CC BY
110
Lingua mobilis № 4 (50), 2014
THE ROLE OF SCHEMATA AND MENTAL MODELS IN DISCOURSE COMPREHENSION: BRIDGING THE GAPS
M. S. Grineva
In cognitive linguistics and psychology of discourse processing two macro-theories have been put forward to account for discourse comprehension and knowledge representation - schema theory and mental model theory. The present paper explores the modern conceptualization, internal organization of schemata and mental models, their respective roles in discourse comprehension, and attempts to bridge schema theory and mental model theory in a holistic approach to discourse comprehension.
Keywords: schema; mental model; knowledge; context; discourse comprehension.
Cognitive Interface and Representational Comprehension
Representational theory of mind postulates that external and psychological reality (real and “possible worlds” in Johnson-Laird’s terms) is reflected cognitively in the form of mental representations. Putnam emphasizes the symbolic character of mental representations, arguing that there is no intrinsic connection between a representation and its referent [7, p. 21]. Representations are not passive copies of phenomenal experience, isomorphic with the external world, but the product of personal interpretation. Lakoff posits immanent connection between a mental construct and the human body: “mental structures are intrinsically meaningful by virtue of their connection to our bodies and our embodied experience” [6, p. 77]. Being physically embodied, mental representations are grounded in the local and global socio-cultural context. Thus, all mental representations have three inextricable dimensions: cognitive, social and cultural. Mental representations constitute a cognitive interface, which, according to van Dijk, is the locus of discourse comprehension.
Discourse comprehension is a complex process that draws information from multiple sources: linguistic input, context, background knowledge. Van Dijk claims that the interaction between this informa-
86
Лингвистика и перевод
tion is mediated by a cognitive interface - a bridge between individual cognitions (personal knowledge, beliefs, goals, opinions, affect) and social cognitions (general knowledge, attitudes, ideologies, norms, values) [13, p. 89].
Language users construe and interpret symbolic discourses with the help of colossal amount of “represented” knowledge: generic world knowledge, personal knowledge, linguistic knowledge, metalinguistic knowledge. Discourse interpretation makes use of two types of knowledge: generic conceptual knowledge, and similar past experiences that constitute personal knowledge. Generic knowledge is socially shared, tacit, formative of the presupposition pool or global common ground binding all “well-informed” members of an epistemic community. Personal knowledge includes unique personal and interpersonal memories of specific events (including communicative events). It is overtly asserted in discourse (as in personal stories) and may become interpersonal knowledge or local common ground.
According to van Dijk, generic conceptual knowledge is schematic in nature and is represented in various types of schemata (frames, scripts) in semantic memory whereas a mental model is a form of representation, reserved for memories of personal experiences in autobiographical memory [13, p. 91-92].
Schemata
Historically, the origins of schema theory go as far back as Plato and Aristotle. Kant in his “Critique of Pure Reason” (1781) put forward the notion of a knowledge unit that represented idealized conceptual categories. Modern schema theory was put forward by Bartlett (1932), the precursor of modern cognitive science. He developed his notion of schema based on the criterion of embodied, intentional, socio-culturally situated consciousness. Investigating the recall of folktales, Bartlett reported systematic recall errors which conventionalized the original story so as to increase conformity to Western cultural stereotypes. Bartlett proposed that human beings have substantial amounts of generic socio-cultural knowledge about the world in the form of unconscious mental structures - schemata. Bartlett defined a schema as “an active organization of past reactions, or of past experiences” operating as “a unitary mass” [1, p. 201]. In comprehension individuals actively and unconsciously relate new input data to knowledge schemata, whereby assigning meaning to linguistic
87
Lingua mobilis № 4 (50), 2014
signs - a process Bartlett called “effort after meaning”, a universal tendency to interpret new information in the light of prior knowledge and experience. [1, p. 227] According to Bartlett, comprehension and recall are achieved by “the organism turning round upon its own schemata and constructing them afresh” [1, p. 206], a process similar to what Rumel-hart referred to as “schema instantiation” [8, p. 105].
