October 23, 2007

Stylistics Meets Cognitive Science: Studying Style in Fiction and Readers’ Attention From an Interdisciplinary Perspective

By Emmott, Catherine Sanford, Anthony J; Dawydiak, Eugene J

Introduction Readers' attention to text has been studied for a number of years in both the Humanities and Cognitive Science. Within the Humanities, attention is discussed in stylistics using the notion of foregrounding (Mukafovsky), with the emphasis being on textual devices and, in consequence, the effect on readers often being taken for granted. Some stylisticians, such as Leech and Short, have nevertheless emphasised the fact that the "psychological prominence" of textual features is an important precursor to literary interpretation in models of foregrounding, and researchers in the area known as the "Empirical Study of Literature" have studied the impact of foregrounding devices on readers experimentally (e.g. van Peer; Hakemulder). Within Cognitive Science, psychologists in the field of "depth of processing" research have looked empirically at the amount of detail that readers notice when reading (e.g. Sanford and Sturt). This work shows how readers can be made more or less attentive depending on the type of linguistic structure information appears in, but this work has focused on a limited range of stylistic devices and has not drawn on observations about how stylistic features operate in texts to highlight key aspects of stories such as plot and theme.

This article aims to bring together these research traditions from different disciplines. We combine stylistic analysis of items which we intuitively feel to be attention-capturing devices in narrative texts with psychological testing to determine whether the selected stylistic features do indeed capture the attention of readers, in the sense of making them notice more. The theory of foregrounding has provided a productive framework for empirical work on attention within the Humanities, but this type of work nevertheless often relies heavily on readers' subjective reports of their beliefs about the extent to which particular stylistic devices have an impact on them (see Section 1 ). We show how empirical work within psychology offers testing methods that are less dependent on subjective reports, summarising a body of previous work in depth of processing research termed anomaly testing (see Section 2) and explaining a new experimental technique, the text change detection method, developed by Sanford and colleagues (Sturt et al.) (see Section 3), both designed to test the amount of detail that readers notice as they read specific passages.

We summarise a broad range of experiments conducted using the text change detection method, studying syntactic, semantic, and graphological features including clefting, sentence and paragraph length and punctuation, vocabulary choice, and use of italics (see Section 3). In addition to these stylistic devices, we also use the technique to examine narratological indicators of potential key events in narratives, such as pre-announcements (e.g. "Then this happened"), explicit signals of importance and surprise, and signals of the emotional impact of events on characters (see Section 3). We then examine the results (see Section 3 (iv) and Appendix 1 ) and offer some observations on the possibilities and limitations of this type of psychological empirical work for studying foregrounding in narrative texts (see Section 4).

1. Stylistics background: Style in fiction and the study of foregrounding

The research described in Section 3 has been jointly conducted by a stylistician and a group of experimental psychologists. The significance of the study for these different disciplines is different, in terms of its relation to the existing body of research in each field and the potential for providing fresh insights into the nature of reading in each area. In this section and the following sections, we provide the context in stylistics and psychology respectively for the empirical work that we conducted.

Within stylistics, the original motivation for the work was to examine how stylistic features in fiction can apparently be used to highlight key plot and thematic moments (Emmott, "Reading for Pleasure" and "Responding to Style"), guiding the interpretation of readers at these points. Both literary texts and popular fiction can make use of stylistic devices which appear to be used by writers with the intention of capturing the attention of readers at crucial points in stories. So, in Example 1, Roald Dahl combines initial under-specification of the leg of lamb (i.e. the initial use of pronouns in order to build up interest) with the use of a short sentence fragment (a noun phrase punctuated as if it were a full sentence) and a mini-paragraph (a single item alone in a paragraph) when the leg of lamb is first mentioned in full. This is the point at which the character sees the object which is to become the murder weapon in the story, so it is heavily plot significant:

Example 1

Everything was automatic now - down the stairs to the cellar, the light switch, the deep freeze, the hand inside the cabinet taking hold of the first object it met. She lifted it out, and looked at it. It was wrapped in paper, so she took off the paper and looked at it again.

A leg of lamb.

All right then, they would have lamb for supper.

(Dahl 25, our emphasis)

Likewise, in A2 below below, the sentence describing the amount of the bill is set in its own mini-paragraph and both this sentence and the earlier one describing the bill are short compared with other surrounding sentences describing the restaurant scene. The selected sentence contains information that is often elided in the presentation of restaurant scenes (see Schank and Abelson for a discussion of the restaurant schema), but is nevertheless of key thematic importance in this episode because the central character Martha has previously commented on the extravagance of the meal and the fact that if the bill comes to more than five pounds (a large sum of money in this story set shortly after the second World War), it will be more than some of her working class friends earn in a week.

