7 Methods and issues in stylistic analysis

7.1 Methodological considerations

Teaching people to ‘do’ stylistics is a very difficult task. This is partly because stylistics draws on a wide range of theories and methods from linguistics, and as a result does not have a single set of parameters which define the discipline.¹ This eclecticism is not a weakness, but a theoretically-legitimate strength. The purpose of theories is to shed light on the subject under consideration and as a result they tend to produce models which are simpler in some respects than the data they relate to. This is in order to generate fuller understanding of particular aspects of the data separately. Trying to capture the whole ‘truth’ about the data in one single unified theory of textual meaning would be unilluminating in its complexity.

As a result of this theoretical eclecticism, the question of how to go about a stylistic study is a complicated one, and requires the researcher to answer a number of questions, which will be introduced and discussed in the sections below. Note that any piece of stylistic research should aim to make clear the basis on which the analysis and interpretation is made, so that others are in a position to judge the results with an understanding of where they originated. This requirement – that stylistics should be as objective as possible in being rigorous and transparent – has at times been questioned by those who saw in this desire for clarity a claim to be ‘scientific’.

We have already seen (Chapter 3) Stanley Fish's argument against the claim of stylistics to be scientific, and the responses from Simpson (1993) and Toolan (1996) defending the aims of stylistic analysis from his attacks. Here, we quote Short et al.'s (1998) explanation of the stance of stylistics towards this subject at some length as a summary of the approach we also subscribe to:

For a stylistician, then, being objective means to be detailed, systematic and explicit in analysis, to lay one's interpretative cards, as it were, clearly on the table. If you believe that the number of interpretations that a text can hold is not indefinitely large (see Alderson and Short 1989 and Short and van Peer 1989 for empirical evidence to support such a view), then interpretative argumentation and testing will have to depend not upon something as unreliable as rhetorical persuasion, but on analysis of the linguistic structure of texts in relation to what we know about the psychological and social processes involved in textual understanding. This is what stylistics has traditionally involved. Of course, . . . we cannot expunge our personal response from our analyses, and would never want to. Like the natural and social scientists, we are human analysts, not machines. But, like them (…), we do think that it is incumbent upon us (a) to produce proper evidence and argumentation for our views, and to take counter-evidence into account when making our interpretative claims, (b) to make claims which are falsifiable and (c) to be explicit and open about our claims and the evidence for them. This does not constitute a claim to be natural scientists, but merely to be systematic, open, honest and rational.

The following sections aim to guide the student/researcher through the kinds of decisions that will need to be taken at the outset of a project in stylistics in order to make sure that the research fulfils these expectations.

7.1.1 Research questions – what are they and how do you develop them?

One of the best ways of approaching a new project is to develop a set of research questions that you intend to answer through your analyses. This might be a small set (about 3) of equally important questions, or it might be a single main question which is quite broad, and some subordinate questions specifying the detailed issues that are involved. Here are some hypothetical examples, reflecting these different possibilities:

Note that example 1 is a quite limited study of a small number of poems, perhaps suitable for a single assessed piece of work for a student, or a journal article for a senior researcher. Example 2 on the other hand has the potential to be a very large study of two or more novels. Example 2 also demonstrates that the list of supplementary questions can be expanded or varied, depending on the tools of analysis that the researcher thinks it appropriate to use. A main research question about something as general as ‘the style of X’ is almost bound to need quite a lot of focusing in the supplementary questions, which will determine the kinds of tool that will be used. Thus, a general question about style may be followed by particular questions about formal features (phonology, morphology etc.) or by questions about the kinds of conceptual metaphor, blending or other cognitive aspects of the texts concerned. These decisions will reflect the preferences of the researcher, who may find some tools of analysis more enlightening than others, but it may also reflect hypotheses that the researcher has developed about the data from a more informal survey of its linguistic features.

Note that the discipline of formulating research questions is a very useful one to acquire as it requires the researcher to think clearly about what the study is attempting to discover. Although one might first of all develop a hypothesis, such as ‘Alan Bennett's prose works use the same stylistic features as his plays’, this is only a beginning, and needs to be broken down into specific research questions before you can start planning the actual research itself.

7.1.2 Quantitative and qualitative approaches

Once the research questions are written, they act as an anchor for the rest of the project planning that is needed before the research proper can begin. Though many research projects in stylistics combine some aspects of both quantitative and qualitative research, it is certainly worthwhile being clear from the start about the extent to which, and where, each of these approaches will be used.

It is possible to use quantitative methods on any type or amount of data, but it is important to be clear from the start that once a study begins to count things, it must take seriously the need to test the significance of any statistical findings it intends to use. Thus, the counting of, say, pre-modifying adjectives in a short lyric poem would probably not produce enough statistical information to be reliable, whereas comparing the occurrence of certain grammatical features (e.g. intensive verbs) across two novels would certainly be a statistic worthy of testing.

Note that it remains important to recognise that stylistics, whatever its detractors say, does not ‘read off’ meanings from linguistic features, so the discovery of a significant (statistical) difference between two texts, though interesting, does not in itself constitute an interpretative finding. The question of why this difference is there often requires more detailed analysis both of the co-text and the context, and this is time-consuming work. The result is that one common way for research projects to progress is to start with a quantitative study of a large corpus of data to discover large patterns, and then to extract samples of the data in order to carry out in-depth qualitative analyses.

