Murray Singer (1993) Minimalism: a Hedged Analysis of Restricted Inference Processing. Psycoloquy: 4(01) Reading Inference (2)

Volume: 4 (next, prev) Issue: 01 (next, prev) Article: 2 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 4(01): Minimalism: a Hedged Analysis of Restricted Inference Processing

Commentary on Garnham and on Glenberg & Mathew on Reading-Inference

Murray Singer
Department of Psychology
University of Manitoba
Winnipeg, Canada R3T 2N2


In their articles Garnham (1992) and Glenberg and Mathew (1992) agreed that a constructionist/mental-model approach provides a comprehensive framework for the analysis of discourse inference processes and that constructionist analysis does not entail "complete" global representations. The minimal inference hypothesis has several flaws and difficult to properly evaluate. Because of information processing constraints, a better theory of restricted inference processing is warranted.


constructionism, inference, mental models, minimalism, reading, text comprehension.


1. In their recent critiques of the minimal inference analysis of McKoon and Ratcliff (M&R, 1992), Garnham (1992) and Glenberg and Mathew (1992) agreed that a constructionist/mental-model approach provides a comprehensive framework for the analysis of discourse inference processes and that the constructionist analysis, contrary to M&R's proposal, does not entail "complete" global representations.

2. Garnham also argued that the focus on automaticity (see also Glenberg & Mathew) and the nature of the "availability" criterion of minimalism "hedged" the minimalist analysis. Further scrutiny reveals three additional principles that qualify the minimalist analysis. First, M&R's proposal (pp. 441, 444) that causal relations contribute to local coherence tacitly acknowledges that readers routinely construct causal situations. Second, the proposal that local INcoherence invokes global information (M&R: p. 445) has a similar impact. Third, M&R (p. 440) indicated that when readers adopt special reading strategies, some global inferences may be "nearly as easy as minimal inferences." Like the Availability hedge, this proposal prohibits the falsification of minimalism. The goal of this commentary is to elaborate upon all of these qualifications except "Availability."


3. M&R (p. 441) used the term "automatic" in its classic sense of referring to processes that are fast, free from conscious awareness, and low in their demands on cognitive resources (e.g., Schneider & Shiffrin, 1977). However, with regard to automaticity, the minimalist hypothesis involves possible circularities and paradoxes; and not all of the inference processes that are claimed to be automatic fit that category. These issues will each be considered.

4. Circularity: M&R's (p. 440) definition of minimal inference implies that the only inferences that automatically accompany comprehension are those that meet either the coherence or the availability criterion. However, "availability" does little more than restate the speed characteristic of automatic processes. It is as though M&R have proposed that the only inferences that are drawn automatically during reading are automatic ones.

5. The paradox of "automatic strategic" processes: Conversely, the meaning of minimal inference is obscured by M&R's treatment of strategic processes. Consider their statement, "In the absence of specific, goal-directed strategic processes, inferences of only two kinds are constructed" (M&R: p. 440). This implies that, in the PRESENCE of strategic processes, other types of inference will be automatically constructed. Insofar as automatic and strategic processing form a dichotomy, this proposal is paradoxical. Indeed, M&R proceeded to state that "some strategic inferences may be... nearly as easy as minimal inferences" (p. 440). Although this proposal might have been articulated as a production system representation of processing strategies (e.g., Anderson, 1983), M&R did not do so. What remains is a very blurred distinction between automatic and strategic processes.


6. In the framework of minimal inferences that preserve local coherence, M&R (pp. 443-445) discussed anaphoric resolution, as well as bridging inferences that specify the links among text ideas. A crucial question is whether these inference are automatic. M&R, in fact, expressed some doubt as whether BRIDGING INFERENCES are automatic. Indeed, local bridging inferences fail two crucial criteria of automatic processing, namely, they are both slow and resource-demanding (van Dijk & Kintsch, 1983, p. 45). This is probably because the reader engages in bridging only after a failed reinstatement search, itself a controlled process (Kintsch & van Dijk, 1978; Lesgold, Roth & Curtis, 1979).

7. The case of ANAPHORIC RESOLUTION is more complex. Of the evidence reviewed by M&R, some (e.g. Swinney & Osterhout, 1990) fits the minimal inference definition well. In contrast, the consistency of the data of Dell, McKoon, and Ratcliff (1983) with the minimalist hypothesis is debatable. This can be demonstrated by comparing pronoun resolution with thematic inferences about text. For example, the sentence, "The townspeople were amazed to find that all the buildings had collapsed except the mint" permits a thematic inference about an earthquake (Till, Mross & Kintsch, 1988). M&R denied that these thematic inferences are automatically encoded. They did so presumably because it was only for stimulus onset asynchronies of 1000 ms or more, relative to the preceding word, "mint," that these inferences showed significant facilitation (Till et al., Figure 5).

