In this Reply I address four issues raised in the commentaries. First, that Glenberg & Mathew (1992) were "unfair" in one of their empirical criticisms of McKoon & Ratcliff's (1992) salience account. Second, a call for a consideration of nonspatial mental models. Third, I will use these ideas to elaborate on a distinction between inferences proper and information available from the representation that does not seem to have a direct counterpart in the text. Fourth, I will extend discussion of the "survivor effect." The ideas I sketch, along with an analysis provided by Barton & Sanford, provide a way of thinking about how this can occur. Finally I discuss at least two ways in which space can be important in language comprehension. First there is brute force: given the right incentives, we can form mental models of space, map arbitrary nonspatial dimensions into spatial ones, and derive spatial inferences. Second, according to Lakoff's Spatialization of Form hypothesis, abstract ideas can be understood by conceptualizing them using structured K-schemas.
1.1 In this Reply I will address four issues raised in the commentaries on Glenberg & Mathew (1992). First, Fernandez & Carriedo (1993) suggest that Glenberg & Mathew (1992) were "unfair" in one of their empirical criticisms of the McKoon & Ratcliff (1992) salience account. I acknowledge that that was the case. Familiarity with Glenberg & Mathew will facilitate understanding the section (2.0) dealing with this issue, but the remainder of this reply is self-contained. Second, Haberlandt (1993) and Zwaan & Graesser (1993) call for a consideration of nonspatial mental models. Here I sketch some ideas based on analyses of Lakoff (1987) that seem to bridge the gap between spatial and nonspatial mental models. Third, I will use these ideas to elaborate on a distinction between inferences proper and information available from the representation that does not seem to have a direct counterpart in the text. Fourth, I will extend Carreiras's (1993) discussion of the "survivor effect." As documented by Barton & Sanford (in press), this effect demonstrates that local INcoherence can be ignored if there is global coherence. The ideas I sketch, along with an analysis provided by Barton & Sanford, provide a way of thinking about how this can occur while respecting some of the intuitions that lay behind the notion of minimalism.
2.1 Glenberg & Mathew (1992) discuss several criticisms of the McKoon Ratcliff (1992) salience account of Glenberg, Meyer & Lindem (1987). One such criticism (section 4.4 of Glenberg & Mathew) is that the salience account cannot explain the absence of an effect of the associated/dissociated variable when the recognition probe occurs immediately after the critical sentence that instantiates the associated/dissociated variable. Fernandez & Carriedo (1993, Section 4.3) point out that the null effect can be explained easily as due to verbatim memory for the probed item presented in the critical sentence. In fact, a similar verbatim memory account is used by Glenberg & Mathew to explain the null effect from the mental model perspective! Clearly, Fernandez & Carriedo are right: the null effect cannot be used to discriminate between the salience and mental-model approaches.
3.1 Haberlandt (1993, section 3) points out that research and thinking about mental models stresses the spatial domain. Zwaan & Graesser (1993, section 3) discuss the need to incorporate nonspatial information into models derived from text. I address several aspects related to these points in the next few paragraphs.
3.2 Space is special. We have to survive in a dangerous, three- dimensional spatial environment. Given our species' size (and thus vulnerability to gravity, predation, injury), if we had no ability to represent and reason about space, that survival would be unlikely. However, we can represent space, and in fact it seems that our perceptual system recognizes the importance of space in that many subsystems (e.g., vision, audition, proprioception) provide information useful in forming those representations (see Freksa, 1991, for further development of this theme). In other words, real survival of individuals and species depends on being able to represent and reason about the natural environment. It would seem a good bet, then, that our conceptual system has sophisticated mechanisms for encoding space. Keep this idea in mind.
3.3 A background assumption in many theories of cognition is that representations of environmental information are, for the most part, formed of the same stuff as representations of linguistically derived information, namely, propositions formed out of abstract symbols that seem to be similar to words. No doubt this reliance on verbal-like propositions stems from several sources, such as a concern for language, the preponderance of experimental techniques derived from the verbal learning tradition, and the elegance and power of propositional systems. There is beginning to be an accumulation of evidence, however, that propositional representations are not the stuff of cognitive representations (Barsalou, in press; Lakoff, 1987; Sanford & Moxey, 1993).