Let us say, for example, that the project aims to find out what kind of discourse presentation is used in the detective novel genre, and whether this differs from other novel genres. One way to approach such research questions would be to assemble a corpus of detective fiction, annotate it for types of discourse (speech, thought and writing) presentation, and then compare this statistically with existing corpora of other fiction. This process would reveal any large-scale differences in types of discourse presentation between this genre and fiction in general. The next stage would probably be to consider whether the detail was also different. For example, even where the amounts of FIS or NRSA were similar in the two corpora, there may be differences in the type of context in which they are used. This kind of analysis, which takes context as part of the process, is bound to be qualitative, as there is unlikely to be a pre-ordained set of categories into which each example would fit, even if there was time to look at enough examples to warrant statistical analysis. Though there may be some conclusions which demonstrate patterning, there may also be none, with each example appearing to differ from the others. This kind of research is basically inductive (i.e. bottom-up) as it does not adopt any pre-conceived ideas as to the likely outcome – in other words, it has no specific hypotheses about what will be found.

Before we consider the nature of qualitative analysis, it is worth reiterating the importance of statistical testing in quantitative studies. Whilst percentages and pie charts etc. often look impressive, the differences they purport to show may in fact be insignificant and may be exaggerated by the manner in which they are presented. The tools of quantitative studies are increasingly automated as large corpora become ever more accessible and software is produced to analyse them. This software (see section 7.2.2) usually includes significance testing in producing its results, so the researcher does not usually need to carry out complex calculations her/himself. Nevertheless, it is useful to understand basic concepts, such as normal distribution, how significance testing works and the relevance of levels of significance to a project's findings in order to truly understand the results of your research.²

Note also that corpus stylistics is not the only quantitative research that can be said to belong under the umbrella of stylistic research. There has been a distinguished tradition of empirical study (see for example van Peer et al. 2007) within stylistics since its inception, much of which involves the measuring and comparison of reader-responses using statistical methodologies.

As for qualitative research, this has been the basis of most of the stylistics of the last hundred years or so, until recently, when computer technology started to make corpus stylistics more attractive. Qualitative work allows the analyst to describe features that are not category-driven, and as we saw in Chapter 3, many of the functional systems of language have prototypical, rather than absolute, categories. Thus, a corpus study which uses these systems (such as transitivity or modality) would need to make sometimes arbitrary decisions about the categorisation of a particular example in order to complete the annotation of the corpus. This is not necessarily a problem if all the data is treated in the same way, though it precludes discussion of the subtleties that are often raised by the difficulties of categorisation.

Another reason why a researcher may choose qualitative research over quantitative is the nature of the data and/or the scope of the research questions. For example, the project may aim to describe in some detail the foregrounded features of a single poem or a small number of poems simply in order to demonstrate the linguistic basis of the foregrounding and perhaps to illuminate the process(es) by which the poem(s) contribute to the reader's meaning. Such a study would be unlikely to require any quantifying of features, and the more appropriate technique would be to provide a commentary on the poem(s) either line by line or under linguistic headings (phonology, morphology, syntax, semantics etc.). Similarly, a student could wish to explain linguistically why a particular passage from fiction seemed to have a particular literary effect, and this would also be a task best suited to qualitative analysis.

In the next section, we will consider the nature of ‘data’ in stylistics, and how it relates to the questions of qualitative and quantitative methods that we have been considering here.

7.1.3 Data

We saw in Chapter 1 that the data of stylistics, though typically literary, can in fact be potentially any text which uses language.³ The question of how much data and what kind of data should be analysed is difficult to answer in general terms as it is dependent on the research questions, the methodology and the tools of analysis to be used. In qualitative studies, the more data to be collected, the more sparse the analysis will be to cover it all. Conversely, a study looking only at one linguistic feature (such as speech presentation) will be able to cover a larger collection of texts than one where a number of tools of analysis are to be used.

The question of how much and which data will be analysed is also related to the question of how many tools of analysis will be applied. It is not straightforward to apply many of the stylistic tools to data as there are decisions to be made about how intensively the data will be analysed. For example, if the study is to look at transitivity choices, does that mean the main clauses only, or does it include subordinate transitivity choices too? If the latter, will this include single participles which are barely even separate clauses, and catenative verbs whose status as subordinate clauses is also debatable? If the study is about speech presentation, is it concerned with only certain character-interactions, or all interactions?

All such decisions about the scope of the data and how it is to be treated will follow from the research questions, and should be made clearly and explicitly at the outset, so that they can be described to others. Sometimes, these decisions need to be changed when problems arise from earlier decisions, but in such cases the analysis will begin again.

7.1.4 Stylistic models and tools of analysis

The kind of stylistic analysis which is carried out in any study will depend partly on the underlying assumptions that the researcher makes about the nature of textual meaning. Thus, there are some researchers who are keen to emphasise the reader's role in the creation of textual meaning, and who therefore take a cognitive model of textual meaning as their starting point. There are others who, whilst acknowledging the importance of the experience and understanding that readers bring to textual meaning, nevertheless wish to focus on the properties of the text itself. Some researchers may favour each of these approaches for different projects.