Example 2

The waiter was bringing the bill. The restaurant was full now, it was about ten o'clock, and had more than ever the atmosphere of a family, of people who were at one with each other. And they were off guard now, with a licensed childishness about them, as if, threatened outside, here they found refuge. Across the room, a man with a heightened colour and a rakish look flicked bread pellets at a girl in a fluffy pink sweater, who flicked them back, giggling, while waiters watched indulgently.

The bill was for six pounds.

(Lessing 43, our emphasis)

Normally, sty listicians take these type of formatting devices for granted in terms of the effect they have on the reader, but the point of our study is to test whether they really capture the readers' attention in the way suggested above.

The notion of foregrounding has been used extensively in stylistics to explain these types of phenomena. The term comes originally from Garvin's translation of Mukafovsky and refers to how unusual linguistic features are thought to place emphasis on the form of the text, prompting a fresh perspective on the meaning of the text (see van Peer and Hakemulder for a summary of research on foregrounding). Leech and Short's work applied this idea to fictional prose at a time when foregrounding had been heavily associated with the study of poetry (e.g. Leech). In addition, Leech and Short highlighted the role of the reader in interpreting foregrounded items by including "psychological prominence" in their model of stylistic use, as shown in Figure 1.

In the practical analysis of their book, the emphasis of Leech and Short's work is heavily on textual features ("deviance") and literary interpretation ("relevance"), but Leech and Short nevertheless recognise the importance of the textual features having an impact on the reader, referring to how stylistic elements "register on a reader's mind" and reach "the threshold of response" (48; 49).

Foregrounding has been studied extensively in Humanities research in the area known as Empirical Study of Literature (ESL). Much of this research has depended on subjective judgements by readers about the impact they think stylistic features have on them. For example, in one of the classic studies in ESL, van Peer asks readers to assess the "strikingness" of stylistic features in poetry by underlining part of the poem that they find most "striking" (37-8). Subjective judgements may be appropriate at the level of overall interpretation, but may not necessarily be appropriate at the level of judging the psychological prominence of specific features. A reader will not necessarily know whether he/she is noticing information more in normal reading when information is in a cleft structure (e.g. "It was the woman who died first") rather than a non- cleft structure (e.g. "The woman died first"). Linguists assume that this is the case, but psychological testing often shows that our actual reading processes can be different from our assumptions about how we read. Also, a reader will not necessarily be able to report accurately on the degree of effect that a stylistic feature has on them (e.g. whether a cleft structure is more effective than italics for highlighting a specific piece of information), but this type of information can be found from psychological testing. A second problem for the type of methodologies typically used in the Empirical Study of Literature (ESL) is that even when responses are assessed by task-based tests rather than subjective reports, it can be difficult to isolate the effect of specific linguistic features on readers. Van Peer performed a memory-task experiment, using poems from authors such as Emily Dickinson and e.e.cummings for his materials (36-7). Informants were asked to read four poems carefully twice, each time being presented afterwards with the same poem with cloze-type gaps (i.e. selected words deleted) for completion, as in the following example from one line of Emily Dickinson's poem "The Brain is Wider than the Sky" (the whole poem was presented with selected gaps of this type).

Van Peer hypothesised that features which experts judged to be foregrounded would be remembered better and have with a higher cloze completion rate. The results showed the opposite, but van Peer discounts these results on the basis that too many factors were being tested simultaneously. For example, van Peer points out that the cloze-type test confuses the reader's memory of the poem with their general expectations about the language due to their knowledge of frequent collocations such as "side by side." He says that:

This contamination of different variables, together with the lack of experimental control over some of them, makes it impossible to dissociate the real cause(s) from concurring factors ... Further research, in which other factors influencing ease of memorability are systematically controlled by the experimenter, is needed to arrive at more conclusive evidence. In the light of what is known about the multitude of variables influencing recall, however, the design of such empirical tests will not be an easy matter. (99)

The text change detection method that Sanford and colleagues have developed (see section 3) aims to resolve some of the problems that van Peer encountered by more adequately controlling the experimental design (albeit in relation to the testing of attention rather than memory).

2. Psychology Background: Depth of Processing and Anomaly Testing

In psycholinguistics, "depth of processing" is a framework used to account for how much readers notice when reading a text (see Sanford; Sanford and Sturt for summaries of this research).1 Whereas Humanities researchers focus particularly on the upper end of the attention scale, there has been a great deal of interest in recent years in psychology in relation to the lower end of the attention scale. Prior to the depth of processing research, psychologists had assumed that a full propositional representation of a text is built up by a reader from the sentences of the text, although this representation of the text might subsequently degrade due to the limits of memory. The radically different assumption of depth of processing research is that the prepositional representation is incomplete on first reading, since readers are remarkably inattentive to the details of what they read. Empirical work within psychology has shown that readers often appear to ignore the full semantic information in what they read, this being referred to as "shallow processing."

A major test of shallow processing has been to see whether readers notice semantic anomalies. Readers are presented with sentences such as the following (e.g. Erickson and Matteson; Barton and Sanford; Sanford and Sturt):

(i) After an aircrash, where should the survivors be buried?

