Anthony Sanford,1 Andrew Stewart,2 Patrick Sturt1 & Annie Archambault2
1 University of Glasgow, 2 Unilever Research Port Sunlight
It is well-established that the representation of discourse in the minds of readers is centered on gist. There is also a growing amount of evidence that semantic processes and interpretation can be shallow, rather than deep and complete. The patchy literature on semantic illusions, in which anomalies in meaning are not detected, is testament to at least occasional shallow semantic processing, as is the literature on the pragmatic normalisation of interpretation, and apparent failures to fully repair faulty interpretations due garden-pathing. In this paper, we shall review some key aspects of these findings, including evidence that hints as focus as a possible controller of depth of semantic interpretation.
We report a preliminary series of experiments on a text-change detection paradigm, inspired by the use of change blindness in vision science. In the first method, participants read a text, and were then presented with the same text but for a single word change. After that, the text changed back to the original, and so on, through a series of cycles. In one experiment, critical changes were made to a verb that appeared either in the main clause or a subordinate clause of a two-clause complex sentence, embedded in a three sentence text. The change in verb was either to a semantically close verb (e.g., finished → completed), or to a distant verb (e.g., finished → started):
(1) The newsagent had just hired a new paperboy to cover the downtown area. After the paperboy finished (completed/started) his rounds, he ate his breakfast. There were a lot of deliveries to be made. [finished in Subordinate]
(2) The newsagent had just hired a new paperboy to cover the downtown area. The paperboy finished (completed/started) his rounds after he ate his breakfast. There were a lot of deliveries to be made. [finished in Main]
Changes to distant verbs were more readily detected than to close verbs, and changes in main clauses were more easily detected than changes in subordinate clauses. In a second experiment, changes to instruments associated with actions were made, demonstrating simple effects of pragmatic plausibility.
In the other technique, a single change was investigated, under a variety of conditions, for cleft constructions embedded in short texts. This provided another means of manipulating focus. It was found that large semantic changes to key nouns were detected more readily than small ones, and that detection rates were higher for noun phrases in focussed positions than those in unfocussed positions. We shall conclude by arguing the general utility of the technique as a way of investigating the level of representation of discourse during reading.