81
Published in: Gerbig, Andrea & Anja Müller-Wood (eds.). 2006. How Globalization Affects the
Teaching of English: Studying Culture Through Texts. Lampeter: E. Mellen Press. 81-109.
Where the Computer Meets Language, Literature, and
Pedagogy: Corpus Analysis in English Studies
Ute Römer
University of Hanover
1 Introduction: Corpora and English Studies
In the field of linguistics, electronic corpora (i.e. large systematic collections of
spoken and/or written text stored on the computer) and corpus-analytic tools have
been widely used for more than forty years now. For very good reasons, linguists
all over the world draw on corpora in language analysis and description, and it is
part of the daily working routine of a steadily growing number of linguistic
researchers to run software programs in order to get access to language patterns,
collocations, and word frequencies, or simply to retrieve examples of a particular
lexical item in use. Since they enable, to a high degree, objective views on the
language, these resources can help us discover what language is really like, which
means that we no longer have to speculate and rely on our (or other individuals')
intuition as the sole source of data.
The aim of this essay is to show that corpora and corpus-analytic tools can
not only be extremely valuable in linguistics but also in language pedagogy and in
literary studies. It will be explored in what ways some basic techniques of corpus
analysis can provide insights into language and literature – insights that intuitive
approaches to the same objects of study may fail to create. Hence, one of the
questions I will address in the following is, "Can corpus analysis help to improve
interdisciplinarity in English Studies?" In other words, referring to the title of the
present volume, it will be asked, "Can corpora contribute to a reconciliation of
82
literature, linguistics, and pedagogy?" These three branches of English as a
subject are usually studied separately and, more often than not, isolated in
research and teaching. In addressing these questions, I mainly aim to open up new
perspectives on working with texts to the non-(corpus)linguist, but I also wish to
put some emphasis on the student perspective. As the editors of this volume
rightly note in their introduction, we as researchers and teachers of English
linguistics, didactics, or literary and cultural studies often fail to explain to our
students what and where the common ground is between these disciplines. In
research and teaching we strictly separate what, in fact, belongs together and
should form an integrated whole. It is hence not surprising that students find it
difficult to spot the existing, but often well-hidden, relations between the branches
of the subject. In the present paper I shall argue that the profitable use of one
particular tool across these separated branches can mean one important step
towards their reconciliation. The corpus and everything that is directly related to
its use is put forward as a reconciliatory means and as a key tool in the study of
English in general.
Before the universality of the application of corpora and corpus-analytic
methods in English Studies will be illustrated by means of a few case studies from
the individual subfields, however, I shall provide an overview of some central
steps in the corpus-analytic exploration of text.
2 Corpus Analysis: Some Basic Techniques
As mentioned above, a corpus is essentially a computerised collection of text used
for linguistic analysis, no more, no less. With the help of available software
programs, this text collection on the computer can be accessed in a number of
different ways (cf. Barlow 2004).
In corpus linguistics access to the texts stored in a corpus is enabled by socalled
concordance programs or concordancers, i.e. software packages which
provide a range of functions to analyse a collection of texts, the most important
83
one being the concordance function (exemplified below). Without such
concordance programs, corpora would be of no use other than being repositories
of texts – texts that could then be read on screen in the normal linear fashion. But
what exactly does this corpus analysis software do? In Hunston and Francis'
(2000: 15) terms, it "selects, sorts, matches, counts and calculates". With
reference to one particular software package, WordSmith Tools (Scott 1999), I
will illustrate how this is done, and present some basic techniques in accessing
and analysing corpora.1
Following Barlow (2004: 205), I will regard corpus or text
analysis as corpus or text transformation. I will try to show in what ways different
types of transformation can draw attention to different aspects of a text.
The corpus that will be used to illustrate these basic analytic (or
transformational) procedures consists of only one text: T.S. Eliot's poem "The
Love Song of J. Alfred Prufrock" (henceforth "Prufrock"). The first ten lines of
the poem are displayed in figure 1. Admittedly, this is a tiny corpus and a very
specialised one too, but it will fully serve our illustratory purpose.