(ii) Can a man marry his widow's sister?

(iii) How many animals of each sort did Moses put in the Ark?

The last example is very well-known in the psychological literature and hence these types of examples are generally known as "Moses-type illusions." Readers frequently fail to observe that these questions are problematic since (i) dead people are buried; survivors are not, (ii) a man with a widow would be dead and so cannot marry anyone, and (iii) it was Noah who put the animals in the Ark, not Moses. Indeed, readers often try to answer these questions (e.g. by saying that the survivors should be buried where the relatives wish) rather than challenging the presuppositions. It seems that readers have a general awareness of whether words are appropriate to a context, since they are more likely to notice the anomaly if a non-biblical character such as a politician is mentioned in (iii) rather than Moses (see van Oostendorp and de Mul for experiments of this type) or where the context is changed to make the anomaly less plausible (Barton and Sanford), such as if a bicycle accident was mentioned in (i) rather than an aircrash. Hence, readers appear to be utilising the overall schema for the situation and, for example, recognising that survivors may be discussed in an aircrash scenario, rather than noting the full semantic details of the noun in relation to the verb it co-occurs with.

The significance of depth of processing for the current research is that clearly if readers are generally inattentive to what they read, stylistic devices can be used to make them more attentive at key points in a text. The Moses-type illusions are used to demonstrate this empirically. Researchers have shown that cleft structures can be used to focus the attention of readers on anomalies (e.g. Bredart and Modolo). Hence when readers are exposed to sentences of the following types, people typically answer "true" for (i), but are more likely to answer "false" or challenge the presupposition for the cleft sentence in (ii).

(i) Moses put two of each sort of animal in the Ark. True or false?

(ii) It was Moses who put two of each sort of animal in the Ark. True or false?

Conversely, attention can be decreased by embedding information in subordinate clauses so that readers notice less anomalies. In the following sentences (Baker and Wagner), the anomaly (that the liver is found only in humans) was noticed less frequently when in the subordinate clause in (i) than when in the main clause in (ii):

(i) Is the following true or false? The liver, which is an organ found only in humans, is often damaged by heavy drinking.

(ii) Is the following true or false? The liver, which is an organ often damaged by heavy drinking, is found only in humans.

The Moses-type illusions hence provide an illustration of how much detail readers notice, but these illusions are difficult to design in the large numbers needed for comprehensive testing of stylistic features. In addition, participants in such experiments may become aware of the fact that they are being repeatedly exposed to anomalies which may affect their reading performance. For these reasons, the text change detection method (see Section 3) was designed as an alternative technique to examine depth of processing. Using this method, we looked at a broader range of stylistic features than in previous research. In cases where we have tested a feature that had already been tested by the "illusion" technique (e.g. for clefting), the text change detection method can provide extra support for the psychological findings of the previous researchers.

3. The text change detection experiments

(i) Overview of the technique

The text change detection method is a new experimental technique developed by Sanford and colleagues (e.g. Sturt et al.) to study readers' attention for textual detail. The idea for this technique comes from vision research where a change detection method for visual details in pictures was already in use (see Simons and Levin for a summary of this work). Vision researchers show experimental participants a picture, then withdraw the picture for a brief delay of 100-500 milliseconds, then display a slightly adapted version of the picture (sometimes flickering between the two versions). Participants are asked to spot the difference. Even when significant parts of the picture have been removed (e.g. a large engine on an airplane, see Rensink et al.2), viewers are slow in identifying the change and when they do identify the change are often surprised at how they have managed to miss such an apparently obvious difference. Participants are termed "change blind" because of these perceptual limitations, but the degree of change blindness can be manipulated by making the changes more or less significant (e.g. by the placement of the item in central or peripheral vision, or by changing its size, colour, etc.).

The text change detection test works on the same principle, but with texts rather than pictures. The experiment uses pairs of textual materials which are either identical or identical apart from a change of one word. The reader is asked whether or not a change has taken place. The detection or lack of detection of any changed word provides a task to check how attentive the reader is being to the text at that point. Each experiment generally tests one or two stylistic/narratological features that are thought to be "attention- controlling," as listed in Section 3(ii) below. The hypothesis is that readers will be more attentive and therefore more likely to detect changes when they are reading materials containing the selected stylistic/ narratological features rather than when reading the controls.

(ii) Stylistic and narratological features examined

We examined our own database of 30 novels and short stories, selecting features which appeared to be used as "attention- capturing" devices. We judged this intuitively on the basis of devices used to highlight key plot moments and key thematic points (see Section 1 above) and also other reasons, such as emphasizing a sudden switch in rhetorical direction, providing a coda to a story or book chapter, or stressing information about a particular character (see Emmott et al. for examples). We tested a selection of stylistic features where the form of the items varied from control items at a number of different linguistic levels: graphological, lexical, grammatical, sentence structure and paragraph formatting), as in Group A below. In addition, we tested narratological cues where a pre-announcement3 was made with or without an indication that events were important, surprising, or had had an emotional impact on the narrator and/or characters, as in Group B below. Group B items are of interest due to their content because they contain explicit propositional information which we suspected was directly relevant to controlling attention. Group A. Potential attention capturers - stylistic form

1. Graphological - italics:

I beg of you not to assault me if I ask you one more question: Are you perfectly certain that you did not leave the trunk unlocked?

