Psychology Of Knowledge Activation Research Paper

One of the earliest and still most influential models of the comprehension process was developed by Kintsch and van Dijk (1978). According to this model, text is processed over a series of cycles. On each cycle, a small portion of the text is transformed into a set of propositions that are connected in working memory on the basis of argument overlap. Once connected, a subset of these propositions is held over in working memory in order to provide possible connections for the set of propositions input on the next cycle. As long as each set of input propositions can be connected to those currently in working memory, the text is considered locally coherent and comprehension proceeds smoothly. If, however, there are no readily available connections in working memory, the reader searches the long-term memory portion of the representation in order to find a proposition that provides a connection. Once found, that proposition is returned to active memory and the connection is made.

Many of these assumptions concerning when and if readers search memory, as well as what information is readily available to the reader, set the stage for our current theories and understanding of the knowledge activation component of the comprehension process. For example, many studies (e.g., Dell et al. 1983) have demonstrated that when a reader experiences a local coherence break, a search of long-term memory does indeed take place. Inactive portions of the representation are reactivated, providing connections that facilitate the re-establishment of local coherence. However, we are still grappling with the conditions under which this knowledge activation process occurs, and whether the activation process is active or passive.

For a brief period of time, these issues were best captured by contrasting the minimalist (e.g., McKoon and Ratcliff 1992) and constructionist (e.g., Graesser et al. 1994) views of how a reader gains access to earlier portions of a discourse, as well as information from general world knowledge. In the original paper presenting the minimalist view, McKoon and Ratcliff argued that the amount of information available to the reader was quite limited—primarily the immediately preceding one or two sentences. As long as this ‘readily available’ information allowed the reader to maintain local coherence, no further search or activation processes were necessary, nor would they occur. In contrast, Graesser et al. argued that the reader was motivated by the ‘search after meaning’ principle. A critical assumption of this principle was that readers were not satisfied with maintaining local coherence; they also sought to maintain global coherence. Maintaining global coherence involves the activation and integration of relevant information from earlier portions of the episodic memory representation of the text, as well as knowledge-based relations derived from general world knowledge.

Consistent with the constructionist view, there have been several demonstrations that information presented earlier in a text may be available even when a text is locally coherent. For example, Albrecht and O’Brien (1993) presented participants with passages containing a description of a particular characteristic of a protagonist (e.g., ‘Mary was a strict vegetarian’). Following several sentences that served to background the characteristic, participants were presented with a target sentence in which the protagonist engaged in an action that was locally coherent but contradicted the earlier-described characteristic (e.g., ‘Mary ordered a cheeseburger and fries’). Reading times on this target sentence were inflated when contrasted against appropriate controls. This indicated that, although not necessary for local coherence, the backgrounded characteristic was reactivated and disrupted the comprehension process.

In response to these kinds of findings, McKoon and colleagues proposed the memory-based text processing view (e.g., McKoon and Ratcliff 1995). According to this view, concepts currently in working memory make contact with other concepts processed earlier, as well as related concepts from general world knowledge through a fast-acting, passive resonance process (e.g., Ratcliff 1978). The primary difference between this view and the earlier minimalist view was refined as a result of the definition of ‘readily available’ to now include easy access to vast amounts of related information without the need to invoke any sort of strategic search processes on the part of the reader. With respect to the constructionist view, this passive activation component is an uncontroversial bottom-up component of the full comprehension process. Inclusion and acknowledgment of this component have muted much of the debate between the two contrasting views and shifted the focus toward a combined attempt to assess the kinds of information that are likely to be readily available to a reader. This bottom-up view of the activation component is shared by several prominent theories of memory activation during reading (e.g., Ericsson and Kintsch 1995, Goldman and Varma 1995, Kintsch 1988a, 1998b, Myers and O’Brien 1998).

It is easiest to explain the strongest assumptions of the memory-based text processing view within the context of the resonance model proposed by Myers and O’Brien (1998). It is important to note that the model is limited to processes and factors governing only the (re)activation of concepts and propositions during reading. Other models do extend this process to how that information is evaluated and ultimately integrated into the evolving text representation, most notably Kintsch’s (1998b) Construction-Integration (C-I) model and Goldman and Varma’s (1995) CAPSCI model. The resonance model is based on the assumption that concepts and propositions derived from the sentence currently being processed, or information residing in working memory as a result of reading earlier portions of text, serve as signals to all of memory. The intensity of those signals may depend on the degree of attention given to the text elements they reflect but the process initiated by the signals is not under the control of the reader; it is both passive and dumb. As in many models of memory (e.g., Ratcliff 1978), Myers and O’Brien assume that this process is one in which concepts and propositions in the discourse representation and in the reader’s knowledge base resonate as a function of the degree of match to the input. This match depends on the featural overlap among concepts, and on the argument overlap of propositions. Memory elements that are contacted by the initial signal in turn signal to other memory elements. During this resonance process, activation builds and when the process stabilizes, the most active elements enter working memory.