Modern schema theories were developed in cognitive linguistics (Chafe, 1987; Fillmore, 1975; Lakoff, 1987), cognitive psychology (Rumelhart, 1975, 1980; Minsky, 1975; Neisser, 1976; Brewer, 1984) as elaborations of Bartlett’s original idea. Brewer defined schemata broadly as “higher-order unconscious cognitive structures and processes that underlie the molar aspects of human knowledge and skill” [2, p. 120]. Schema has been metaphorically conceptualized as a “developing pattern” (Bartlett), an “organized setting” (Bartlett), a “conceptual skeleton” (Rumelhart), a “prototype” (Tannen), a “stage setting” (Bransford), a “scene” (Fillmore) a “play with a script” (Driscoll), a “computer program” (Driscoll). These conceptualizations suggest that a schema is an idealized flexible pattern tentatively applied to the linguistic input “on a trial basis” whereby new information is assimilated and conventionalized in terms of prior generic knowledge to ensure comprehension.
Schemata are recurrently characterized by Bartlett and other schema theorists as unconscious, active, purposive, task-oriented, context-sensitive, self-regulatory, “malleable” (Rumelhart), socio-culturally embodied (Johnson), culture-specific (Palmer), metaphorical (Lakoff & Johnson), at once conventional and idiosyncratic, interest-driven, affective, prototypical (Tannen), non-deterministic, admitting of uncertainty and vulnerable to distortions.
Minsky’s and Rumelhart’s subsequent elaborations of Bartlett’s theory viewed schemata as atomistic contrary to Bartlett’s holistic approach: a schema was no longer seen as “an organized mass” but divisible into discrete conceptual units - “nodes”, “slots” (Minsky, 1975) or “variables” and “values” (Rumelhart, 1980). Rumelhart defined schemata as “data structures for representing the generic concepts [...] underlying objects, situations, events, sequences of events, actions and sequences of actions” [8, p. 101]. The common characteristic uniting all post-Bartlett schema theories is their hierarchical organization. Schemata have variables that can be “bound by” different values “deter-
88
Лингвистика и перевод
mined by aspects of the environment, that is, by contextual and situational factors, as well as the to-be-comprehended stimulus” [8, p. 102]. E.g.: ASK schema has 3 variables: 1) Agent; 2) Resource; 3) Addressee. Schema variables correlate with semantic “cases” in Fillmore’s (1975) frame semantics (cf.: Agent, Objective, Patient).
Variables exert constraints on the values they can take on: each variable specifies the type of referent that can be realistically assigned to it and a distribution of possible default values. E.g. : Agent in ASK schema: a person, an institution, a phenomenon. Certain values are more typical, i.e. closer to the “average” of the distribution. In ASK schema a person is a more typical value than a phenomenon, cf.: She asked John to change the flat tire. - This job asks a lot of patience. Internal relations among the conceptual constituents of the schema remain constant regardless the particular referential values assigned to them: in ASK schema Agent requires Resource from Addressee. Assigning default values to the variables of a schema, conditional on the values of the already filled variables, based on the “assumed normality” principle, is what establishes the automatic “missing link” [3, p. 258] inference.
Once the assignment of definite values to variables has been made, the schema is instantiated [8, p. 105]. According to Rumelhart, variable constraints are seldom absolute: The “binding of a schema” (in Rumel-hart’s terms) allows for distortions and deviations from the stereotype which may become the basis for a heuristic effort trying to account for them, i.e. “bridging inferences” (Clark, 1977).
According to Chafe, when a variable conceptual constituent of a schema has been activated, the whole schema with the rest of its constituents is invoked. E.g.: I need Peter’s telephone number. - Call his brother. He must have it. Explicit mentioning of Resource and Addressee variables activate ASK schema.
Inferential potential of schemata allows them to be defined as a “cluster of interrelated expectations” (Chafe 1987, p.29), “structures of expectations” [9, p. 21], a “set of anticipations” (Neisser 1976, p.22). According to Tannen, relating new input to schemata generates expectations and hypotheses about situations [9]. Schema-based expectations explicate relevant presuppositions as part of the emergent common ground between the producer and the interpreter of the text, contribute to making default inferences in case of insufficient information, inferring pragmatic forces.