(Christie, "The Million Dollar Bond Robbery" 113, Christie's italics)

2. Lexical - low frequency and/or long words:

. . . the termagant concierge would have come raging up the stairs . . .

(Bonfiglioli 14, our added emphasis and Bonfiglioli's italics)

3. Grammatical - Clefting:

It was the second male prisoner who had gone in the other car . . .

(Christie, "The Kidnapped Prime Minister" 107, our emphasis)

4. Sentence fragments and very short sentences:

The child. His child. That, too, he sacrificed and injured - doomed to degradation.

(Gilman 94, our emphasis)

The wailing and the memories, as opposed to their meaning, were continuous, and, before he was conscious of the knowledge, he had pushed away his guards and started to run, helmet strap flapping, towards the noise and the fact.

Hyppolita was dead. Also she had been mutilated: her breast had been slashed over and over again in what appeared to be a genuine attempt to obliterate it.

(Maitland, "Hyppolita" 31, our emphasis)

5. Mini-paragraphs (containing a single sentence or a fragment):

He let himself in with his latch-key, and stole softly upstairs, to surprise his wife.

No wife was there.

He rang the bell. No servant answered it.

(Gilman 94, our emphasis)

Arthur let out a low groan. He was horrified to discover that the kick through hyperspace hadn't killed him. He was now six light years from the place that the Earth would have been if it had still existed.

The Earth.

Visions of it swam sickeningly through his nauseated mind. There was no way his imagination could feel the impact of the whole Earth having gone, it was too big.

(Adams 53, our emphasis)

Group B. Potential attention capturera - text content (signals of key narrative events)

1. Pre-announcements:

. . . she sucked and groped at his feet and knew that something was going to happen. (Maitland, "The Tale of the Beautiful Princess Kalito" 106, our emphasis)

2. Statements of surprise, emotion and importance, with or without pre-announcements:

I went into the dingy cafe and asked for dinner, and when I sat down to it I found to my supreme astonishment that there was another Englishman quietly eating his dinner opposite.

(Chesterton 52, our emphasis)

We expected all the stylistic and narratological features listed above to act as attention-capturing devices, but, as will be seen below, the results were more mixed than we had anticipated.

(iii) Methodology and experimental materials

The experiments were run over a period of several years, by a number of different researchers, rather than as a single set. Most of the work discussed here has been conducted as part of the University of Glasgow's STAGS project (Stylistics, Text Analysis and Cognitive Science: Interdisciplinary Perspectives on the Nature of Reading), including collaborative work with University of Strathclyde researchers (Alison J.S. Sanford et ai), but we also refer to work conducted in Sanford's University of Glasgow laboratory prior to the start of the STAGS project (Sturt et al.). There are some methodological differences between experiments as a result of differences in experimental design due to the type and number of stylistic features being tested, and also because the methodology itself was being developed during the period of experimentation. Hence the description below just gives a general overview of the type of methodology used.4

Table 1 provides an example of the type of materials used in the experiments, in this case for testing short items (short sentences or fragments of sentences) and mini-paragraphs (Emmott et al.). Each potential "attention-controlling" stylistic variant is said to be in a different "condition" by psychologists and there is also usually a control condition to provide a means of comparison (in the example below, Condition 1 is the control, since the phrase is not given as a separate sentence fragment or separate paragraph, but incorporated into a full sentence). Every text in each condition has two forms that are either identical or identical apart from one word. In Table 1, ([arrow right] white) indicates that the word "brown" was sometimes changed to "white" when the material was presented the second time.5 The hypothesis is that when reading materials in the different conditions, readers will detect more changes in the case of the potential "attention-controlling" stylistic variants (Conditions 2 and 3 in Table 1) than for the control (Condition 1 in Table 1).

Each experiment had sufficient versions of the materials and sufficient participants to ensure that statistical significance could be judged. In Emmott et al., for example, we tested materials of the type shown in Table 1 using thirty-six experimental materials, each in three conditions, with an additional 112 "filler" materials (passages which are included simply to prevent participants recognising recurring patterns in what they read, which might influence their reading behaviour). Reading in the experiment described in Emmott et al. was self-paced and text pairs (with or without changes) were presented with a 500 millisecond delay, with participants using a button on a response box to identify whether or not they detected a change and, if they noticed a change, reporting verbally to the experimenter what they thought it was. This type of methodology (with minor differences) was used for all the text change experiments discussed in this article. Table 1 has shown a sample of one set of materials for one of the experiments to test Group A items. As a further example, Table 2 shows sample materials for two of the narratological features in Group B that were tested together, preannouncements with and without emotion statements.