The student/researcher embarking on a project should take a view of the kinds of theoretical position s/he wishes to take as a starting-point, though in many cases the careful elaboration of research questions will make a particular theoretical standpoint inevitable. Thus, a project wishing to address the question of what interpretative common ground there is for readers of a poem will place emphasis on readers’ meaning, though there remains the question of whether the empirical collecting of reader responses or the application of cognitive models of understanding is to be chosen. In either case, this being stylistics, there will be close and continual reference to the text at the centre of the process.

If an initial hypothesis is more definitive than this, and suggests that readers will overlap in their interpretation of the surface meaning of a poem but differ in their broader interpretative strategies, then there is a latent assumption that one can distinguish theoretically between what might be called semantic and pragmatic meanings of such texts. Note that this distinction, between semantics and pragmatics, is a contentious area of debate, and one that is normally negotiated in relation to spoken face-to-face interaction. Stylistics, then, can appropriate techniques and debates where relevant, but also contribute to them by introducing new kinds of data, in this case written literary texts.

7.2 Stylistic studies

Having described in section 7.1 some of the key elements of stylistic analysis, our aim in this section is to give a flavour of the kind of studies carried out within stylistics, and of the range of methodologies employed by stylisticians. We hope that this will serve as a useful indicator of the range of approaches taken to the stylistic analysis of texts. Of course, in the space available we can do no more than provide a brief summary of typical studies and approaches. Nonetheless, we believe it is useful for the beginning student to have at least some measure of the analytical and methodological possibilities available. We have chosen to focus on three main areas – the qualitative analysis of literature, corpus stylistics, and responses to texts – though there remain, of course, a number of other approaches to stylistic analysis. For these, the reader is referred to the further reading section at the end of this chapter.

7.2.1 The qualitative analysis of literature

The qualitative analysis of literary texts has traditionally been the mainstay of stylistic analysis and will no doubt continue to be so. The importance of qualitative analysis can be seen in the fact that even the corpus-based and protocol analyses described in the subsequent two sections do not ignore this aspect of the analytical process. As West (2008: 137) notes in an article assessing the current state of English Studies, ‘the object of study of English is literature written in the English language’. While stylistic analysis may be carried out on languages other than English, West's point is clear: for stylistics, the text is key.

If we take a look at the range of qualitative stylistic studies that have been reported in recent years in Language and Literature, the main international stylistics journal,⁴ we find that, unsurprisingly, there are some that investigate the style of a single poem. These include Freeman (2005), which takes a conceptual metaphor approach to Sylvia Plath's poem ‘The Applicant’, and Melrose (2006), which looks at the stylistic ambiguities in Robert Browning's ‘My Last Duchess’. What is noticeable about these studies is that they each take a specific stylistic approach to the poem, and that in each case they are comparing their approaches with others’. Thus, Freeman is comparing her analysis with that of Semino (1997), who applies schema theory to the same poem, and Melrose is using stylistic analysis to explain the range of existing literary interpretations of the Browning poem.

Other studies of poetry include those which appear to try to encompass rather large bodies of material, such as Duffell (2002), who surveys the use of Italian metrical lines in English poetry after Chaucer. This work is inevitably reliant on his encyclopaedic knowledge of English poetry of a very long historical range, but can also, of course, draw upon metrical studies by others, since there is a long tradition of such work in English literature. Much more central to stylistics are those studies which focus on a particular poet, such as Somacarrera (2000), who investigates the effects of parallelism in a collection of poems by Margaret Atwood, or Goodblatt (2000), who compares three poems by three poets (Whitman, Williams and Hughes) to demonstrate stylistically the phenomenon noted by critics that after Whitman the unitary poetic (monologic) voice seems to have been displaced by multiple voices (Bakhtin's 1981 concept of ‘heteroglossia’). Perhaps the most focused article on poetry is Cauldwell (1999), which looks at just a few lines of Larkin's ‘Mr Bleaney’, and in particular attempts to relate Larkin's own varied renderings of these lines in recordings to the range of interpretations that have been attached to the poem by others.

Qualitative studies of larger bodies of data, such as prose fiction, are also frequently found in Language and Literature. Short fiction is one way to keep the amount of data manageable, and this is illustrated by Ryder (2003), who considers two short stories about time travel, and Malmkjær (2004), who takes a stylistic approach to questions of translation, in particular translations of Hans Christian Andersen's stories. There are also those which attempt to characterise aspects of the style of a whole novel, such as Chapman and Routledge (1999), who take a pragmatic approach to the detective novelist Paul Auster's City of Glass, Hidalgo-Downing (2000), who applies text world theory to Joseph Heller's Catch-22, and Wallhead (2003), who considers metaphors for the self in A. S. Byatt's The Biographer's Tale. Notice that each of these studies has a different way of making the large amount of data manageable, by taking a particular strand of the whole to investigate, usually from a particular theoretical angle. Another way to do this is to specify an extract or extracts to study, as in Cecconi's (2008) study of the trial scene in Dickens's Pickwick Papers. As we saw in relation to poetry, some researchers wish to compare versions of texts, either because they have been performed a number of times (see Cauldwell 1999) or because the author drafted and re-drafted the work. Sopcák (2007) considers foregrounding in the drafts of Joyce's Ulysses, and this, by its nature, limits the sections of the work that needed investigation, as Joyce will not have changed all of it entirely.