89
Lingua mobilis № 4 (50), 2014
Lakoff proposed a concept similar to schema - Idealized Cognitive Models - oversimplified, prototypical representations of phenomena, with respect to which concepts are defined. Lakoff & Johnson introduced a group of basic non-propositional context-independent metaphorical schemata: image schemata, spatial schemata, sensory-motor schemata, emergent at the level of bodily experience, that enable metaphorical conceptualization and visualization, spatial and temporal coordination of events in discourse comprehension [6].
Mental Models
The concept of a mental model was introduced by Johnson-Laird (1980, 1983) and elaborated for discourse analysis by van Dijk. John-son-Laird applied mental model theory to syllogistic reasoning and in-ferencing, and was the first to postulate the intrinsic connection between understanding and constructing mental models: “models are the natural way in which the human mind constructs reality, conceives alternatives to it, and searches out the consequences of assumptions” [5, p. 999]. Mental models are representations of “states of affairs”, experienced directly (via perception) or indirectly (via discourse). Linguistic comprehension yields a model of the situation “akin to one created by perceiving or imagining events” [5, p. 999]. According to Johnson-Laird, the structure of a model “mirrors the relevant aspects of the corresponding state of affairs” [4, p. 98].
Johnson-Laird posits two levels of discourse representation: a propositional text representation, close to the surface linguistic structure, and a model, close to the structure of the events described in the discourse. Symbolic propositional code is a means of constructing iconic three-dimensional models of the world. In Johnson-Laird’s view, propositions serve as “instructions for the construction of models” [5, p. 1000]. There is a one-to-one correspondence between the symbols of a propositional representation and lexical items of natural language (Kintsch, 1974, 1964; Fodor, 1975). Thus, words act as “cues” in constructing a model. According to van Dijk, semantic information encoded in the text does not have a function of its own, but conveys information about a possible world [12, p. 241]. The goal of a mental model is to analogically simulate and interpret discursive states of affairs, partially determined by the linguistic properties of utterances, partially by non-linguistic contextual assumptions. There is no primacy of imagery over propositions or vice
90
Лингвистика и перевод
versa. However, Chafe (1994) rejects the term “proposition” in linguistic analysis altogether on the grounds of it being generativistic and suggestive of formal logical analysis.
Van Dijk expanded on Johnson-Laird’s model theory and developed a typology of mental models applicable to discourse production and comprehension. Van Dijk defined mental models as “subjective representations in episodic memory” and laid particular stress on their biased character: “naturally partial.. .fraught with social and cognitive imperfections” [10, p. 383]. Mental models are ad-hoc, unique, non-generic, defining one specific interpretation of one specific text at a specific moment.
As Johnson-Laird van Dijk included a referential dimension into models by drawing a distinction between an episodic text representation (i.e. “what was said” -cf.: Kintsch’s “textbase” [14]) and a model of the state of affairs (i.e. ‘what happened’) referred to by the producer of the text [12, p. 241]. The model of discourse is richer than a textual representation as it incorporates not only the new relevant information explicitly expressed in the text, but also indexes presupposed socio-cultural knowledge, instantiated schematic knowledge, fragments of personal knowledge, schema-based inferences, evaluative beliefs, goals, affect. Van Dijk compared a text representation to “the tip of the vast iceberg” which is a model of the micro-world of discourse [10, p. 395].
Van Dijk and Kintsch first proposed the notion of a “situation model” as a bridge between social knowledge and semantic representation of the text to account for discourse coherence [14].Van Dijk further differentiated experience models and event models. Experience models represent the “ongoing subjective interpretation of everyday episodes in the lives of social actors” [11, p. 125]. Experience models are convertible into more general personal knowledge and generic knowledge schemata through generalization, decontextualization, and normalization.
Context models are a special case of experience models that represent the subjective interpretation of the relevant features of the objective social situation of communication. Van Dijk, Halliday, Johnson-Laird stressed that objective context has no bearing whatsoever on discourse interpretation - contextual parameters (setting, circumstances, social identity and personal characteristics of participants) only influence discourse interpretation as long as they are represented in a context model. The criterion of inclusion of each contextual category into a context model is subjective salience and systematic relevance in similar situa-
91
Lingua mobilis № 4 (50), 2014
tions in the past. Each individual has a unique interpretation of contextual features that affects his comprehension of the text.
