Roth, E.M., and Shoben, E.J. (1983). The effect of context on the structure of categories. Cognitive Psychology, 15, 346 – 378.

Purpose

The purpose of this article was to describe the relationship between typicality of ratings in a specified context, compared to context-free typicality that has previously characterized the concept literature. One of the methods used to analyze this was anaphoric resolution in silent reading.

Anaphoric resolution

Fluent readers use a number of strategies to understand what they are reading. One such strategy is the use of anaphoric reference. Anaphora is the process of referring back to information already given in an earlier clause or sentence (the antecedent). Anaphora is used within sentences or across sentence boundaries as a cohesive linking device. For example the sentence: Daniel said he wanted to leave, is the simplest anaphoric sentence. This article focused on how readers use semantic memory (memories of word meaning) to understand what is being referred to by the anaphoric expression. For example, reading the pronoun he, readers know that the sentence is referring to a male, without further clues.

Anaphora may be the same word as the referent, pronouns, different nouns, or noun phrases. When different words are used, the context is often used to access meaning. The closer the anaphor and referent are in meaning, the more quickly the sentences are read (Garrod and Sanford, 1977). It has further been established that reading time is faster if the anaphor is a typical exemplar of the referent, than an atypical exemplar. Thus, if the referent were tree, oak would be read faster than palm.

Similar results have been found in research into semantic memory in which exemplars judged as more typical are verified as belonging to a category much faster than those judged atypical  For example, robin is a more typical exemplar of bird than is penguin. Further, exemplars can be graded by typicality. That is, they are rated by how well they fit the category example. This typicality, or goodness-of-example (GOE) rating predicts how quickly the object will be verified as belonging to a category.

In reading, categories and objects appear in a context that influences which exemplars are most easily accessed. For example, if the context is a barnyard, a chicken is a more typical example of a bird than a robin.

Instantiation hypothesis

According to the instantiation hypotheses, the context of some sentences is so narrow that the target exemplar is obvious. In these situations, only one exemplar is likely. For example banana is the obvious example in the sentence: The monkey peeled and ate the yellow fruit. This demonstrates the instantiation hypothesis and accounts for situations in which only one exemplar is likely. 

Refocusing hypothesis

In most contexts, many possible exemplars would be acceptable.  The context may activate a typical example, but there is a set of other similar objects that would also be suitable. These examples are graded based on their similarity to a prototype. The set of exemplars is, however, constrained by the context. In the refocusing hypothesis, many possible exemplars are activated in a specific context. The most likely example is the one that is most similar to the best exemplar. In a sentence such as: The musician tuned the strings of his instrument before playing the classical piece, the prototype violin may be activated. However, other choices are acceptable and will vary in how representative they are by how similar they are to the prototype, e.g., viola, cello, bass, or harp.

Restructuring hypothesis

The third possible way in which anaphoric resolution occurs is described by the restructuring hypothesis. In this case, the entire GOE and the different exemplars are activated according to the information provided by the context. Representativeness is not based on similarity to the best exemplar but rather GOE is based on appropriateness of examples in that context.

Experimental Work

Experiment 1

First, context sentences were rated for how well they suggested the target exemplar. Then typicality of the target exemplar was rated. Context sentences were provided for 30 typical and 30 atypical target exemplars, and there were three different context conditions for each target exemplar.

Subjects were presented with the 3 context sentences (spaced apart), a target sentence, and a test sentence. Reading time for the target sentence was measured. There were three conditions for the context sentences, Bias +, Bias -, or neutral. In the Bias + condition, information in the context sentence suggested the exemplar that was given in the target sentence. In the Bias – condition, context sentences provided category information and the target sentence contained an exemplar that was an unlikely, though possible, member of the category. In the neutral context condition sentences provided no context information. The test sentence was given to ensure that reading was done for comprehension.

Reading time of the target sentence was fastest for the Bias + condition and slowest for the Bias – condition for all (typical and atypical) target exemplars. These results showed that context had an effect on comprehension.

Experiment 2

The restructuring and refocusing hypotheses were contrasted.

2a

In the first part of experiment 2, subjects rated exemplars in different contexts. The rating patterns showed that in a specific context, categories do have GOE distributions. These results suggest that the GOE distributions in context are generated in a similar way to concepts in isolation. In context, however, the order of representativeness of exemplars changes.

Experiment 2b

The purpose of this experiment was to examine whether the GOE ratings obtained in a context predict the speed with which a referent is identified as belonging to a category. That is, whether GOE distributions in context are similar to concepts in isolation but in a different order. Reading time was measured as subjects read context sentences. Reading times were faster for sentences in which the target exemplar was more representative in both congruent and incongruent contexts. Both category terms in context and category terms in isolation have a graded GOE distribution. However, the order of representativeness is not necessarily the same. This result is consistent with the restructuring hypothesis which asserts that it is not just that a best exemplar of a category is changed in a specific context, but the whole GOE distribution is reordered due to the information supplied by the context.

Experiment 3

Mental processes used to establish membership in a context-constrained category are similar to the processes used to establish membership in a category presented in isolations as well. In this experiment subjects were required to respond to context sentences, and response time was measured. This experiment used semantic memory to measure category member decision tasks by asking subjects to verify whether an exemplar belonged in a category in a specific context. Response time was measured. These exemplars were all members of the category, but half of them were not possible.

In addition, typicality measures that occur in the absence of context had no effect on verification time for in context true or false items. In half the cases when both typical and atypical items were presented in context, the true item that is more typical for that category in the absence of context was verified more slowly that the less typical but more context appropriate one. Thus, once items appear in context, typicality is unimportant.

Conclusion

The typicality of items in categories is completely restructured when the categories appear in a context. It is important, therefore, to extend semantic memory models so that criteria for category membership include context.

There appears to be no one clear way in which anaphora in text are comprehended. In some highly constraining context sentences, only one exemplar is anticipated and thus there is instantiation. There may be other acceptable alternatives but there is a best example. For unlikely examples, the text requires bridging in order to make sense. For example, a bird that is going to be consumed for Thanksgiving dinner is most likely a turkey, but goose or duck are acceptable. Quail, however, would require a bridging sentence in order to facilitate comprehension.

Instantiation is, however, not the only process involved in anaphoric resolution. People do not always anticipate the item when we read a category name. Sometimes the context is too vague. In these cases, GOE structure may be more important.

This page was last updated:
07/18/2006 00:36