Text Comprehension by Cue-based Constraint Satisfaction

How do readers comprehend text? How do they build a representation of the event to which the text refers? How do they infer important causal information that the text does not explicitly mention, and how is information from different past experiences combi

To sum up, cue-based constraint satisfaction provides a new approach to storycomprehension. Some of the difficulties found in other models of story comprehension areameliorated by the qualities of this approach. There is much to story comprehension that themodel does not yet do, and the model faces some difficulties in learning speed andarchitecture, but the fundamental characteristics of the approach are successful andpromising.

References

Bartlett, F. C. (1932). Remembering: An experimental and social study. Cambridge: Cambridge University

Press.

Charniak, E. (1983). Passing markers: A theory of contextual influence in language comprehension.

Cognitive Science, 7, 171-190.

Graesser, A. C. (1981). Prose comprehension beyond the word. New York, NY: Springer-Verlag.

Hinton, G. E., (1987). Connectionist learning procedures, technical report #CMU-CS-87-115, Computer

Science Department, Carnegie-Mellon University.

Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-integration model.

Psychological Review, 95, 163-182.

McClelland, J. L. & Rumelhart, D. E. (1985). Distributed memory and the representation of general and

specific information. Journal of Experimental Psychology: General, 114, 159-188.

McKoon, G. & Ratcliff, R. (1986). Inferences about predictable events. Journal of Experimental

Psychology: Learning, Memory, and Cognition, 12, 82-91.

Miikkulainen, R. & Dyer, M. G. (1989). A modular neural network architecture for sequential paraphrasing

of script-based stories. Technical report UCLA-AI-89-02, Artificial Intelligence Laboratory,Computer Science Department, University of California, Los Angeles, CA.

Posner, M. I., Keele, S. W. (1968). On the genesis of abstract ideas. Journal of experimental psychology,

77, 353-363.

Potts, G. R., Keenan, J. M., Golding, J. M. (1988). Assessing the occurrence of elaborative inferences:

Lexical decision versus naming. Journal of Memory and Language, 27, 399-415.

Rumelhart, D. E. (1981). Understanding understanding. In H. W. Dechert and M. Raupach (Eds.),

Psycholinguistic models of production. Norwood, NJ: Ablex Publishing.

Rumelhart, D. E., Smolensky, P., McClelland, J. L., & Hinton, G. E. (1986). Schemata and sequential

thought processes in PDP models. In D. E. Rumelhart, J. L. McClelland, and the PDP ResearchGroup (Eds.), Parallel distributed processing: Explorations in the microstructure of cognition, Volume

2. Cambridge, MA.: MIT Press.

Schank, R. C. (1981). Language and memory. In D. A. Norman (Ed.) Perspectives on cognitive science.

Norwood, NJ: Ablex Publishing.

How do readers comprehend text? How do they build a representation of the event to which the text refers? How do they infer important causal information that the text does not explicitly mention, and how is information from different past experiences combi

more sophisticated than simple pattern completion since it is capable of binding characters totheir roles in a script.

A second idea is that these constraints can be graded, and their support of aninterpretation can be graded. Pieces of text often may be only partially reliable as cues for aninterpretation. A polite or obnoxious pass, for instance, is only partially reliable at predictingthe outcome in the bar stories. Despite its inconsistency, it is valuable information. It hassome predictive value and could potentially be combined with other information to produce agood interpretation. Allowing constraints to take on strength values according to theirreliability, therefore, allows the model to use less than completely reliable information to itsadvantage.

The constraint view also carries with it the idea that each constraint makes its owncontribution to the interpretation. Each new constraint adds its own support to the evolvinginterpretation. This independence of constraints allows the model to update itsinterpretation as each new constraint is processed.

Additionally, the idea of independent constraints shows how novel stories can beunderstood. Each constraint adds its own support and influence to the whole interpretation.Understanding a novel story involves combining the memory constraints from each part of theinput. The associative constraints from each part add extra information, but they maysupport conflicting information. Misremembering a story to better fit prior experience can beone result of conflicting associative constraints (Bartlett, 1932).

Finally, the constraints imposed by cues can be learned. The graded strengths ofconstraints allows them to be learned incrementally with experience. The model does nothave to decide all-or-none whether a constraint is valuable. It simply assigns a strength tothat constraint commensurate with its reliability.

How do readers comprehend text? How do they build a representation of the event to which the text refers? How do they infer important causal information that the text does not explicitly mention, and how is information from different past experiences combi

Conclusions

Several of the model’s specific features are problematic. One problem is that theproposition representation is rather clumsy. First, each story can use a particular predicateonly once because the predicates must be used as unambiguous questions for training.Second, the seven thematic roles are both arbitrary and confining. Ideally, the model wouldlearn its own representation for propositions just as it learns …… 此处隐藏：31473字，全部文档内容请下载后查看。喜欢就下载吧 ……