Other comparative studies may involve two (or more) novels. Fraser Gupta (2000), for example, compares two Singaporean novels, whilst Heywood et al. (2002) compare extracts from a popular and a serious novel, to identify the different challenges they raise for metaphor identification. Studies which range even wider across prose fiction are Harvey (2000), which investigates the representation of ‘camp talk’ in post-war English and French fiction, and Rash (2000), which considers the question of language use in Swiss (German-language) literature. Each of these studies makes a manageable project by taking a single theme which thereby reduces the amount of data that needs to be considered. There remain, however, significant challenges in ensuring that such wide-ranging studies are rigorous, replicable and explicit.

Recent developments in qualitative stylistics have included consideration of how stylistic methods may help us to interpret and analyse plays and films. These often take the textual aspects of the script or screenplay as their starting point. So, for example, Culpeper (2000) considers the cognitive stylistic features which give us insights into Shakespeare's characterisation of Katherina in The Taming of the Shrew, Ivanchenko (2007) takes a conversation analysis approach to overlap in Caryl Churchill's Top Girls and McIntyre (2004) considers how point of view is indicated linguistically in the text of Dennis Potter's Brimstone and Treacle. Others compare novels and their realisation in film versions. Thus, Clark (1999) compares Trevor Griffiths’s Fatherland with Ken Loach's film and Forceville (2002a) compares styles of narration in the novel and film of The Comfort of Strangers. These latter exemplify the increasing move in stylistics towards multimodal analysis, though progress is relatively slow in this field, there being a lack of clear analytical procedures for non-linguistic aspects of texts, which makes comparisons difficult.

The final type of qualitative stylistic study that occurs in Language and Literature is the investigation of non-literary texts. In the same way that a researcher may investigate everything from an individual poem (or a few lines of a poem) to a whole poetic movement or era, these studies vary in their scope. As with all stylistic study, the broader the scope, the less detail it is possible to include and the less comprehensive it is possible to be. An example of a broad study is by Beal (2000), who takes the representation of the Geordie dialect as her theme, and investigates it across a wide historical range of data including poems, prose and cartoons. A similarly broad study is Seargeant's (2007) study of the representation of epidemic disease in the texts (sermons, pamphlets and so on) of Early Modern England. The focus in this case, on the representation of disease, is what makes the project (relatively) manageable.

More focused studies often depend on corpora of data, though not all of them use quantitative methods to examine this data. Lambrou (2003), for example, draws on a set of recordings of oral narratives to consider the way in which speakers may move from one speech genre to another and Fitzmaurice (2000) investigates politeness phenomena in the letters of Margaret Cavendish. More focused still are those studies which concentrate on a single text or small number of texts, such as Pearce (2001), who examines a single party election broadcast by the British Labour Party, and Montgomery (1999), who analyses the reactions of Tony Blair, the Queen and Earl Spencer to the death of Princess Diana, and the public reception of these reactions. Another way to limit the data that one is investigating is to concentrate on a single word, and this is the strategy of Montgomery (2005), who studies the context and meanings of the word ‘war’ immediately after the attack on the World Trade Center on 11 September 2001.

In addition to the journal articles explored above, and those of other relevant journals, such as Style, Journal of Literary Semantics and English Text Construction, there are book-length studies which include stylistic analysis of work by genre, author, period and so on. The ‘Language of Literature’ series (Palgrave), for example, includes Blake (1989), Todd (1989), Jeffries (1993), Fowler (1995) and Lester (1996). These books of course have little chance of being comprehensive, and each has different ways of presenting a digest of the stylistic analysis available on their chosen topics. More research-project-based books still dependent on a qualitative, rather than a quantitative, approach include Stockwell's (2000) book on the poetics of science fiction; McIntyre's (2006) book on style in drama, including a detailed analysis of Alan Bennett's The Lady in the Van; Semino (1997), which makes detailed studies of individual poems; Gregoriou (2007), which investigates deviance in detective fiction; and Mandala (2007), which compares dramatic dialogue with ordinary conversation. Further work on the style of particular authors and text-types is forthcoming in the book series ‘Advances in Stylistics’, including West (forthcoming) on I. A. Richards, Sotirova (forthcoming) on D. H. Lawrence, Ho (forthcoming) on John Fowles, Montoro (forthcoming) on chick-lit and Piazza (forthcoming) on cinematic discourse.

This rather brief survey of some of the qualitative stylistic work which has been published demonstrates its range and variability in terms of scope and topic. It may be that some of the work discussed above will in future be replaced by more quantitative corpus studies, but there remains a place for detailed qualitative analysis in most stylistic research, not least because many of the most interesting aspects of style are not searchable automatically by computer. In addition, short texts, such as poems or individual advertisements, will always lend themselves to qualitative analysis of the following kind. Consider the following short poem:

McGough's poem appears to be a playful comment on the nature of getting old and suggests that, sadly, a reduction in the excitement experienced in life accompanies this. We are likely to see the word ‘italic’ as referring to some sexual highpoint and the lowercase letters as somehow symbolic of the speaker's now dull life. Clearly, the poem is not a complex one. Nonetheless, a qualitative stylistic analysis can help to determine how this intuitive response to the poem comes about. In this way, a stylistic analysis can help to support an interpretation of the text. The interpretation in effect constitutes a hypothesis that the accompanying stylistic analysis tests. With regard to analytical tools, we will examine the creation of foregrounding effects in the poem via the concepts of deviation and parallelism. Here we suggest that any other stylistician attempting to identify foregrounding via an analysis of linguistic levels (see Chapter 2) would uncover the same foregrounded elements. In this respect, the analysis we present is potentially replicable.