Let us go then, you and I,
When the evening is spread out against the sky
Like a patient etherized upon a table;
Let us go, through certain half-deserted streets,
The muttering retreats
Of restless nights in one-night cheap hotels
And sawdust restaurants with oyster-shells:
Streets that follow like a tedious argument
Of insidious intent
To lead you to an overwhelming question...
Figure 1: Lines 1-10 of T.S. Eliot's poem "The Love Song of J. Alfred Prufrock"
1
Another well-known concordance package used in corpus analysis is MonoConc Pro (Barlow
2000). For a comparison of features of MonoConc Pro and WordSmith Tools, readers are referred
to a review published in Language Learning & Technology (Reppen 2001), available online at
http://llt.msu.edu/vol5num3/review4/default.html (consulted: Oct. 3rd 2004).
84
I downloaded "Prufrock" from the web, more precisely from the Project
Gutenberg website,2
and saved it on the harddrive of my computer in text file
format as 'prufrock.txt'. This text file could then be analysed with the software
package WordSmith Tools. The analytic procedure was carried out in five steps.
Step 1: Word Listing and Counting – Tearing the Text Apart
As Barlow (2004: 207) notes, "[p]robably the most radical transformation of a text
used in linguistic analysis is to, in effect, rip it apart to produce a wordlist." This is
what I did with Eliot's dramatic monologue in the first corpus-analytic step. Using
WordSmith's 'WordList' tool, the statistics of the poem were determined and a
frequency wordlist was created (cf. figure 2).
We can see in the right-hand screenshot in figure 2 that our mini-corpus
consists of 405 different words (types) which altogether occur 1,063 times (i.e. the
corpus contains 1,063 tokens). We also find further statistical information on
average word and sentence lengths and on the numbers of 1-, 2-, 3-, etc. letter
words in the poem. Perhaps more interesting than statistics, however, is the
frequency ranking of the words found to occur in the corpus. The left-hand part of
figure 2 shows that, with 74 occurrences, the is the most frequent word in
"Prufrock" and that the word makes up 6.96% of the whole text. Also particularly
common in the text are and, I, a, to, have, that, and of.
2
Project Gutenberg produces free electronic texts on the internet. More than 12,000 e-books are
currently available for download from http://www.gutenberg.net/ (consulted: Oct. 3rd 2004). A
large number of literary texts is also available from the Oxford Text Archive at
http://ota.ahds.ac.uk/ (consulted: Oct. 3rd 2004).
85
Figure 2: Frequency-ordered wordlist based on T.S. Eliot's "Prufrock" (left) and
statistics of the poem (right)
In the analysis of such a wordlist we now have two options. One option is
to analyse the wordlist manually, i.e. to simply scroll down the list, in order to see
which words, in particular which content words (such as nouns, verbs, and
adjectives), are especially common, and hence to trace some peculiarities of the
underlying text. The second option is to compare this wordlist, which is based on
a very specialised corpus (the poem "Prufrock" by T.S. Eliot), with another
wordlist derived from a more general so-called reference corpus, i.e. a corpus that
is considerably bigger and to a high degree representative of written English in
general. In the WordSmith Tools program suite this can be done by means of the
'compare 2 wordlists' command in the WordList tool. As a result, we get a list of
keywords in the smaller corpus, ordered by their 'keyness' in the text as compared
to the selected reference corpus, which was in our case the 90 million-word
written part of the British National Corpus (BNC_written).
86
Figure 3 displays the thus retrieved keywordlist of "Prufrock" with the
most outstanding words topping the list. Even though they are very frequent in the
poem (cf. figure 2), words like the, a, to, and of do not appear in this list. This is
because these words are generally extremely common in written English and not
key items of Eliot's poem. The items listed in figure 3 in order of keyness (I,
panes, have, dare, presume, upon, etc.) are all more frequent in "Prufrock" than
we would expect them to be on the basis of general language use. Knowing which
words in a text, or in a collection of texts, are unexpectedly or unusually frequent
can help us determine what the text is concerned with. The compilation of a
keywordlist like the one given in figure 3 is hence a very useful step in the
characterisation of any text or text collection.
