Mental model building
The interpretation of longer stretches of text or speech, e.g. of more than one sentence, requires both the integration of information from different parts of the input and the construction of a mental model (Johnson-Laird, 1983) of what is being read or listened to. What evidence is there that listeners or readers build their own mental models First, early studies showed that listeners integrate information that they have heard, rather than remembering it as separate pieces of information. For example, Bransford and Franks (1971) presented participants with jumbled lists of sentences, including (12.1)–(12.3). Later the participants reported that they had heard the sentence in (12.4), which had not in fact been presented to them. This clearly indicates the integration of the sentences in (12.1)–(12.3) into a single mental representation.
Second, further studies also showed that the mental representations formed during listening or reading involve more than just the integration of information in the input. They are also based on the constructive use of world knowledge and the making of inferences about the situation being portrayed in the text Bransford, Barclay Franks, (1972). So, for instance, participants who heard the sentence in (12.5) confused it in a later memory test with (12.6), but participants who had heard (12.7) did not confuse that sentence with (12.8). This difference arises because the mental models associated with (12.7) and (12.8) are different – in (12.7) the fish swam beneath the log, but not beneath the turtles, who were to one side.
So discourse comprehension involves the construction of an abstract representation, using world knowledge and inferencing skills.
Given this evidence that the words and phrases of the input become interpreted as part of a more abstract mental model, it is interesting to speculate how long information from the text remains available during subsequent processing. ell e al tested this in a self-paced reading task using groups of four connected sentences (Dell, McKoon & Ratcliff, 1983). The first three sentences, e.g. (12.9) to (12.11), appeared in their entirety one after another, and a final sentence, (12.12), was presented word-by word. This final sentence was interrupted by a probe word, which appeared at one of the points marked by subscripts in (12.12) at different points for different participant groups. Participants had to decide whether the probe word had appeared anywhere in what they had read over this series of four sentences.
Two of the probe words were burglar, which is referred to explicitly in the final sentence, and garage, which is in the same earlier sentence as burglar. A third probe word was taken from a different sentence in the sequence, in this case the word bottles from the second sentence. The processing measure from the experiment is how much more rapidly each probe word was responded to, relative to a baseline condition. The result showed that there was an immediate benefit for both the repeated word (burglar) and the word from its earlier sentence(garage), as soon as the noun phrase the burglar is read i.e. at point 2, but the advantage for the word from the same earlier sentence garage drops off rapidly points 3 through 5. There was no such advantage at any stage for the control word bottles from a different earlier sentence. This indicates that in the reader’s mental model there is still an association of burglar with garage, which has been set up in the first sentence, so that garage is reactivated when burglar is repeated.
Neuroimaging studies of discourse processing have compared brain activity during the comprehension of isolated sentences with that found during the comprehension of stories and/or of texts longer than individual sentences, such as interconnected pairs of sentences see discussion of inferences and anaphora below. The general consensus is that discourse processing involves additional brain areas and that many of these additional areas are also involved in more general cognitive processing, including those that are typically involved when people attribute mental states to others (so-called Theory of Mind’ processes Bornkessel Schlesewsky & Friederici, 2007). This pattern of neural activity fits well with the notion that in developing an understanding of connected text we need to build a mental model of the discourse, based on our own experience and often projecting beyond it.