Perhaps the first noticeable element of the language of the poem is its graphology. The text is divided into two pairs of lines. The first two lines are graphologically parallel in that they consist entirely of uppercase letters. The second pair of lines are also graphologically parallel, since they consist entirely of lowercase letters. However, we can also note the presence of graphological deviation, as the common conventions of writing are that we don't write consistently in capitals or consistently in lowercase. Notice too that the poem is not left-justified but centred. This perhaps breaks our prototypical expectations about poetry and therefore constitutes a further graphological deviation. There is, then, a substantial amount of foregrounding in the poem as a result of this deviation and parallelism. The next task is to identify how the foregrounding connects to our interpretation of the poem. Since the poem is short, we can analyse it line by line:

Line 1 ‘ONCE I LIVED IN CAPITALS’

Living life ‘in capitals’ suggests living life to the full. This is reinforced by the uppercase letters. The fact that they are literally big and bold is graphologically symbolic. We can also note the polysemy of ‘capitals’. In addition to referring to uppercase letters, we might also interpret this to mean capital cities. The appropriateness of this interpretation is that geographical capitals are often thought of as being exciting and vibrant places to live, and this connects with the speaker's implicature of having lived an exciting life.

Line 2 ‘MY LIFE INTENSELY PHALLIC’

This line indicates more specifically the kind of life that the speaker claims to have lived. ‘My life intensely phallic’ suggests a life filled with sexual pleasures. The connection between the speaker's suggestion that his life was exciting and the notion that this excitement was sexual comes about because of the graphological parallelism of these first two lines. This pushes us to see a relationship of equivalence between the propositional content of the lines.

Line 3 ‘but now i’m sadly lowercase’

We can note here that the lowercase letters contrast with the uppercase letters, and that this appears to reflect the fact that the speaker's life is now the opposite of what it used to be. The graphological parallelism of the first and second pairs of lines now sets up a relationship of opposition between them, generated by a conventional assumption of capital letters as the opposite of lowercase letters (see section 3.2.5 for a detailed discussion of categories of opposition). If living life ‘in capitals’ is leading a busy, exciting life, then being ‘lowercase’ seems to mean the opposite, and this is reflected in the graphology.

Line 4 ‘with the occasional italic’

The final line contains arguably another piece of sexual innuendo. Despite the fact that the speaker's life is now fairly boring, there is still the occasional high spot. The literal indication of this comes via the graphological deviation inherent in the italicisation of ‘italic’, creating a graphologically symbolic effect (Short 2000). That we should see this as referring to sexual activity comes about because of the phonological parallelism of ‘italic’ and ‘phallic’. The rhyme pushes us to see the propositional meaning of these two words as being somehow equivalent.

This qualitative analysis uses the concept of foregrounding to identify the source of our initial interpretation of the text, explaining how deviation and parallelism generate particular stylistic effects which push us to interpret the poem in a particular way. We would suggest that anyone else applying foregrounding theory in an analysis of this text would uncover the same kinds of effects as we have noted here, and that in this sense what we have presented is a replicable analysis, albeit of a fairly simple text. Of course, it would be possible to use other stylistic frameworks than foregrounding in the analysis of the poem. Conceptual metaphor, for instance, would seem to be important, in the speaker's conceptualisation of himself as akin to particular kinds of writing. It is also the case that a fuller stylistic analysis could take into account aspects of context that we have not covered here. The point here is that any analysis that did so convincingly would supersede the one we have presented here. This, of course, is a hallmark of objectivity – the willingness to change one's mind if contrary evidence or a better explanation is presented.

Analyses of short texts such as the one we have just seen are a staple of stylistics, and may stand alone or form part of a study bringing together many such analyses, perhaps of parts of a larger corpus which has been excavated automatically to generate short texts to focus upon. In the next section, we will consider the development of computer-aided and computer-driven stylistic research.

7.2.2 Corpus stylistics

In recent years stylistics has profited considerably from the insights offered by corpus linguistics. Indeed, such has been the influence that there is now a developing sub-area of stylistics commonly known as corpus stylistics. Corpus linguistics is best viewed as a methodology for the analysis of large quantities of language data, as opposed to being a sub-discipline of linguistics in its own right. Although the origins of corpus linguistics can be traced back to the early years of the twentieth century and the work of the American field linguists (McEnery and Wilson 2001), corpus linguistics only came into its own with the advent of computing, which gave corpus linguists the capacity to store and search large electronic collections of text. Such databases are termed corpora (corpus is the singular noun). Modern corpora run to many millions of words and cover a wide variety of genres and text-types. Examples include the British National Corpus of 100,000,000 words of written and spoken British English from the 1990s, the Bank of English (stored at the University of Birmingham and the source of the COBUILD series of dictionaries and grammars), and the ICE (International Corpus of English) collection of corpora of international varieties of English.