Figure 3: Keywordlist based on T.S. Eliot's "Prufrock" and BNC_written as
reference corpus
Step 2: Tracing the Repeated Occurrence of an Item in a Text –
Examining Dispersion Plots
If we do not only want to know what the most common (or the most unusually
common) words in our corpus are but also where they occur in the text, we can
87
make use of the 'dispersion plot' function in WordSmith Tools (comparable to the
'distribution of hits' function on MonoConc Pro). This function serves to visualise
the distribution of a selected item across a text.
Figure 4: Dispersion plot of time in T.S. Eliot's "Prufrock"
Figure 4 shows the dispersion plot of the most frequent noun, the noun
time, in "Prufrock".3
Each of the eleven occurrences of the word in the text is
represented by a thin black line in the 'plot' window. The first line in the plot
column (on the left) indicates the beginning, the last line on the right indicates the
end of the text. As we can see, time clusters in the first third of "Prufrock". There
are no instances of the word at the very beginning of the poem and none in the
second half of it. We can therefore infer that time only features in one or two
stanzas in the first third of "Prufrock" and that it is not a theme traceable
throughout the entire poem.
Step 3: Compiling a Concordance – Putting Words Back into Context
In the third analytic step we are moving towards the centre of each kind of corpuslinguistic
activity: the concordance. Having torn the text apart in step 1, we will
now reverse this process and provide a contextualised view of those items in our
corpus in which we are most interested (maybe some of the particularly common
or key words).
As fittingly stated by Barnbrook (1996: 65), "[t]he concordance
provides a simple way of placing each word back in its original context, so that
the details of its use and behaviour can be properly examined." A concordance of
3
Time is in fact not only the most frequent noun in "Prufrock" but also in English in general.
Hence the item does not appear in the keywordlist displayed in figure 3.
88
the word time based on our "Prufrock" corpus is given in figure 5. Generally, in
such a concordance all occurrences (tokens) of a word are shown in the context of
a concordance line, usually displayed in the so-called 'KWIC' (key word in
context) format, with the searchword (here time) in the middle of the screen and
some context on the left and on the right of it. In our case the concordance shows
all eleven instances of time in "Prufrock" in random order.
Figure 5: Unsorted concordance of time in T.S. Eliot's "Prufrock"
One major advantage of this concordance function is that we do not have
to read through the whole text if we want to see how a particular word is used. All
we need is one mouse-click and the required textual evidence is displayed on the
screen right in front of us. However, access to the usage patterns in such a
concordance display can be further optimised, as the next step in our analysis
shows.
Step 4: Sorting the Context in a Concordance – Uncovering Patterns
Concordance programs usually offer the possibility of sorting or re-sorting the
concordance, which means that the order of the displayed concordance lines is
rearranged according to certain predefined sorting criteria. In the analysis of
89
"Prufrock" I decided to sort the context in the time concordance alphabetically to
the left. The result of this sorting step can be seen in figure 6.
Figure 6: Left-sorted concordance of time in T.S. Eliot's "Prufrock"
In contrast to the unsorted concordance in figure 5, the sorted display in
figure 6 enables us to spot at a glance repeated word combinations in the text, e.g.
"and time" (three times) or "there will be time" (five times). Sorting to the right of
the searchword, which is of course also possible, would highlight different
patterns, such as "time for" (five times) and "time to" (four times). Access to
language patterns is absolutely central in all sorts of text analysis, since we can
only fully understand what is expressed by a word when we see it in combination
with other words. In Wittgenstein's terms, "the meaning of a word is its use in the
language" (Philosophical Investigations, §43). Granting us direct access to
patterns of language use, sorted concordances can thus help us determine how
meanings are made. The fifth and final corpus-analytic step that I would like to
demonstrate will further focus on the important issue of repeated combinations
and co-occurrences of words.