(iv) Results

Table 3 provides an overview of the results, indicating whether readers detected significantly more of the text changes for each stylistic and narratological feature. In addition, Appendix 1 shows the statistical analysis of the results for each experiment, grouping together items that were tested together.

The results for the features where the stylistic form changed (Group A in Section 3(ii)) show that readers were more attentive, in the sense that they noticed a higher percentage of text changes, when the stylistic devices were used. This applies to italics, clefting, use of a long and/or low frequency word, and use of a short sentence or sentence fragment with or without a mini- paragraph. By contrast, for the narratological cues (Group B in Section 3(ii)), which involve a change to the prepositional content, there was no increased text change detection. This applied to pre- announcements with or without indications of emotion and statements of importance and surprise. In the case of indications of surprise, the results showed significantly fewer text change detections for indicators of surprise than for the control.

4. Discussion of the text change experiment results and further research

For the Group A devices, the results matched our hypothesis. The stylistic features which we felt intuitively to be attention- capturing devices were confirmed to be so, in terms of increasing the amount that the readers noticed. This matched the earlier findings in the depth of processing literature for anomaly detection experiments on cleft structures (see Section 2). Also the fact that attention increased when information was split off into short sentences or sentence fragments makes sense in the context of anomaly researchers finding that information embedded in more complex structures is less detectable.

For Group B devices, the results were, by contrast, unexpected. We anticipated that if readers were told to expect a narrative event or given an indication that something was important, surprising, or of emotional significance, that this would be attention-capturing. Not only, however, was there no significant increase in text change detection, but stating that something was surprising made readers notice less.

These results require further exploration. Text change detection experiments require readers to do a very specific task involving noticing small details at word level (e.g. the change of a word from "white" to "brown"). It may be that changes to the stylistic form of the text (e.g. use of italics) make readers more sensitive to such details, whereas with the narratological cues, readers may be too busy engaging with what is happening next in the story to be focusing on details of wording. We may, therefore, need a psychological model of attention that separates the process of observing detail from the process of engaging with the propositional information such as signals of the narratological structure of the text and statements of the significance of events. It seems unlikely that the Group B narratological cues we have looked at are not attention-capturing in some way, so it may be that they are operating at a different level from the Group A stylistic devices and that other methods are necessary for testing the Group B items. One possibility is to use story continuation tests where readers are required to read materials similar to those in Table 2 and continue writing the story themselves after they read the narrative cue, such as a pre-announcement. If readers are more involved in the story, this cuing can make a difference to the type of continuation they write, such as yielding more detail and/or more indications of emotional engagement in the subsequent text. Another possibility is to use reading time experiments, which indicate whether a reader is taking longer to process a stretch of text which may suggest more involvement as a result of responding to narrative cues. So far, we have tested6 this with one type of cue, signals of importance, and have found that reading time does increase in such cases.7 More generally, there are other ways that this research could be extended. The stylistic features discussed here are only a selection of the potential attentiongrabbing devices that could be examined. We selected features that were relevant in the narrative texts we were examining, but devices which are particularly characteristic of other genres such as rhyme and alliteration in poetry could be investigated too. Also, the study of graphical features using the text change detection method could be extended to include other aspects of formatting such as capitalisation, size of print and colour of print. This might be useful for studying experimental styles of writing, but might also be used for graphic design purposes, such as judging the effect of different print formats in web communication, newspapers and advertising. In addition, the story continuation method can be used for examining the effect of under-specifying forms (e.g. the pronouns in Example 1) to see whether they can create interest in the object or person they refer to prior to the details being revealed.

The text change detection method could also be used to investigate in more detail the relation between different types of stylistic feature and also the relevance of context in interpreting a device. In principle, we could compare the relative effects of two devices of a similar type, such as the graphological devices of italicisation and capitalisation. We could also examine the impact that a stylistic device has according to its position. It may be the case that information embedded in the middle of a paragraph has less impact than information at the beginning or end or a paragraph, and likewise, it may be the case that information is handled differently depending on whether it comes at the beginning, middle or end of a whole story. Also, we could examine the effect of repeated exposure to a particular device. For example, in Margaret Atwood's children's book Rude Ramsay and the Roaring Radishes, the entire story is written with alliterative use of the letter "r." We do not know whether this device would have the same impact on readers when they first encounter it at the start of the book as when they become accustomed to the device, but this possible difference could be investigated empirically. In addition, more investigation is needed in relation to the different functions of stylistic devices. Italics, for example, can be used for different types of emphasis, such as simply highlighting a word for contrastive effect or contributing towards highlighting a key plot moment. Furthermore, italics are used for highly conventional purposes such as signalling a book title or indicating that a passage in a story is quoted from another source such as a letter. More research would be needed to see whether the more conventional uses of formatting prompt our attention initially and whether this is then suppressed as we make our overall stylistic interpretation of a text.

