One of the main challenges of cognitive science has been the understanding of the way meaning is constructed during text understanding. It is commonly accepted that during text comprehension people construct a micro-world of characters, objects, spatial layouts, actions, events, emotions, etc.. The important questions relates to the kind of representations involved in this micro world as well as the type of information in those. Dijk & Kintsch (1983) came up with a model involving at least five levels of representations during text comprehension; surface code, textbase, situation model, pregmatic communicative context, and text genre.
The surface code preserves the meaning of the sentences. It does not stay for long in the memory, and experiments have shown that readers remember the surface codes of recent clauses, in most cases. Surface codes could change while retaining the same meaning. The textbase contains explicit propositions with regard to state, event or action. It is always in a reduced format, where the meaning is preserved (but not the surface code). Propositions contain predicates and arguments. In addition, textbases also include a small number of inferences which helps in text coherence by tying up explicit propositions. The situation models are the micro worlds of contents of the text. They include information on characters, objects, spatial settings, events, actions, etc.. In order to create situation models, readers many times go beyond explicit information and create their own interpretations depending on the knowledge available, which scientists term as memory intrusions. Think-aloud protocols, memory data, reading time etc.. have made conclusive proof of memory intrusions available to the researchers. Pragmatic communicative contexts involve an implicit dialogue between the author and the reader, where the communicative exchanges during text comprehension is facilitated through pragmatic principles. Such principles dictate the presence of surface cues which helps in the process of comprehension. Text genres deal with the generic classification of texts into descriptive, narrative, expositional and persuasive categories, with each category following certain structural component and features unique to its class.
The kind of inference readers construct during the process of text comprehension has been a matter of debate. One position, called the "promiscuous inference generation hypothesis", proposes that the inferences constructed include the goals and plans that motivate characters" actions, character traits, characters" knowledge, character emotions, causes and consequences of events etc., which in essence constructs a kind of video image of the situation model in the mind. The reader, according to this hypothesis, generate thousand of inferences for each text segment. The model, however, seems unreal considering the high volume of information involved with it. During 1980s Roger Schank proposed a computer model that was based on scripts. They argued that inference generation involves accessing and using information stored in natural packages of world knowledge (scripts, stereotypes etc.). However, the model is not well tested. A third hypothesis called the "minimal hypothesis" was proposed by McKoon & Ratcliff (1992) as an alternative to the first model but, once again, the model lacks test data in realistic scenario. It is believed that a more realistic model lies somewhere between the two extremes. Graesser (1990) proposed the constructionist theory which argues that four assumptions determine the class of inference generated during comprehension; reader goals, convergence and constraint satisfaction, explanations, and local and global coherence. None of the hypotheses, however, have succeeded so far in providing a realistic model.