90
Step 5: Examining the Context of a Word – Looking for Collocations
There are two functions in the WordSmith 'Concord' tool (the tool I used to create
the time concordance) that serve to highlight co-occurrence patterns in texts: the
'show collocates' and the 'clusters' function. By collocates we mean words that
occur near the searchword, usually in a contextual span of four or five words to
the left and to the right of it.
Figure 7: Collocates of time in T.S. Eliot's "Prufrock"
The activation of the 'show collocates' function, based on the concordance
of time, retrieves the collocation plot given in figure 7. The different columns list
the most common collocates of time and their numbers of occurrence on either
side of the searchword, split up according to (in our case) five positions to the left
(L1 to L5) and five positions to the right (R1 to R5). This display of numbers tells
us, for instance, that of the six co-occurrences of time and will, will is found five
times in L2 position, that is two places on the left of time as in "there will be
time". The 'clusters' function facilitates access to such collocation patterns in that
it automatically extracts all recurrent word combinations from a concordance that
matches a certain predefined cluster span. I defined a cluster span of '3' to extract
all three-word clusters that occur at least twice in the time concordance (see figure
8).
91
Figure 8: Repeatedly occurring three-item clusters based on the concordance of
time in T.S. Eliot's "Prufrock"
Also important in this context of investigating typical collocations and cooccurrence
patterns of words is the notion of 'semantic prosody'. When we look at
the items on the right-hand side of time in figure 5, we note a number of words
and phrases with rather negative connotations. There is time "to murder", "for a
hundred indecisions", "for the yellow smoke", or time "to turn back and descend
the stair". These words and phrases all contribute to the meanings of time in
"Prufrock". The, in this case, negatively connotated meaning they create is called
semantic prosody. Sinclair (2003: 117) refers to semantic prosodies as the
"[h]idden meanings" of words, which are "essential for the understanding of
language text" (ibid.) and which can be uncovered by means of corpus-analytic
techniques (cf. also Bublitz 1996, Esser 1999, Louw 1993, and Partington 2004).
To sum up, the central steps in the analysis and thus in the transformation
of a corpus are: counting and listing words, tracing their distribution across texts,
compiling concordances, sorting concordances, and searching for collocations and
patterns (including the determination of semantic prosodies).
92
3 The Use of Corpus-Analytic Techniques in English Studies
3.1 Corpus Analysis and Linguistics
Probably the most obvious field of application of the outlined steps in corpus and
text analysis is linguistics. Linguists study language, its production, its structure,
and its use, and they need a certain amount of language data on which to base
their research. Corpora therefore present welcome tools in all traditionally
distinguished fields of linguistics, be it phonology, morphology, lexicology,
syntax, semantics, pragmatics, historical linguistics, sociolinguistics, or
psycholinguistics.4
To provide a rough idea of what kinds of insights the use of
corpora and corpus-analytic techniques may provide in linguistics, I shall now
briefly sketch a research project I carried out on English modal verbs (for details
cf. Klages & Römer 2002 and Römer 2004a).
The project centres on the use of the central English modals (can, could,
may, might, will, would, shall, should, must, and ought to) in spoken British
English. Two key questions I dealt with in this project were, "How are different
modal verbs distributed?" and "Which meanings do they express?" Of major
interest in tackling both questions were frequencies of occurrence of the modals
and their meanings in actual language use. The first step in my analysis thus
involved some word counting. With the help of a concordancer I determined the
frequencies of the above-listed modal verbs in the spoken component of the
British National Corpus (including their negative forms and excluding instances in
which the items do not function as modals, e.g. can in "I'll put a can of that juice
in") and established the frequency ranking displayed in figure 9. We can see in
this figure that in spoken British English will, would, and can are by far most
4
For overviews on the use of corpora in various linguistic sub branches the reader is referred to
the following introductory corpus-linguistic textbooks: Biber, Conrad & Reppen 1998, Hunston
2002, McEnery & Wilson 2
2001, Meyer 2002, and Partington 1998.