3.4 Let's now view the background assumption (that perceptual and linguistic information share a common representational format) from a different perspective, namely, that space is special, and thus that the representation of the natural environment is the primary job of the conceptual system. From this perspective the assumption that perceptual and linguistic information share a common representational format implies that linguistic information is encoded by means of representations designed primarily to deal with space. Thus, from this perspective, language comprehension is intrinsically spatial.
3.5 How can this be? Clearly, not all language is about space, so how can a system evolved to deal with space represent nonspatial information such as time and causality? One solution is suggested by Glenberg & Langston (1992) and Glenberg, Kruley & Langston (in press). They propose that language understanders can learn to map nonspatial dimensions onto spatial ones and that this mapping affords spatial reasoning. This spatial reasoning may often take the form of "noticing," using the analog of spatial contiguity in the mental model to infer new relations (see Glenberg & Langston, 1992; Glenberg et al., in press). There is empirical evidence that this form of mapping can be done. Nonetheless, this form of mapping also appears to be quite effortful and to depend on special strategies or conditions (e.g., when pictures are available to suggest the appropriate mapping).
3.6 A second, more sophisticated, solution is described by Lakoff (1987) and Johnson (1987). They suggest that common, repeated, human experiences result in the formation of basic conceptual structures that are universal. These structures (which bear a family resemblance to the Piagetian notion of sensory-motor schemata) are termed kinesthetic image schemas (K-schemas here). As examples, we have a source-path-goal K-schema to represent movement in space from a source, along a path, and to a goal. We have a container K-schema to represent the idea of containment. In addition to incorporating spatial information developed through interaction with the environment, K-schemas are highly structured (nothing like pictures) and provide a type of logic. Thus, the container schema includes structural information such as inside, outside, and boundary, as well as logical notions such as EITHER inside OR outside, and transitivity of embedded containers. According to Lakoff and Johnson, these sorts of K-schemas can be used to represent a great variety of objects and ideas. Thus, the container schema is used in thinking about human bodies, cups, cars, lakes, and universes.
3.7 Furthermore, Lakoff and Johnson develop the Spatialization of Form hypothesis, according to which abstract ideas and complex arrays of loosely structured experiences (e.g., experiences of love) can be mapped onto more concrete K-schemas when there is a correlation between the structure of the K-schema and the structure of the abstract idea or experience. Thus, whereas Glenberg & Langston (1992) and Glenberg et al. (in press) propose a mapping of single dimensions, Lakoff and Johnson propose a mapping of correlated structures. For example, we come to view the abstract notion of "mind" as a container for ideas, and we say things such as, "Keep this idea IN mind" (end of section 3.1). Similarly, we can conceptualize abstract theories as buildings (a type of container) with foundations (assumptions) that can be viewed from different perspectives (beginning of section 3.4). Furthermore, even prototypically abstract relations such as "cause" appear (!) to be representable using structures very similar to K-schemas (see Talmy, 1988). In short, it is not inconceivable that what we have defined as abstract can be conceptualized as concrete and spatially structured. In fact, it is quite conceivable that this should occur, given that space is special.
3.8 Why is the sort of mapping that Glenberg & Langston (1992) describe strategic and effortful, whereas the sort of mapping that Lakoff and Johnson describe is natural and easy? I suspect that the difference has to do with structure and familiarity. In the Glenberg & Langston situation, subjects must map arbitrary temporal relations onto arbitrary spatial relations. At the very least, there are well-known constraints on immediate memory capacity that make this difficult. In the Lakoff and Johnson situation, the relations are nonarbitrary; spatialization of form occurs when there is a (perceived) correlation between the structure of the abstract domain and the structure of the concrete domain. Thus, at the very least, the ability to handle chunks of information as single units will make the Lakoff and Johnson situation easier.