The above suggestions yield a large programme of psychological research for the future. The type of work we are describing here is nevertheless restricted to a specific aspect of foregrounding. In relation to Leech and Short's model (Figure 1 ), our emphasis has been on psychological prominence which has been underinvestigated in the past. There are, of course, other aspects of this model, statistical deviance and literary relevance, which we are not addressing here, but which are an important part of understanding attention during reading. We have selected our stylistic features largely on the basis of intuition," but these could also be selected using frequency analysis to judge statistical deviance. This type of work is already well described in books such as Leech and Short and has been extensively studied in corpus stylistics, so it might be fruitful to bring together this type of computational approach with our psychological method.

Cognitive psychology may also be able to contribute in certain respects to studying the literary relevance of foregrounded items. One reason why we have been interested in the heavy use of attention- controlling devices at key plot moments is that heavy foregrounding may sometimes seem to prompt readers to make inferences about the plot and any inference-making could be examined by psychological methods (see Emmott "Reading for Pleasure" and Emmott et al. for discussions of how this might be relevant to responding to clues in detective novels). In other respects, however, it may be necessary to use Humanities techniques from the Empirical Study of Literature to study literary relevance, since ESL researchers specialise in this aspect of reading. ESL researchers have examined issues such as how readers change their moral attitudes as a result of reading foregrounded passages in literary texts (e.g. Hakemulder), how they respond to foregrounded sections of a text in terms of its literary discussion value (e.g. van Peer 30), and how foregrounding inter- relates with aesthetic appreciation and ratings of affect (Dixon et al.; Hakemulder). We are not, therefore, claiming that the type of psychological techniques described here can answer all questions concerning foregrounding, but they may be of particular value in studying the middle level in Leech and Short's model termed "psychological prominence."


This article brings together research traditions in two distinct fields which can nevertheless be viewed as complementary: the stylistics work on foregrounding and the psychology work on depth of processing. We have presented a new experimental technique, the text change detection method, which can be used to test the effect on readers' attention of a wide range of foregrounding devices used in fiction at different linguistic levels. The results suggest that stylistic features involving graphical, lexical, and grammatical form and sentence/paragraph structure affect the amount of detail that readers notice as they read, in contrast with propositional cues such as pre-announcements and indicators of surprise, emotion, and importance, which do not. We have also indicated the need for a programme of research to explore this distinction further, in order to determine the relative effect of stylistic features on noticeability in different contexts of use and to find out in what way (if any) propositional cues gain the attention of readers. The research makes a contribution to stylistics in terms of presenting a method of testing the psychological prominence of foregrounding devices, and provides a new approach in Cognitive Science by making psychologists more aware of the range of stylistic features that can be studied in texts.

Readers' attention has been studied in stylistics using notions such as foregrounding (Mukafovsky) and psychological prominence (Leech and Short). In this article, we offer a fresh perspective on this topic from Cognitive Science. Our research draws on the psychological framework of "depth of processing" (e.g. Sanford and Sturt), which provides a context for studying different degrees of attention during reading. We identify stylistic features and narratological cues in fiction which we intuitively feel to be "attentioncapturing devices." We then use a new psycholinguistic technique, the text-change detection method (Sturt et al), to test whether the stylistic features we have selected really do make readers more attentive, in the sense of making them more alert to textual detail.


The text change detection experiments described in this article (with the exception of the experiment on clefting (Sturt et al.)) were funded by the Arts and Humanities Research Council (B/IA/ AN8799/APN13740 and B/RG/AN8799/ APN19525) and a British Academy grant (SG-36219) for our project, Stylistics, Text Analysis, and Cognitive Science: interdisciplinary Perspectives on the Nature of Reading (The STACS Project). The authors are grateful to the AHRC and the British Academy for their support.

1 The term "depth of processing" is used here as it is used in the research on attention (as in Sanford and Sturt and the other work discussed in this section). The term has a different use in memory research (Craik and Lockhart), which is not relevant here.

2 See Nichols and Stich (71) for reproductions of relevant photographs.

3 This term is used in Conversation Analysis for an utterance that gives an advance indication of an announcement, such as newsworthy information or a story. We are applying the term here to indications of forthcoming dramatic events in written fiction.

4 See references in Table 3 for the detailed accounts of specific experiments, where published.

5 The emphasis in the materials is added simply for discussion purposes here. The participants in the experiments saw the text without added emphasis.

6 In this unpublished experiment, we found that the reading time increased in the two sentences after the signal of importance (using t-tests, for the two sentences combined: t1(23) = 2.046, p

7 See also Miall and Kuiken for results showing increased reading time for passages containing more foregrounding. 8 We did, nevertheless, use the British National Corpus in our research to judge the frequency of items when we were specifically testing word frequency (Dawydiak et al).

