1.3.2. Generating Inferences during Discourse Processing

To achieve successful comprehension, the comprehender must often fill in details that are not explicitly presented in the discourse. That is, the comprehender must generate inferences about the events being described in order to maintain a coherent representation of the discourse. For example, when they encounter a pair of sentences such as ‘‘It was very cold that winter morning. Joe slipped on the sidewalk,’’ comprehenders need to draw the inference that there may have been snow or ice on the sidewalk in order to integrate the two statements into a coherent discourse model. The processes through which inferences are generated have been explored extensively.

One of the major research questions about inference generation during discourse processing has centered upon the issue of when inferences are generated and when they are not. Comprehenders mostly attempt to maintain local coherence when they process a discourse. The only inferences that are routinely generated by comprehenders are those that are required to maintain local coherence (such as the inference that there was ice on the sidewalk from the previous example).

Comprehenders are capable of drawing more global inferences from discourse, but these inferences are only drawn under certain circumstances (e.g., when the comprehender is attempting to process the discourse more deeply than usual). In contrast, the promiscuous position maintains that comprehenders routinely generate a wide range of inferences from the discourse, including those that are not strictly necessary to ensure local coherence. These inferences include inferences about the goals of the characters, the emotional state of the characters, the cause-and-effect relationship between events in the discourse, the intent of the writer in conveying particular pieces of information, and so on.

A compromise between these extreme positions is the constructivist position. The constructivist position holds that comprehenders routinely draw inferences that meet their goals as comprehenders, inferences that maintain the coherence of the discourse, and inferences that explain why different events in the discourse are taking place. On this view, comprehenders may appear to behave, depending on the nature of the discourse and the goals they have in comprehending that discourse. For example, if the comprehender is attempting to skim the discourse in an effort to quickly glean information, they may draw few inferences (in keeping with the minimalist position). On the other hand, if individuals are reading a discourse for enjoyment (say, if they are reading a detective novel) or if they are trying to learn about a new field of study, they may read the discourse more carefully and draw a wider range of inferences.

Another topic of import with regard to inference generation during discourse processing is the mechanism(s) through which the inferences are drawn. It is undoubtedly the case that many kinds of inferences (especially those relating to the local coherence of the discourse) are drawn by combining information from the incoming discourse with ‘world knowledge’ retrieved from long-term memory. For instance, when we read that it is cold outside, we can infer that it is probably wintertime. When we read that the character in the discourse has slipped on the sidewalk, the knowledge that snow and ice can be on the sidewalk during the winter is retrieved from memory, supporting the inference that the snow or ice has caused the character to fall. This sort of inference generation can be explained via the mechanisms of many of the theories outlined earlier. For example, the construction-integration model proposes that inference generation is supported by the rapid, automatic retrieval of world knowledge during the construction stage of comprehension. On the other hand, much less is known about how more global inferences (e.g., inferences about the goals of the characters, the purpose of the author in writing the discourse, and the theme of the discourse) are generated. It seems likely that world knowledge of the sort described above contributes to the generation of these global inferences. Additionally, metalinguistic knowledge about different genres of discourse (e.g., that newspapers are supposed to be balanced and factual, that detective stories should build suspense) may also help the comprehender to draw inferences about the purpose and structure of the discourse.