The advantage of such corpora for stylisticians is that they can provide some measure of what might constitute the norms of language. In this respect, they offer an opportunity to test our intuitions concerning what might be the foregrounded elements of a text. It follows from this that using corpora in stylistic analysis provides another means of achieving the objectivity that stylisticians aim for. Interestingly, while some early stylisticians foresaw the advantages that having access to such large quantities of text could offer, others were dismissive of the value of such statistical information. Freeman (1970), for example, claimed that even if it were possible to ascertain frequency information for particular elements of language, ‘they would constitute no revealing insight into either natural language or style’ (Freeman 1970: 3). Louw (1993) rightly criticises Freeman for this remarkably short-sighted view, which has since been demonstrated to be erroneous (see, for example, Semino and Short 2004, Stubbs 2005, O’Halloran 2007a, 2007b).

The potential applications of corpus linguistics to stylistics are considerable. Aside from the possibility of discerning frequency information and thus a statistical measure of foregrounding, software for corpus linguistics also makes possible the analysis of complete texts. This has enabled stylisticians to circumvent the problems identified by Leech and Short (1981) concerning how to analyse complete texts in the requisite amount of detail demanded by stylistics, without such analyses being impossibly time-consuming and thus increasingly open to human error.

As an example of what is possible, Semino and Short (2004) report on a large-scale project carried out at Lancaster University to test the model of speech and thought presentation originally outlined in Leech and Short (1981) on a 250,000-word corpus of English writing. The model of speech and thought presentation proposed in Leech and Short (see Chapter 3 for a summary of this) was developed initially through the qualitative analysis of examples drawn from the close reading of texts. Short, Semino and other members of their project team tested this model by constructing a corpus of fiction and non-fiction texts and manually annotating these texts using the categories of discourse presentation presented in Leech and Short's model. As a result of this they found it necessary to modify the model slightly, since the data did not always support the suppositions that Leech and Short had made about how discourse presentation works in English. Semino and Short's corpus work thus refined the original model of speech and thought presentation, as well as providing a statistical measure of which categories were most common. From this, it became possible to state more confidently what constitutes the norm for speech and thought presentation, and thus to identify with greater certainty what constitutes deviation from this norm. Their corpus analysis also revealed more clearly the stylistic effects associated with each category on the cline.

Semino and Short's (2004) study relied on manual annotation of their data, though it is also possible to analyse large quantities of language stylistically without having to do this. Mahlberg (2007), for example, analyses Dickens's Great Expectations using the WordSmith Tools software package (Scott 2004) to extract data relating to repeated patterns of usage to determine what she terms their local textual function. The patterns that Mahlberg studies are five-word clusters: that is, repeated sequences of five words. (Clusters are also known as n-grams, where ‘n’ is simply any number.) Mahlberg identifies twenty-one different five-word clusters, each of which she categorises as belonging to one of the following groups: label clusters (e.g. ‘the Old Green Copper Rope-Walk’), speech clusters (‘it appeared to me that’), as if clusters (‘as if he had been’), time and place clusters (‘at the end of the’) and body-part clusters (‘his head on one side’). Her analysis of these patterns provides support for more literary-oriented and intuitive critiques of Dickens's language such as those provided by Brook (1970) and Quirk (1961). For example, on the basis of a qualitative analysis of Chapter 38 of the novel (particularly the pattern ‘I saw in this’), Quirk (1961) suggests that Pip's experience causes him to believe that what he sees is unequivocally correct, and that this contrasts with an inherent theme of the novel – namely that seeing does not equate to knowing. Mahlberg demonstrates that the as if clusters in Great Expectations contribute to the theme that Quirk identifies, noting, for example, that the clusters are used when the I-narrator Pip is describing how people look at someone or something (e.g. ‘looking dejectedly at me, as if he thought it really might have…’). In this way she is able to provide objectively garnered evidence for a previously subjective opinion.

Mahlberg's analysis is primarily what Adolphs (2006: 65) describes as intra-textual; that is, it is an analysis that examines a particular piece of language data in order to extrapolate information pertaining specifically to that data. This may be contrasted with inter-textual analysis. This latter analytical approach compares the linguistic features of the target text with those of a control text or collection of texts (what corpus linguists would term a reference corpus). Corpus-stylistic methods are not just appropriate for the intra-textual analysis of longer texts, of course. While this might be a primary application of the corpus methodology, corpora can also be used to provide supporting evidence for stylistic analyses via inter-textual analysis, often corroborating or disproving the results of qualitative analysis. It can, for instance, be used to discern why a particular linguistic choice is deviant. For example, in Chapter 1 we discussed an extract from a University of Huddersfield staff newsletter, in which the retirement announcement of a member of staff closed with the sentence: ‘Stephen intends to spend more time with his wife and caravan.’ The unintentionally humorous effect here clearly comes about as a result of the fact that we do not expect to see the word caravan turning up in the slot following the phrase wife and. As we noted in Chapter 1, on the basis of intuition we are much more likely to expect a word like children or family. But how are we to know that our intuition is either reliable or shared by other readers? A corpus allows us to check such an intuitive response. Searching for the phrase WIFE AND in the British National Corpus reveals that the most common collocates of this phrase are indeed nouns associated with family. (A collocate is a word that typically turns up in close proximity to our target word more than would be expected by chance alone.) Indeed, the only non-family-related word that turns up is thief, and a closer inspection reveals that this is part of a film title, The Cook, the Thief, his Wife and her Lover. Even so, thief refers to a human. Caravan is clearly a semantically deviant choice, then, being unrelated to either humans or families. A corpus can thus allow us to check our intuitions and can be particularly useful for analysts investigating a language of which they are a non-native speaker.