Works Cited

Adams, Douglas. "The Hitch Hiker's Guide to the Galaxy." The Hitch Hiker's Guide to the Galaxy: A Trilogy in Four Parts. London: William Heinemann, 1986. 13-149.

Atwood, Margaret. Rude Ramsay and the Roaring Radishes. London: Bloomsbury Publishing, 2005.

Baker, Linda, and Jody L. Wagner. "Evaluating Information for Truthfulness: The Effects of Logical Subordination." Memory and Cognition 15(1987): 247-55.

Barton, Stephen B., and Anthony J. Sanford. "A case Study of Anomaly Detection: Shallow Semantic Processing and Cohesion Establishment." Memory and Cognition 21 (1993): 477-87.

Bonfiglioli, Kyril. "Don't Point that Thing at Me." The Mortdecai Trilogy. London: Penguin, 2001. 1-168.

Bredart, Serge, and Karin Modolo. "Moses Strikes Again: Focalization Effect on a Semantic Illusion." Acta Psychologica 67 (1988): 135-44.

Chesterton, Gilbert K. "The Curious Englishman." Daylight and Nightmare: Uncollected Stories and Fables. Ed. Marie Smith. London: Xanadu Publications, 1986. 50-3.

Christie, Agatha. "The Kidnapped Prime Minister." Hercule Poirot: The Complete Short Stories. London: HarperCollinsPublishers, 1999. 94-109.

_____. "The Million Dollar Bond Robbery." Hercule Poirot: The Complete Short Stones. London: HarperCollinsPublishers, 1999. 110- 17.

Craik, Fergus I.M., and Robert S. Lockhart. "Levels of Processing: A Framework for Memory Research." Journal of Verbal Learning and Verbal Behavior 11 (1972): 671-84.

Dahl, Roald. "Lamb to the Slaughter." Tales of the Unexpected. Harmondsworth: Penguin, 1979.22-31.

Dawydiak, Eugene J., Anthony J. Sanford, and Catherine Emmott. "External and Internal Sources of Attention Control: Findings from Change Detection." The Society for Text and Discourse. University of Minnesota, Minneapolis. 13-15 July 2006.

Dixon, Peter, Marisa Bortulussi, Leslie C. Twilley, and Alice Leung. "Literary Processing and Interpretation: Toward Empirical Foundations." Poetics 22 (1993): 5-33.

Emmott, Catherine. "Reading for Pleasure: A Cognitive Poetic Analysis of 'Twists in the Tale' and other Plot Reversals in Narrative Texts." Cognitive Poetics in Practice. Eds. Joanna Gavins and Gerard Steen. London: Routledge, 2003. 145-59.

_____. "Responding to Style: Cohesion, Foregrounding and Thematic Interpretation." Thematics: Interdisciplinary Studies. Eds. Max Louwerse and Willie van Peer. Amsterdam: John Benjamins, 2002. 91- 117.

Emmott, Catherine, Anthony J. Sanford, and Lorna I. Morrow. "Capturing the Attention of Readers? Stylistic and Psychological Perspectives on the Use and Effect of Text Fragmentation in Narratives." Journal of Literary Semantics 35.1 (2006): 1-30.

Erickson, Thomas A., and Mark E. Matteson. "From Words to Meanings: A Semantic Illusion." Journal of Verbal Learning and Verbal Behavior 20 (1981): 540-52.

Gilman, Charlotte P. "Turned." The Charlotte Perkins Gilman Reader: The Yellow Wallpaper and Other Fiction. Ed. Ann J. Lane. London: The Women's Press, 1980.87-97.

Hakemulder, Jemeljan F. "Foregrounding and its Effect on Readers' Perceptions." Discourse Processes 38.2 (2004): 193-218.

Leech, Geoffrey N. A Linguistic Guide to English Poetry. London: Longman, 1969.

Leech, Geoffrey N., and Michael H. Short. Style in Fiction: A Linguistic Introduction to English Fictional Prose. London: Longman, 1981.

Lessing, Doris. The Four-Gated City. London: Grafton Books, 1972.

Maitland, Sara. "Hyppolita." Telling Tales. London and West Nyack, N.Y.: The Journeyman Press, 1983. 31-39.

_____. "The Tale of the Beautiful Princess Kalito." Telling Tales. London and West Nyack, N.Y.: The Journeyman Press, 1983. 101- 10.

Miall, David S., and Don Kuiken. "Foregrounding, Defamiliarization, and Affect: Response to Literary Stories." Poetics 22 (1994): 389-407.

Mukafovsky, Jan. "Standard Language and Poetic Language." A Prague School Reader on Esthetics, Literary Structure, and Style. Ed. Paul L. Garvin. Washington, D.C.: Georgetown University Press, 1964. 17-30.

Nichols, Shaun, and Stephen P. Stich. Mindreading: An Integrated Account of Pretence, Self-Awareness, and Understanding Other Minds. Oxford: Oxford University Press, 2003.