Context models function as an executive control system which determines what knowledge schemata to activate for interpretation, and as a “pragmatic filter” which makes an assessment of the producer’s knowledge, monitors the scope of shared knowledge (cf.: van Dijk’s K-device), determines how much knowledge is presupposed as irrelevant, and how much knowledge may be intentionally suppressed, downplayed, implied, hedged or otherwise manipulated by the producer for contextual reasons of appropriateness, observing politeness, impression management, positive self-presentation. Context model is the necessary interface between discourse and knowledge on the one hand, and discourse and social situation, on the other. Context models account for individually variable interpretation of social categories, referential continuity, coherence, expectations and inferences in discourse comprehension.
Knowledge schemata play a twofold role in the interaction with the context. Schemata influence the construction of context models by determining which features of the context are perceived as salient or functionally important. Neisser expressed a somewhat extreme point of view that people only attend to those features of the environment that have been encoded in their schemata. On the other hand, the activation of a particular knowledge schema at the time of comprehension depends on the implicit or explicit contextualization cues in the linguistic input, interpretation of contextual features in a context model.
Event models represent subjective interpretation of the states of affairs derived from various discourses. In comprehension event models are constructed based on the semantic representation of the text, a context model, old experience models, instantiated generic socio-cultural knowledge, featuring evaluative opinions, affect. Event models bridge formal and semantic discourse structures, social reality and knowledge structures.
Event models are multimodal multidimensional analogical constructs. Basic-level schemata are used to structure the dimensions of an event model to situate an event spatially and temporally and coordinate sensory-motor relationship. Spatial dimension is conceptualized metaphorically by an image schema as a three-dimensional container with an inside, a boundary and an outside. Time is conceptualized in terms of motion (motor schema) either as moving by itself or as a static landscape which an observer moves across. Movement is structured by
92
Лингвистика и перевод
a “source-path-goal” and an “up-down” image schemata which determine directionality and trajectory of movement. Causality, intentional -ity are structured by force-dynamic schema. These dimensions interact with each other on the basis of conceptual metaphorical mappings from schema to schema [6].
Linguistic properties of discourse (formal and semantic structures) and its propositional representation are translated into analogical model information: speech acts, topics, lexicalization represent intentionality and protagonists’ identity; deictic expressions stand for temporality, referential, spatial and temporal continuity; word order, sentence structure signal directionality, movement, causality; topics, style and rhetorical devices signal personal opinions, attitudes, emotions of protagonists.
By way of illustration, consider a fragment of narrative discourse:
He turned into the alley, fast. It ran back between two houses, with a boardfence on one side. At that moment the fire siren soundedfor the first time, beginning and mounting to a slow and sustained scream that seemed at last to pass beyond the realm of hearing, into that of sense, like soundless vibration. (Faulkner “Light in August” pp. 345-346). The alley is metaphorically represented in the event model by a container schema (turned into) and a motor schema (it ran back); the sound of fire siren is conceptualized by up-down image-schema metaphorically connected with a motor schema (mounting to a slow scream); hearing and sense are represented as metaphorical containers (beyond the realm of hearing, into that of sense).
Discourse Comprehension System
Discourse comprehension, i.e. the construction of an event model, integrates information from three sources (see Figure 1):
1. Text and its linguistic properties;
2. Context of interpretation;
3. Knowledge structures.
Comprehension consists of the following macro-stages:
1. Construction of a context model that includes a propositional text representation and a representation of the social situation of communication;
2. Activation of relevant knowledge schemata and experience models.
3. Expectations and inferences arise out of the interaction between a context model and knowledge schemata;
93
Lingua mobilis № 4 (50), 2014
4. Construction of a coherent multidimensional event model.
Figure 1. Discourse Comprehension System
A set of candidate schemata is invoked directly upon the presentation of the linguistic input and/or based on the contextual cues. In Rumelhert’s view, comprehension is contingent on the capacity of schemata to “account for” the input situation. Context model specifies the relevant information to derive from discourse and include into an event model, activates generic knowledge schemata and experience models. Convergence of information is achieved by the interaction of bottom-up (input-driven) processing and top-down (schema-driven) processing. Cumulative analysis of linguistic information incrementally coming in from the text interacts with schema-generated conceptual predictions and inferences about the situation, checking and balancing each other. In case the current interpretation is inconsistent with conceptual predictions or vice versa, the interpreter will revise or reject the previous interpretation by making new inferences.