There are two critical features of the resonance process. First, it is continual; a signal is constantly being sent to all of memory, not just when triggered by a coherence break. What is resonating in long-term memory changes constantly with each change in the content of working memory. More important, the resonance process is dumb. Any information that resonates sufficiently is returned to active memory independent of whether that information will ultimately facilitate or hinder comprehension. Note that Kintsch’s (1998b) C-I model has the same assumption; because the construction (or reactivation) phase is both passive and dumb, it is quite possible that the output of the construction phase will be incoherent, and possibly contradictory.

This bottom-up view of knowledge activation eliminates the distinction between local and global coherence. Because most information related to any current input into working memory automatically resonates and is reactivated in response to that input, the reader simply needs to monitor the coherence of the information that is in working memory. By definition this will involve not only recently read information, but also related information from earlier portions of the text, as well as related information from general world knowledge.

There is mounting evidence that this knowledge activation is indeed passive and dumb. Several studies have shown that inactive portions of memory are activated and become part of the integration process, even when that information contradicts and/or disrupts comprehension (e.g., O’Brien et al. 1998). Even stronger evidence exists in studies in which information that is related—in terms of featural overlap— but irrelevant becomes active and disrupts comprehension. For example, Cook et al. (1998) found that if a narrative described a secondary character as a vegetarian (e.g., Joan), a later reference to the primary character (e.g., Mary) eating a hamburger resulted in the reactivation of the vegetarian characteristics assigned to the secondary character. It is difficult to explain these types of findings in terms of a reader actively searching memory for relevant information. But these types of findings become obvious predictions for comprehension models that include a passive activation component.

In many earlier models of knowledge activation there was an order in which the episodic memory trace and general world knowledge were activated. Kintsch and van Dijk (1978) assumed that the episodic memory trace was activated first, whereas van Dijk and Kintsch (1983) assumed that general world knowledge was activated first. Current models (e.g., Kintsch’s 1998b C-I model) now routinely assume that both sources of knowledge are activated in parallel—an assumption that must be made if knowledge activation is passive. In earlier work examining knowledge activation, researchers often examined activation of the episodic memory trace while holding general world knowledge constant (e.g., O’Brien 1987) or vice versa, holding the episodic memory trace constant while varying general world knowledge (e.g., Spilich et al. 1979). Because most models now assume that the two sources are activated in parallel, pressure is mounting to examine the interaction of these two knowledge sources. The most promising work in this area is that of Kintsch and his colleagues’ use of Latent Semantic Analysis (LSA, see the special issues of Discourse Processes, 1998, 25 devoted to LSA). LSA is a method for assessing the impact of general world knowledge on the comprehension process. It involves the building of large, but simple, associative networks of word meanings. This is done by measuring co-occurrences of words in large text corpora. The mechanism for activation is consistent with Kintsch’s C-I model (i.e., passive). LSA has proven quite successful in accounting for several cognitive phenomena including word recognition, sentence-word priming, and text comprehension. It is very exciting and promising that a simple associative network can go so far in accounting for the impact of general knowledge activation on the comprehension process. However, although computationally it provides a good fit to data, its value as a representational theory may be limited (see Perfetti 1998). Nevertheless, it provides an excellent starting point for considering the interaction of the two sources of knowledge activation.

The activation and instantiation of inferences, either necessary or elaborative, plays a crucial role in the comprehension process. Clearly, before any sort of inferential information can be incorporated into an evolving representation of a text, it must be activated (see Kintsch 1998b, for an excellent discussion of the distinction between activation and instantiation, p. 189). It must also be validated against general world knowledge (Singer and Halldorson 1996). By definition this means knowledge-based relations not explicitly stated in the text but already present in general world knowledge must be activated. Two critical questions are. What information is activated? And how does activation occur? Within the memory-based text processing view, any concept that is semantically related to what is currently being read should be passively and automatically activated. This makes a vast amount of information quickly and readily available (e.g., Ericsson and Kintsch 1995), some of it useful to the reader, some of it not. Currently this account of the initial knowledge activation phase can account for the majority of results showing activation of information that ultimately plays a role in inference generation (e.g., Murray et al. 1993). However, there is also evidence that some inferences require a slower-acting conscious process (e.g., Calvo et al. 1999). The memory-based text processing view cannot account for these sorts of active generation processes. The question of active vs. passive inferences remains an open and important issue for future research.