Collocational analysis can also reveal the connotations that a particular word may generate. Consider, for example, the following short poem by Roger McGough:

This is not a complex poem interpretatively and it does not need a detailed stylistic analysis to draw out what is perhaps its main theme. The speaker in the poem appears to be ruing his single status and wishing that he was in a relationship. The comparison the speaker makes of himself to a priest seems to be motivated by the fact that Catholic priests typically spend their lives alone, not marrying or entering into an intimate relationship with anyone. Of course, this is a prototypical view and, since the time the poem was written (1967), certain socio-cultural changes that have occurred might affect this interpretation. Nonetheless, it would be easy to dismiss the poem as trivial, were it not for the other connotations that the word priest generates. One way of accessing these would simply be to speculate on the basis of intuition. However, a corpus analysis of the collocates of priest can give us a statistical insight into which connotations are particularly likely to be triggered for the reader.

The statistical significance of collocations can be determined by calculating a mutual information (MI) score. This is calculated by comparing the observed frequency with which a particular collocate is associated with the node word (in this case, priest) with its expected frequency in the corpus as a whole. Most corpus linguistics software packages will carry out this calculation (or a similar one) automatically. For example, Mark Davies's web front-end to the BNC⁵ calculates MI scores for collocates and an MI score of 3 or over indicates that the collocation in question is not simply a result of chance alone but is statistically significant.

To return now to the McGough poem, the top ten statistically significant collocates of priest in the BNC, in descending order of significance, are lecherous, maxi, ordained, nun, deacon, Jesuit, celibate, fr, atonement and parish. What this list demonstrates is that priest does not have overwhelmingly positive connotations. In fact, the most statistically significant collocate, lecherous, has an obvious negative semantic prosody which may well cause us to re-evaluate our assessment of the poem as twee and whimsical. A word's semantic prosody is the connotations that it takes on as a result of the meanings of the words that it collocates with. A further statistically significant collocate of priest is celibate, which again may cause us to re-focus our assessment of the speaker of the poem, suggesting as it does that the comparison with a priest is intended to convey sexual frustration. Of course, these connotations may have been arrived at via introspection, but a corpus analysis provides some measure of objectivity in determining collocates, as well as a statistical calculation of collocational strength. It thus provides an objective means of finding supportive (or non-supportive) evidence for a particular interpretation.

7.2.3 Responses to texts

Concerns about variability in the meaning-potential of texts and how this affects the value of stylistic analysis have arisen from time to time, particularly whilst the parallel debates took place in literary studies as to the location of textual meaning, and whether it made sense to write about textual meaning in the face of theories that placed the reader at the heart of the meaning-making process. Interestingly, though literary studies traditionally privileged the concept of author meaning, and then abandoned it in favour of reader meaning, stylistics has tended to focus on textual meaning, though this has less frequently been the focus for literary studies.⁶

Stylistics has embraced the notion of the reader's contribution to meaning in a number of ways, which reflect on the one hand the view that there is some value in establishing the extent to which there are collective views of textual meaning-potential and on the other hand the view that individuals may have different, and ultimately infinitely variable, experiences of texts. In this section, we will focus particularly on those studies that have collected and analysed the actual responses of readers.

The study of readers’ responses to literary and other texts has been used in a number of different ways by stylisticians. This technique can be used in studies applying the methods of psychology to the reading process, and has been used with considerable success by, for example, Emmott (1997), who has been concerned with tracking the process by which a reader of narrative fiction will process the participants and events in a story. Her research, then, requires considerable preparation of suitable texts, which have to be invented to allow her to test precisely the kinds of textual feature which produce particular reactions in readers. This kind of empirical testing of reactions to texts is so far difficult to use with independently-occurring texts such as poems and novels, because of the need in experimental conditions to control the possible variables and be sure that the right elements are being tested. (For more on Emmott's work, see the discussion in Chapter 6.)

Other work has used the responses of readers to ‘real’ texts as some kind of evidence of the meaning potential of such texts. This work originates with a technique known as ‘protocol analysis’, pioneered by Short and van Peer (1989) and Alderson and Short (1989), in which the analysts (two in each case) were presented with a poem line by line and externalised their thoughts on the meaning of the poem as each line was read. Although not as experimentally controlled as Emmott's work, these experiments did allow for genuine literary texts to be used. Though not a completely natural reading experience, the resulting protocols were at least produced under identical conditions, so they can legitimately be compared with each other. Here is a sample of the kind of (written) responses in Short and van Peer (1989: 29):

Flutes and low to the lake falls home.

MHS The verb! Metaphorical – so it makes a musical (high pitched?) sound and then falls straight down into the lake (cf. the sound patterning and positioning of low). Home = end; but also anthropomorphic meaning (cf. his)?

WVP A verb. Subject = the fleece of his foam? Falls home is deviant (but cf. to plunge home). Makes a musical sound? And grooves? Note vowel parallelism in coop – comb/flutes – low; also the rhyme (home/foam). Hence this stanza ‘describes’ a stream thundering down from a height and dropping into the lake. In this dive the water seems to come to a standstill: connotations of rest and peace – home, flute, low.

