Both Glenberg & Mathew's (1992) and Garnham (1992) make compelling contributions to the debate. Their arguments, however, seem incoherent of incomplete in a number of ways.
2.1. The minimalists' idea of the theory of mental models is slightly different from what most researchers assume as a standard point of view. Both Garnham (1992) and G&M indicate that automaticity is not an essential characteristic of the mental-models theory. What is important for the mental-models theory is (a) the idea of constructing a representation of the situation described in the text, not of the linguistic expressions used to describe the situation, (b) the incremental construction of a text representation, and (c) the updating process which takes place during reading.
2.2. The construction of a mental model is an incremental updating of the representation on the basis of the present and past input. So the resultant representation in any given moment guides the interpretation of subsequent input. A relevant question not yet answered in the mental-model framework concerns the time course of the integration of previous information with the interpretation of each new sentence (see Sanford & Garrod, 1989). However, the claim that mental models are built online differs from the claim that mental models are built automatically. Comprehension proceeds in a linear fashion and is very efficiently organized in real time. The formation of a mental model is an intentional process, hence not automatic, but it might still occur online. The distinction between automatic and nonautomatic should not be drawn only on the basis of how quickly an inference is generated, as is implicit in the minimalist hypothesis. Because speed alone is used as the criterion for automaticity, the mental-model theory is mischaracterized by the minimalist hypothesis. It is precisely the idea that automaticity and speed are synonymous in the minimalist hypothesis that demands that we know the speed at which we can consider a process as being automatic. It would be very hard to falsify the minimalist hypothesis if we could not specify how fast an inference had to be in order to be considered automatic.
3.1. Characterizing mental models as being both complete representations of the situations described by texts and as "life-like" representations is also misleading. The minimalist hypothesis seems to attack mental-model theory based on definitions of terms not entertained by most other researchers. Most researchers would agree that a representation of a text is not literally a "lifelike" representation. The imagery debate that took place in the 1970's distinctly discarded the notion that images are pictures-in-the-head. Although metaphors are sometimes misleading, G&M clearly indicate that the particular sense of the "lifelike" metaphor is a way to capture the characteristics of functional equivalence and analogical correspondence between the representation and the situation represented.
4.1. Garnham's arguments to demonstrate that the notion of a complete representation is incoherent seem very compelling to me. Processing also seems to be partial or incomplete, at least in some situations. The failure to notice anomalies (local inconsistencies) demonstrated with "the Moses Illusion" (Erickson & Mattson, 1981) and "the survivors effect" (Sanford, 1992) suggests that only a partial analysis of the input is carried out in some situations. These effects also indicate that global expectations -- the situation or mental model -- prevent anomaly detection at the level of local semantics. It seems difficult to explain these results within a minimalist approach which requires local semantic coherence to be established before the phrase is integrated into a larger unit of meaning.
5.1. Another important contribution of G&M's target article is its refutation of M&R's salience arguments. First, the presence of the association/dissociation effect for Glenberg, Meyer & Lindem's (1987) targets but the absence of the same effect for M&R's targets (after excluding six texts in which M&R's targets were not clearly dissociated from the main character) is problematic for the M&R account. Second, the logic and predictions of G&M's second experiment are clear, and the data are suggestive. The data are not clean enough to make very strong claims, however. The second wave of subjects did not replicate the effect observed for the first wave. Besides, the third wave of subjects showed quite a large negative effect for the M&R target, one not predicted by mental-model theory.
Erickson, T. A. & Mattson, M. E. (1981) From words to meaning: A semantic illusion. Journal of Verbal Learning and Verbal Behavior, 20, 540-552
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 comprehension. PSYCOLOQUY 3(64) reading-inference-2.1
Glenberg, A. M., Meyer, M. & Lindem, K. (1987) Mental models contribute to foregrounding during text comprehension. Journal of Memory and Language, 26, 69-83
McKoon, G. & Ratcliff, R. (1992) Inference during reading. Psychological Review, 99, 440-466
Sanford, A.J. (1992) Some mechanisms of coherence. Paper presented at the workshop on coherence in spontaneous discourse. Oregon
Sanford, A.J. & Garrod, S.C. (1989) What, When, and How?. Questions of Immediacy in anaphoric reference resolution. Language and Cognitive Processes, 4, 235-262