8. Dell et al., in contrast, reported that, relative to a control word, the anaphor "criminal" produced recognition facilitation both of its referent, "burglar," and of a companion co-proposition of the referent, "garage." Facilitation obtained in 250 ms, well within the time limit of automatic processing. This confirms that the concept "burglar" was readily available. However, Dell et al. (1983, Figure 1) also reported that it took 1250 ms for the facilitation of the referent, "burglar," to substantially exceed that of its companion, "garage." Only at this point can it be said that "burglar" has been established as the referent of "criminal." If a facilitation lag of 1000 ms indicates that Till et al.'s thematic inferences are controlled, then the 1250 ms lag indicates the same thing concerning pronoun resolution.

9. Doubt about the automaticity of anaphoric resolution is also raised by the findings of Corbett (1984). Reading time for the anaphoric noun phrase, "frozen vegetable," is faster when it has only one possible referent, "frozen asparagus," than when the text also includes the distractor referent, "canned corn." M&R invoked this finding as support for minimalist hypothesis because it reflects the immediate availability of potential referents. If anaphoric resolution were automatic, however, processing time should be independent of the number of distractors. Analogously, in their classic demonstration, Schneider and Shiffrin (1977) reported that the automatic process of detecting a digit among a series of frames of letters was independent of the number of letters on each frame.

10. Finally, even within M&R's local coherence separation limit of two sentences (M&R: p.441), it takes longer to resolve pronouns to less recent than more recent referents. For example, it takes longer to comprehend "It was selling very well" after reading "Yesterday I met a woman who had written a book on viruses; she had studied them for years and years" than after "Yesterday I met a woman who was an expert on viruses; she had written a book on the topic" (e.g., Lesgold et al., 1979). This indicates that M&R have overstated the automatic features of pronoun resolution.


11. According to M&R, local coherence may be established on the basis of causal relations (p. 441, p. 444, p. 455), although this assertion itself is hedged using terms such as "perhaps" and "potential." However, whether a causal representation is derived from connections among nearby or distant text ideas, it is clearly a high level representation of discourse meaning; that is, a situation model (e.g., Trabasso & van den Broek, 1985). The contribution of causal relations to local coherence proposed by M&R contradicts their denial that readers routinely construct situation models.


12. Although Garnham (Abstract; 4.2) argued that M&R overlooked the fact that local coherence depends on constructionist processes, M&R did propose that local incoherence may be repaired with reference to global information (M&R: p. 445). This is a plausible proposal, but it qualifies the minimal inference analysis.

13. First, it advances a "weak" minimal position. A stronger minimal inference view is that people who read in the "morning fog over breakfast" (M&R: p. 440) might well be insensitive to local incoherence and continue on their merry way when it is encountered (Singer, in press). This alternative is consistent with the poor quality of people's comprehension monitoring with reference to the surface, text base, and situation-model levels of text representation (Glenberg, Sanocki, Epstein & Morris, 1987).

14. On the other hand, M&R's proposal that local incoherence invokes global information is consistent with numerous experimental observations. For example, local coherence may be preserved by bridging inferences based on causal relations (Keenan, Baillet & Brown, 1984; Singer & Ferreira, 1983; Trabasso & Sperry, 1985); and by reference to global discourse functions, such as the presence of questions and their answers in a dialogue (Malt, 1985).

15. Although reasonable, the proposal that local incoherence invokes global information creates significant problems for the minimalist hypothesis. First, it includes a presupposition that is antithetical to the minimalist hypothesis, namely, that the reader is constructing a situation model. That is, it would be prohibitive for the reader to preserve coherence on the basis of causal connections, M&R's central realm of evaluating the minimalist hypothesis, if the causal situation had not been gradually developed. Second, Garnham (#5.2) pointed out that, in other circumstances, the global information needed to preserve local coherence is not available at all.

16. In conclusion, the proposal that local incoherence invokes global information advances a weak but perhaps defensible version of the minimal hypothesis. (Singer, in press, however, has reviewed evidence that is inconsistent even with weak minimalism). However, the cornerstone of M&R's treatment is that "readers do not automatically encode the inferences that would make up (a situation) model" (pp. 461-462). This theme is inconsistent with M&R's acknowledgement that readers detect causal connections for nearby ideas; and under local incoherence, even for distant ones. In addition, the possible failure to detect distant causal connections, the focus of M&R's new experiments, is not synonymous with failing to represent global structures. Therefore, possible minimalist characteristics of inference processing have no implications for the construction or nonconstruction of situation models. This is consistent with Garnham's and with Glenberg and Mathew's conclusions that the minimalist-constructionist distinction is a false dichotomy.


17. M&R (p. 440) stated that if the reader adopts "special strategies," inferences other than those strictly stipulated by the minimalist hypothesis might be automatically (in M&R's sense of the term) computed. This qualification has the same impact as Hedge 2, "Availability," rendering the minimalist hypothesis unfalsifiable. That is, it permits any detection of global inferences to be attributed to special processing strategies. In addition, it is likely that causal analysis in reading, a global function, is not a special strategy. In a test of the latter hypothesis (Singer, 1993), I instructed readers to adopt either a summarizing strategy, to promote causal processing (Einstein, McDaniel, Owen & Cote, 1990); to proofread; or to read normally. The subjects read causal sequences, such as "Anne POURED the bucket of water ON the bonfire, The fire went out"; and control temporal sequences, such as "Anne PLACED the bucket of water BESIDE the bonfire, The fire went out." After each experimental sequence, the subjects answered a question such as "Does water extinguish fire?" Answer time was faster in the causal than the temporal condition for both the summarizers and the normal readers, but not for the proofreaders. The causal advantage is attributed to the reader's validation of causal bridging inferences with reference to pertinent knowledge, such as WATER EXTINGUISHES FIRE (Singer, Halldorson, Lear & Andrusiak, 1992). The similarity of the summarizers' and normal readers' answer times discourages the notion that causal processing is a special strategy (see also Masson, in press, Experiment 6).