The issues and principles raised in knowledge activation in text comprehension also apply to issues in problem solving. Within Ericsson and Kintsch’s (1995) model of long-term working memory, activation of the knowledge necessary for problem solving is initially guided by the textual input. Automatic activation processes enrich explicitly stated information. This activation links information in working memory to large, pre-existing structures in long-term memory. The result is a large expansion of the information that is readily available to be used in problem solving. This is an important first step in understanding problem solving processes. However, beyond the initial activation stage, the issues in problem solving become enormously complex and messy. Problem solving is typically assumed to involve complex, conscious, resource demanding operations. Other than in situations involving well-practiced experts, those operations that can, or will, be utilized are difficult to constrain, and as a consequence are difficult to predict (see Kintsch 1998b).

The primary goal of models consistent with the memory-based text processing view is to explain as much of the overall comprehension process as possible without invoking strategic processes on the part of the reader. These models are incomplete because they typically lack a full specification of the mechanisms for evaluating and integrating the large amount of information they assume to be readily available to the reader. Some of those mechanisms are well specified in the top-down components of current constructionist models. Any complete model of the comprehension process will have to include many of the components identified in the memory-based text processing models as well as many of the top-down components identified in constructionist models. The current state of the field of text comprehension has clearly identified these two ends of the comprehension continuum. Future work that begins to identify the gray area between these two end points is likely to prove the most informative to a complete theory of the comprehension process.

Bibliography:

1. Albrecht J E, O’Brien E J 1993 Updating a mental model: Maintaining both local and global coherence. Journal of Experimental Psychology: Learning, Memory, and Cognition 19: 1061–70
2. Calvo M G, Castillo M D, Estevez A 1999 On-line predictive inferences in reading: Processing time during versus after the priming context. Memory & Cognition 27: 834–43
3. Cook A E, Halleran J G, O’Brien E J 1998 What is readily available? A memory-based view of text processing. Discourse Processes 26: 109–29
4. 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: 121–32
5. Ericsson K A, Kintsch W 1995 Long-term working memory. Psychological Review 102: 211–45
6. Goldman S R, Varma S 1995 Capping the construction integration model of discourse comprehension. In: Weaver III C A, Mannes S, Fletcher C R (eds.) Discourse Comprehension: Essays in Honor of Walter Kintsch. L Erlbaum, Hillsdale, NJ
7. Graesser A C, Singer M, Trabasso T 1994 Constructing inferences during narrative text comprehension. Psychological Review 101: 371–95
8. Kintsch W 1988a The role of knowledge in discourse comprehension: A construction-integration model. Psychological Review 95: 163–82
9. Kintsch W 1998b Comprehension: A Paradigm for Cognition. Cambridge University Press, New York
10. Kintsch W, van Dijk T A 1978 Toward a model of text comprehension and production. Psychological Review 85: 363–94
11. McKoon G, Ratcliff R 1992 Inference during reading. Psycho-logical Review 99: 440–66
12. McKoon G, Ratcliff R 1995 The minimalist hypothesis: Directions for research. In: Weaver III C A, Mannes S, Fletcher C R (eds.) Discourse Comprehension: Essays in Honor of Walter Kintsch. L Erlbaum, Hillsdale, NJ
13. Murray J D, Klin C M, Myers J L 1993 Forward inferences in narrative text. Journal of Memory and Language 32: 464–73
14. Myers J L, O’Brien E J 1998 Accessing the discourse representation during reading. Discourse Processes 26: 131–57
15. O’Brien E J 1987 Antecedent search processes and the structure of text. Journal of Experimental Psychology: Learning, Memory, and Cognition 13: 278–90
16. O’Brien E J, Myers J L 1999 Text comprehension: A view from the bottom up. In: Goldman S R, Graesser A C, van den Broek P (eds.) Narrative Comprehension, Causality, and Coherence: Essays in Honor of Tom Trabasso. L Erlbaum, Mahwah, NJ
17. O’Brien E J, Rizzella M L, Albrecht J E, Halleran J G 1998 Updating a situation model: A memory-based text processing view. Journal of Experimental Psychology: Learning, Memory, and Cognition 24: 1200–10
18. Perfetti C A 1998 The limits of co-occurrence: Tools and theories in language research. Discourse Processes 25: 363–77
19. Ratcliff R 1978 A theory of memory retrieval. Psychological Review 85: 59–108
20. Singer M, Halldorson M 1996 Constructing and validating motive bridging inferences. Cognitive Psychology 30: 1–38
21. Spilich G J, Vesonder G T, Chiesi H L, Voss J F 1979 Text processing of domain-related information for individuals with high and low domain knowledge. Journal of Verbal Learning and Verbal Behavior 18: 275–90
22. van Dijk T A, Kintsch W 1983 Strategies of discourse comprehension. Academic Press, New York