4.1 Glenberg & Mathew (1992, section 3.2) tried to develop a distinction between (a) inferences and (b) information that accompanies a representation. The notions that we struggled with are similar to ideas expressed in Garnham (1992, Section 6; 1993, Section 3). I am sorry that distinction was not central to any of the commentaries, because I think it is an important (though difficult) issue. I will try again. When we think about inferences, the prototypical case is in going beyond what is given by applying some sort of logic-like process to the data (assumptions or premises) at hand. Thus, in the domain of reading comprehension, we think of the text as containing some basic ideas and the application of inference processes gets us beyond the basic ideas to inferences. This is wrong in at least two ways.
4.2 First, texts do not have any ideas in them. Ideas can only be in heads. Furthermore, getting even the most basic ideas out of a text requires that the text be interpreted, and this interpretation is always relative to some background (Fauconnier, 1985; Sanford & Garrod, 1981). When an author violates those background assumptions, we are brought up short (see, for example, O'Brien & Albrecht, 1992). Of course, once ideas are formed, logic-like processes can be used to derive various inferences, and in some circumstances this clearly occurs.
4.3 Second, because even the simplest language segment requires some sort of interpretation relative to background knowledge, the process of forming the interpretation will induce information that does not correspond to any of the words in the text. Thus, when we read a text about Sally starting her car in the garage, we don't have to infer that Sally is also in the garage. By virtue of conceptualizing the car and the garage as containers, the transitive "inference" comes along with the conceptualization; it does not have to be derived from the conceptualizations.
5.1 This sort of thinking provides a new perspective on minimalism -- or, perhaps, a new type of minimalism. The basic question becomes: what sort of understanding is achieved when the reader does not engage in inferential processes (as described in 4.1) but simply constructs an interpretation constrained by background assumptions? This type of minimalism helps us understand phenomena like the "survivor effect" (Barton & Sanford, in press) and the "illusion of knowing" (Glenberg, Wilkinson & Epstein, 1982).
5.2 In Barton & Sanford (in press), subjects were asked to write solutions to a version of the following problem (adapted from Barton & Sanford):
(1) A tourist flight crashes in the Pyrenees and wreckage is strewn
equally in France and Spain. Where should the survivors be buried?
Only 66% of the subjects noted that survivors are not the sorts of things that should be buried. Even more striking, when the term "survivors" was replaced by the phrase "surviving dead," only 23% of the subjects noted any anomaly!
5.3 What is going on here? Clearly subjects were "understanding" the text; they were willing to write solutions to the problem and the solutions were, in some sense, coherent. Nonetheless, the subjects did not appear to notice what is a clear anomaly. Even the locally incoherent (and nonsensical) "surviving dead" did not sound any alarms, and appeared to suppress detection of the anomaly. The account that Barton & Sanford offer is that subjects were understanding by developing a global, situational interpretation of the text. To the extent that the coherence at the global level can be maintained (e.g., "dead" is consistent with the global plane-crash situation), local problems are ignored and perhaps not even computed. Consistent with this account, when the global situation was changed (to a bicycle crash instead of a plane crash), detection of the anomaly increased from about 35% to about 80%. Thus we have a situation in which readers are forming a global interpretation but they do not appear to be engaging any formal, global inferencing. In fact, given the failure to notice the problem with "surviving dead," it seems unlikely that the readers are forming text-based propositions as the premises of inference processing.
6.1 There are at least two ways in which space can be important in language comprehension. First there is brute force: given the right incentives, we can form mental models of space, we can map arbitrary nonspatial dimensions into spatial ones, and we can derive spatial inferences. Second, according to Lakoff's Spatialization of Form hypothesis, abstract ideas can be understood by conceptualizing them using structured K-schemas derived from interactions with spatial objects. This sort of conceptualization results in a structured interpretation, and this interpretation may encode information that does not correspond to particular words in the text. On the other hand, this information is not formally inferred. Given that the goal of language comprehension is the construction of coherent, situational interpretations (i.e., mental models), we can begin to understand how one can comprehend (globally) and still miss the (local) point.