18. Because of information processing constraints, a theory of restricted inference processing is warranted. The minimal inference hypothesis, however, has several serious flaws. (1) The "automatic" inferences that it addresses are not automatic in the classic contrast of this term with "controlled" processing. (2) Restricted inference processing has no implications for the construction of text situation models (Garnham, 1992; Glenberg & Mathew, 1992). As M&R acknowledged, causal structure, a situational analysis, contributes both to local coherence and to the repair of local incoherence. Furthermore, it is likely that dynamic features of cognitive processing (Just & Carpenter, 1992) promote global effects. For example, the current state strategy (Fletcher & Bloom, 1988) controls the allocation of working memory resources in just such a way that highlights global causal relations. (3) The minimal hypothesis is so hedged as to make it difficult to properly evaluate.


Anderson, J. R. (1983) The architecture of cognition. Cambridge, MA: Harvard University Press.

Corbett, A. (1984) Prenominal adjectives and the disambiguation of anaphoric nouns. Journal of Verbal Learning and Verbal Behavior, 23, 683-695.

Dell, G. S., McKoon, G. & Ratcliff, R. (1983) The activation of antecedent information during the processing of anaphoric reference in reading. Journal of Verbal Learning and Verbal Behavior, 22, 121-132.

Einstein, G. O., McDaniel, M. A., Owen, P. D. & Cote, N. C. (1990) Encoding and recall of texts: The importance of material appropriate processing. Journal of Memory and Language, 29, 566-582.

Fletcher, C. R. & Bloom, C. P. (1988) Causal reasoning in the comprehension of simple narrative texts. Journal of Memory and Language, 27, 235-244.

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

Glenberg, A. M. & Mathew, S. (1992) When minimalism is not enough: Mental models in reading. PSYCOLOQUY 3(64) reading-inference-2.1

Glenberg, A. M., Sanocki, T., Epstein, W. & Morris, C. (1987) Enhancing calibration of comprehension. Journal of Experimental Psychology: General, 116, 119-136.

Just, M. A. & Carpenter, P. A. (1992) A capacity theory of comprehension: Individual differences in working memory. Psychological Review, 99, 122-149.

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.

Kintsch, W. & van Dijk, T. A. (1978) Toward a model of text comprehension and production. Psychological Review, 85, 363-394.

Lesgold, A. M., Roth, S. F. & Curtis, M. E. (1979) Foregrounding effects in discourse comprehension. Journal of Verbal Learning and Verbal Behavior, 18, 291-308.

Malt, B. C. (1985) The role of discourse structure in understanding anaphora. Journal of Memory and Language, 24, 271-289.

Masson, M. E. J. (in press) Episodically enhanced comprehension fluency. Canadian Journal of Experimental Psychology.

Schneider, W. & Shiffrin, R. M. (1977) Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review, 84, 1-66.

Singer, M. (1980) The role of case-filling inferences in the coherence of brief passages. Discourse Processes, 3,185-201.

Singer, M. (1993) Validation of motive bridging inferences in brief texts. Unpublished manuscript, University of Manitoba.

Singer, M. (in press) Discourse inference processes. In M. Gernsbacher (Ed.), Handbook of psycholinguistics. San Diego: Academic Press.

Singer, M. & Ferreira, F. (1983) Inferring consequences in story comprehension. Journal of Verbal Learning and Verbal Behavior, 22, 437-448.

Singer, M., Halldorson, M., Lear, J. C. & Andrusiak, P. (1992) Validation of causal bridging inferences. Journal of Memory and Language, 31, 507-524.

Swinney, D. A. & Osterhout, L., (1990) Inference generation during auditory language comprehension. In A. Graesser & G. Bower (Eds.), Inferences and text comprehension. San Diego: Academic Press.

Till, R. E., Mross, E. F. & Kintsch, W. (1988) Time course of priming for associate and inference words in a discourse context. Memory & Cognition, 16, 283-298.

Trabasso, T. & Sperry, L. L. (1985) Causal relatedness and importance of story events. Journal of Memory and Language, 24, 595-611.

Trabasso, T. & van den Broek, P. (1985) Causal thinking and the representation of narrative events. Journal of Memory and Language, 24, 612-630.

van Dijk, T. A. & Kintsch, W. (1983) Strategies of discourse comprehension. New York: Academic Press.


This research was supported by grant OGP9800 from the Natural Sciences and Engineering Research Council of Canada. Please address all further correspondence to Murray Singer at

Volume: 4 (next, prev) Issue: 01 (next, prev) Article: 2 (next prev first) Alternate versions: ASCII Summary