The two analysts (MHS and WVP) seem to notice the same kinds of thing, and describe them in similar ways. Thus, they both notice the musical reference, though the line describes the movement of a waterfall, and they both comment on the vowel sounds in relation to the pitch of the music. The best summary of the findings of these studies is to be found in Short and van Peer (1989: 22):

Our most striking findings were that:

a) Our interpretations and strategies for arriving at those interpretations were very similar;

b) We had made very explicit and very similar evaluative remarks on the text;

c) These evaluative statements centred on practically identical text locations.

These studies were carried out, of course, by highly-trained stylisticians who shared certain other background features (e.g. gender, age, ethnicity etc.) which could also have produced similar effects. The authors are aware of these possible criticisms and make clear that what they are trying to do is to pilot a technique that might be used elsewhere in order to establish the nature and extent of common ground that readers have in response to texts. This technique has influenced certain other studies, such as Jeffries (2002), which used a visual text with almost no linguistic content to elicit linguistic responses about the meaning of the text. The informants in this case were students with relatively little stylistic training, and the results were used as texts for analysis in their own right, so that stylistic analysis could be used to draw conclusions about the informants’ responses that may be unavailable to them consciously. This technique is also influenced by the work of Rob Pope (1994), whose use of textual intervention as a pedagogical tool for teaching cultural theory also paves the way for stylistic research using intervention as a way of reflecting on the meaning of texts. Thus, textual intervention is one possible technique of stylistic analysis that could be used by analysts self-consciously or by informants in response to a request to ‘paraphrase’. The following short poem (5), for example, could be re-written in a number of ways (6–7):

The many permutations of this kind of re-writing may illuminate the ‘base text’ by showing what is lost when the language changes. In (6), for example, the loss is one of suspense, which in the base text means that we don't find out what the narrator's reaction is until the main clause in the final line. The rest of the original poem is a subordinate clause, meaning that there is ‘deferred gratification’ for the reader as s/he waits to find out what the proposition of the sentence will be. In (7) we are caused to reflect on the implicit nature of the meaning in the original, as (7) makes explicit (one reading of) the underlying meaning of the poem. The stylistician may, therefore, use such intervention him/herself, either as a way of discovering the linguistic foci of the text or as a way of explaining her/his analysis. The technique may also be used as a way of discovering the extent of overlap of interpretation and understanding between a group of informants.

What is important about this technique, and the response techniques, whether protocols like Short and van Peer's or more open techniques as used in Jeffries (2002), is that they all produce a legitimate response to the text, and that there is a range of response possible, starting from the most agreed-upon consensual meaning, and shading out to the most individual and varied response, which might be based on personal experience. This recognition, that there is a ‘shadow’ cast by every text, and that this shadow is darker (i.e. more consensual) the nearer you get to the text itself, is one that is examined in the introduction to Jeffries et al. (2007):

The idea of a text's ‘shadow’ is taken from terminology employed by criminologists and others to describe the great many unreported and unsolved crimes lying behind the ones we know about. How it relates to textual meaning is that it seems to me that there is a very deep shadow close to the text, representing the communal meaning that most readers would agree on, and a lighter and lighter shadow as we get further away from this consensus, including accidental meanings that relate to the individual circumstances of readers and changing meanings that are anachronistic to the text's context of production but happen to resonate with later historical audiences.

(Jeffries et al. 2007: xi)

The use of responses to text, then, will depend on the nature of the project being undertaken, though it is certainly worth keeping protocol analysis, textual intervention and other responses in mind as a useful tool for the stylistician.

7.3 Summary and conclusions

Our aim in this chapter has been to return to some of the principles of stylistics outlined initially in Chapter 1, in order to demonstrate the range of theoretical and analytical frameworks available for stylistic analysis. We should make clear, of course, that no one method of analysis is intrinsically better or worse than another. The key to a successful stylistic analysis is to choose the methodology and analytical framework that is appropriate to the text or texts in question. It is also the case that many analyses are substantially improved by the application of numerous analytical methods. Such triangulation can often substantially enhance the evidence in support of a particular hypothesis.

Further reading

In addition to the specific studies mentioned in this chapter, the following suggestions are recommended as introductory reading for the intending researcher in stylistics. For the qualitative analysis of poetry, Nowottny (1962) and Leech (1969) are early examples, and Verdonk (1993) and Jeffries (1993) provide further guidance on how to carry out the analysis itself. Cluysenaar (1976) is an early introduction to literary stylistics that covers prose and poetry, while Leech and Short (2007) is a classic text that is arguably the best available introduction to the qualitative stylistic analysis of prose fiction. Short (1996) is a very accessible introduction to the linguistic analysis of poems, plays and prose and is especially suitable for those new to the study of language in relation to literature. Short's (1989) Reading, Analysing and Teaching Literature contains a number of methodological insights into the practice of stylistics. An excellent introduction to the empirical study of language (and, indeed, other elements of the humanities generally) is Muses and Measures (van Peer et al. 2007). Wynne (2006) discusses corpus stylistics in general and Adolphs (2006) is a very accessible guide to the kind of electronic text analysis practised by corpus stylisticians. Further corpus-stylistic work can be found in O’Halloran (2007a, 2007b), O’Keefe and McCarthy (2010) and Mahlberg (2010). Finally, Wales (2001) is an indispensable dictionary of stylistic terms.