Rensink, Ronald A., Kevin O'Regan and James T. Clark. "To see or Not to see: The Need for Attention to Perceive Changes in Scenes." Psychological Science 8.5 (1997): 368-73.

Sanford, Alison J. S., Anthony J. Sanford, Jo Molle, and Catherine Emmott. "Shallow Processing and Attention Capture in Written and Spoken Discourse." Discourse Processes 42.2 (2006): 109- 30.

Sanford, Anthony J. "Context, Attention and Depth of Processing during Interpretation." Mind and Language 17(2002): 188-206.

Sanford, Anthony J., and Patrick Sturt. "Depth of Processing in Language Comprehension: Not Noticing the Evidence." TRENDS in Cognitive Sciences 6 (2002): 382-6.

Schank, Roger C., and Robert P. Abelson. Scripts, Plans, Goals and Understanding. Hillsdale, N.J.: Lawrence Erlbaum, 1977.

Simons, Daniel J., and Daniel T. Levin. "Change Blindness." TRENDS in Cognitive Sciences 1 (1997): 261-7.

Sturt, Patrick, Anthony J. Sanford, Andrew Stewart, and Eugene J. Dawydiak. "Linguistic Focus and Good-Enough Representations: An Application of the Change-Detection Paradigm." Psychonomic Review and Bulletin 11 (2004): 882-8.

van Oostendorp, Herre, and Sjaak de MuI. "Moses Beats Adam: A Semantic Relatedness Effect on a Semantic Illusion." Acta Psychologica 74 ( 1990): 35-6.

van Peer, Willie. Stylistics and Psychology: Investigations of Foregrounding. London: Croom Helm, 1986.

van Peer, Willie, and Jemeljan F. Hakemulder. "Foregrounding." Encyclopedia of Language and Linguistics 2nd ed. Vol. 4. Ed. Keith Brown. Oxford: Elsevier, 2006. 546-51.

Catherine Emmott, Anthony J. Sanford, and Eugene J. Dawydiak

University of Glasgow

Eugene Dawydiak ([email protected]) is a postdoctoral fellow in psychology at the University of Glasgow, Scotland. He is currently a researcher on the STAGS project (Stylistics, Text Analysis and Cognitive Science), funded by the Arts and Humanities Research Council. He is co-author of several psycholinguistic research papers in journals such as Discourse Processes and Journal of Memory and Language.

Catherine Emmott (C.Emmot[email protected]) is a senior lecturer in the department of English language at the University of Glasgow, Scotland, where she teaches courses in discourse analysis, text linguistics, and grammar. She is director of the STAGS Project (Stylistics, Text Analysis and Cognitive Science) and she is assistant editor of the journal Language and Literature. Her publications include her book Narrative Comprehension: A Discourse Perspective ( 1997, Oxford UP), and she is currently writing a new book on Stylistics and psychology with Anthony Sanford.

Anthony Sanford ([email protected]) is a professor of psychology at the University of Glasgow, Scotland. He is a member of the psychology department, and of its Centre for Cognitive Neuroimaging (CCNi). His research interests are in the processes of language comprehension, and the neural mechanisms that enable such understanding to take place. His publications include his book Understanding Written Language (1981, with Simon Garrod), published by John Wiley, which outlined a knowledge-based theory of comprehension. He is currently preparing a new book with Catherine Emmott that presents a new and more refined account of comprehension, ranging in scope from stylistic considerations to cognitive neuroscience.

Appendix 1: Statistical analysis of results (% = percentage of correct text change detections, N = number of participants)

Italics (Alison J. Sanford et al.)

* The materials included italicised versus non-italicised words, tested by asking participants to identify changed items that were semantically close and semantically distant (2 x 2 repeated measures design).

* Italics (close semantic changes) = 73.1 %; no italics (close semantic changes) = 41.8%; italics (distant semantic change) = 94.2%; no italics (distant semantic change) = 82.1%

* The use of italics led to significantly more detections (F1 = 52.03, p

* N = 36

Clefting(Sturt et al.)

* The materials employed it-cleft and pseudo-cleft constructions to focus attention on either the critical word (focused condition) or a character (unfocused condition, detracting attention away from the critical word), tested by asking participants to identify changed items that were semantically close and semantically distant (2 x 2 repeated measures design).

* Focused (close semantic change) = 93%; unfocused (close semantic change) = 84%; focused (distant semantic change) = 96%; unfocused (distant semantic change) = 94%

* The use of focusing devices led to significantly more detections (F1 = 15.79, p0.05 (not significant); F2 = 4.57, p

* N = 40

Word frequency (Dawydiak et al.)

* The materials contained transitions between key words according to their high or low frequency.

* High to high = 25%; high to low = 67%; low to high = 29%; low to low = 77%

* F1 = 53.8, p

Copyright Northern Illinois University, Department of English Summer 2007

(c) 2007 Style. Provided by ProQuest Information and Learning. All rights Reserved.