Conclusions
Discourse comprehension does not consist in linear understanding of words, clauses and sentences, establishing cohesion by making formal inferences; nor is it limited to the activation of relevant semantic knowl-
94
Лингвистика и перевод
edge. Discourse comprehension is an embodied, metaphorical, interpretive act, the result of which is not a propositional but analogical multidimensional mental construct - a mental model emulating the states of affairs referred to in discourse. Event models allow readers/hearers to “identify with the story” or “relate to the protagonists”. Discourse meanings are interpreted under the constraining and facilitating influence of context and knowledge. What follows is that each act of comprehension is innately unique: referentially the same text can be interpreted differently based on the unique properties of personal cognitions, knowledge schemata, context models. Knowledge schemata and mental models enable inferencing, identifying presuppositions, generating hypotheses.
Schema theory alone is not sufficient for explaining discourse comprehension, whereas mental model theory cannot be used as a kind of “skeleton key” theory to account for discourse processing, knowledge representation, understanding narratives, expository texts, reasoning, sensory-motor coordination. Thus, it is deemed methodologically imperative to elaborate a holistic cognitive approach to discourse comprehension bridging such constructs as schemata and mental models.
References
1. Bartlett, F.C. (1995). Remembering: A Study in Experimental and Social Psychology Cambridge: Cambridge University Press. - 317 p.
2. Brewer, W. F., & Nakamura, G. V. (1984). The Nature and Functions of Schemas. In R. S. Wyer & T. K. Srull (Eds.), Handbook of social cognition. Hillsdale, NJ: Erlbaum. Vol. 1, pp. 119-160.
3. Brown G., Yule, G. (1983.) Discourse Analysis. Cambridge University Press. - 288 p.
4. Johnson-Laird, P.N. (1980). Mental Models in Cognitive Science. In Cognitive Science 4, pp. 71-115.
5. Johnson-Laird, P.N. (1995). Mental Models, Deductive Reasoning, and the Brain. In Michael S. Gazzaniga (ed.), The Cognitive Neurosciences. Mit Press. pp. 999-1008.
6. Lakoff, G. & Johnson, M. (1999). Philosophy in the Flesh: The Embodied Mind and its Challenge to Western Thought. New York: Basic Books. - 624 p.
7. Putnam, H. (1991). Representation and Reality. MIT Press, - 136 p.
8. Rumelhart, D.E., Ortony, A. (1977). The Representation of Knowledge in Memory. In R.C. Anderson, R.J.Spiro & W.E. Montague
95
Lingua mobilis № 4 (50), 2014
(Eds.). Schooling and the acquisition of knowledge. Hillsdale, N.J.: Erlbaum. pp. 99-135.
9. Tannen, D. (1993). Framing in Discourse. Oxford University Press, 263 p.
10. van Dijk (2012). On macrostructures, Mental Models, and Other Inventions: a Brief Personal History of the Kintsch-van Dijk Theory. In C. A. Weaver, S. Mannes, & C. R. Fletcher (Eds.), Discourse comprehension: Essays in honor of Walter Kintsch. Hillsdale, NJ: Erlbaum. pp. 141-156.
11. van Dijk, T. (1999). Context Models in Discourse Processing. In H. van Oostendorp & S. Goldman (Eds.) The construction of mental representations during reading. Mahwah, NJ: Erlbaum. pp. 123-148.
12. van Dijk, T. (1980). Macrostructures: An Interdisciplinary Study of Global Structures in Discourse, Interaction, and Cognition. Hillsdale, N.J.: Erlbaum 317 p.
13. van Dijk, T. (2003). The Discourse-Knowledge Interface. In G. Weiss & R. Wodak (Eds.), Critical Discourse Analysis. Theory and Iter-disciplinarity. Palgrave MacMillan. pp. 85-109.
14. van Dijk, T., Kintsch, W. (1983). Strategies of Discourse Comprehension. Academic Press. 418 p.
96
