding the causal conjunction consists simply in accepting that there is a causal relation, without deriving the underlying proposition that accounts for the coherence between the sentences. This distinction is made by Noordman & Vonk (1992) as well as by Garnham. If one defines a causal inference as "accepting that there is a causal relation," then one could say that both experts and novices make the inference. The results are then in agreement both with the minimalist claim that inferences establishing local coherence are made and the minimalist claim that inferences based on available knowledge are made, since both groups of readers at least have knowledge about the meaning of the conjunction "because." In our opinion, however, this superficial way of interpreting the conjunction "because" cannot be considered an inference. So the results support a minimalist theory of inferences, with minimalism defined in terms of available knowledge (where the inferences come from) and not in terms of local coherence (where the inferences lead to).

10. We would like to make two additional points. The first concerns the question of how "dichotomous" inference making is. Both M&R and Garnham claim that making an inference is not a simple yes/no matter. Inferences can be made partial (McKoon & Ratcliff, 1986); they may be implicitly encoded (Garnham, 1992). That inference making is not a simple, dichotomous notion is also clear from some of our own studies that point to a distinction between the temporary activation of an inferred proposition and the encoding of the proposition (Noordman & Vonk, 1992; Simons, 1993). In the expert/novice experiments discussed earlier, the timing data for experts indicated that they made the inferences during reading. However, we also found a difference in the verification task that was administered after the reading task. The interpretation was that experts do activate the inferred proposition, but the proposition is not stored in the same way as information that is explicit in the text. Additional support for this claim was found in a subsequent experiment: the inferred information was more available for the experts when the it was verified immediately after reading the target sentence rather than after reading the whole text.

11. The last point concerns the notion of automaticity that M&R use in the minimalist hypothesis: "According to [the minimalist] hypothesis, the only inferences that are encoded automatically... are those that are based on easily available information... and those that are required to make statements in the text locally coherent." The notion of automaticity is confusing. Two different questions are actually involved in the discussion about inferences. The first concerns the control of inferences: what factors determine whether inferences are made during (nonstrategic) reading? This is the question addressed by M&R. The other concerns whether inferences are made automatically. We presume that this is the distinction made by Garnham in his section 6.5. That the two questions are different is also clear from our own studies. We have pointed out that the experts made the inferences spontaneously during reading because they had the relevant knowledge. In this sense inferences are controlled by the reader's knowledge. But one cannot conclude that the inferences are made because they do not require time for experts who have that knowledge available. On the contrary, although the knowledge was available, the inferences did require extra time. In that sense, the inferences were not made automatically.


Garnham, A. (1992) Minimalism versus constructionism: A false dichotomy in theories of inference during reading. PSYCOLOQUY 3(63) reading-inference-1.1

McKoon, G. & Ratcliff, R. (1986) Inferences about predictable events. Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 82-91.

McKoon, G. & Ratcliff, R. (1992) Inference during reading. Psychological Review, 99, 440-466.

Noordman, L.G.M. & Vonk, W. (1992) Readers' knowledge and the control of inferences in reading. Language and Cognitive Processes, 7, 373-391.

Noordman, L.G.M., Vonk, W. & Kempff, H.J. (1992) Causal inferences during the reading of expository texts. Journal of Memory and Language, 31, 573-590.

Simons, W.H.G. (1993) De regulering van inferenties door kennis van de lezer. PhD. Thesis Nijmegen University.

Vonk, W. & Noordman, L.G.M. (1990) On the control of inferences in text understanding. In D.A. Balota, G.B. Flores d'Arcais & K. Rayner (Eds.), Comprehension Processes in Reading. Hillsdale: Lawrence Erlbaum.

Volume: 4 (next, prev) Issue: 08 (next, prev) Article: 9 (next prev first) Alternate versions: ASCII Summary