REFERENCES:
Barsalou, L. B. (in press). Flexibility, structure, and linguistic vagary in concepts: Manifestations of a compositional system of perceptual symbols. In A. C. Collins, S. E. Gathercole, M. A. Conway & P. E. M. Morris (Eds.), Theories of Memory. Hillsdale, NJ: Lawrence Erlbaum Associates.
Barton, S. B. and Sanford, A. J. (in press). A case-study of anomaly detection: shallow semantic processing and cohesion establishment. Memory & Cognition.
Carreiras, M. (1993). Minimalist misconceptions of mental models. PSYCOLOQUY 4(6) reading-inference.7.
Fauconnier, G. (1985). Mental spaces: Aspects of meaning construction in natural language. London: MIT Press.
Fernandez, A. & Carriedo, N. (1993). Reading Comprehension: In defense of a mental model approach. PSYCOLOQUY 4(7) reading-inference.8.
Freksa, C. (1991). Qualitative spatial reasoning. In D. M. Mark & A. U. Frank (Eds.) Cognitive and Linguistic Aspects of Geographic pace. pp. 361- 372. Dordrecht, The Netherlands: Kluwer Academic Publishers.
Garnham, A. (1992). Minimalism versus constructionism: A false dichotomy in theories of inference during reading. PSYCOLOQUY 3(63) reading-inference-1.
Garnham, A. (1993). An impartial view of inference making. PSYCOLOQUY 4(17) reading-inference.17.
Glenberg, A. M., Kruley, P. & Langston, W. E. (in press). Analogical processes in comprehension: Simulation of a mental model. In M. A. Gernsbacher (Ed.), Handbook of Psycholinguistics.
Glenberg, A. M. & Langston, W. E. (1992). Comprehension of illustrated text: Pictures help to build mental models. Journal of Memory and Language, 31, 129-151.
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., Meyer, M. & Lindem, K. (1987). Mental models contribute to foregrounding during text comprehension. Journal of Memory and Language, 26, 69-83.
Glenberg, A. M., Wilkinson, A. C., and Epstein, W. (1982). The illusion of knowing: Failure in the self-assessment of comprehension. Memory & Cognition, 10, 597-602.
Haberlandt, K. (1993). Understanding mental models and inferences. PSYCOLOQUY 4(5) reading-inference.6.
Johnson, M. (1987). The Body in the Mind: The Bodily Basis of Meaning, Imagination, and Reason. Chicago: The University of Chicago Press.
Lakoff, G. (1987). Women, Fire, and Dangerous Things: What Categories Reveal about the Mind. Chicago: The University of Chicago Press.
McKoon, G. & Ratcliff, R. (1992). Inference during reading. Psychological Review, 99, 440-466.
O'Brien, E. J. & Albrecht, J. E. (1992). Comprehension strategies in the development of a mental model. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 777-784.
Sanford, A. J. & Garrod, S. C. (1981). Understanding Written Language. Chichester: Wiley.
Sanford, A. J. & Moxey, L. M. (in press). Aspects of coherence in written language: A psychological perspective. In T. Givon and M. A. Gernsbacher (Eds.), Coherence in Spontaneous text. Philadelphia, PA: John Benjamins.
Talmy, L. (1988). Force dynamics in language and cognition. Cognitive Science, 12, 49-100.
Zwaan, R. A. & Graesser, A. C. (1993). Reading goals and situation models. PSYCOLOQUY 4(3) reading-inference.4.
Note: I thank the Zentrum fur interdisziplinare Forschung (ZiF) of the University of Bielefeld for support while preparing this article, Tony Sanford for many fruitful and enjoyable discussions about these issues, and Morti Gernsbacher for her insightful comments on an earlier draft. Art Glenberg ZiF, University of Bielefeld, Wellenberg 1, Apt. 10/18 D-4800 Bielefeld 1, Germany