93
common among all modal verbs and that verbs like may, must, or shall are
comparatively infrequent.
47,400
7,940
1,082
45,771
21,924
2,612
4,525
12,287
5,801
52,511
0
10,000
20,000
30,000
40,000
50,000
60,000
will would can could should might must may shall ought
to
Figure 9: Overall frequencies of the central modal verbs in BNC_spoken
As each modal can express different meanings, depending on the
immediate lexical context in which it occurs, I considered it important to move
one step further in the analysis, from overall frequencies to the distribution of
these different modal meanings. Again, a concordance program was consulted and
concordances were compiled in order to retrieve a large number of examples of
modal verbs in context. An extended context window in the concordances (of 200
characters per line) enabled me to attach a meaning label to almost every example5
and so to determine the shares of different functions for each modal. Figure 10
visualises the results of this meaning attribution in the case of may, a modal that
expresses possibility and permission, however not with equal shares.
As figure 10 shows, instances of may in which a speaker grants or asks for
permission (as in "May I ask a question?") are much less common in spoken
British English than occurrences of 'possibility' may (e.g. "It may well be that their
5
Mainly due to the often fragmentary nature of spoken English, there were a couple of cases in the
datasets which could not be clearly attributed to either of the defined meaning categories. These
examples were labelled 'unclear'.
94
expectation is wrong."). Besides, the manual analysis of the context in the sorted
concordance of may (and the other modals) gave access to co-occurrence patterns
and typical collocations. Among the things I observed were a rather high share of
if-clauses in the may concordance set, and a repeated occurrence of the cluster "if
I may".
4%
13%
83%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%
unclear
permission
possibility
Figure 10: The distribution of different meanings of may in BNC_spoken
These are just some of the findings I was able to retrieve using corpusanalytic
techniques in my approach to an object of linguistic study. Such
empirical findings are not only relevant for a systematic stock-taking of the
structure and use of a language, they can also have important implications for
language learning and teaching – a field that forms another part of the subject
'English' as it is taught and studied in a large number of English departments
across Europe. Before further investigating the impact of corpora and corpus
analysis on language pedagogy in section 3.3 of this essay, I will first turn to the
application of corpus-analytic procedures in the study of literature.
3.2 Corpus Analysis and Literature
The aim of this section is not to give an introduction to or provide an overview of
work done in the field of stylistics. Rather than asking questions like "What is
style?" and "How can it be captured in linguistic terms?", I will address the
question "How can corpus analysis help to shed light on aspects about an author's
95
style?"6
Taking for granted that linguistic approaches to literary language can be
fruitful and illuminating, I will try to exemplify the specific surplus value of
corpus-analytic techniques in the study of literature and show how the steps
described in section 2 of this paper can be applied to a corpus of literary works. I
wish to stress here that corpus analysis is not supposed to replace, but to
complement, traditional approaches in literary analysis.
The corpus in question consists of the six major novels written by the early
nineteenth century British novelist Jane Austen (the Jane Austen Corpus,
henceforth JAC) and was compiled by Katrin Oltmann as part of a university
research project in corpus linguistics.7
In the following I will largely draw upon
Oltmann (2004) and present selected results from her study on irony in Austen's
writing, complemented by findings from some additional queries based on JAC
and a larger corpus of eighteenth and nineteenth century English novels (the
Novels Reference Corpus, henceforth NRC).8, 9
Jane Austen and her Major Concerns – Keywords in JAC
As pointed out above (cf. section 2), the comparison of a frequency wordlist from
a small and specialised corpus with a frequency wordlist based on a larger
reference corpus can highlight items in the smaller corpus ('keywords') that are
unexpectedly common and that are likely to characterise the texts in this corpus.
The type of quick analysis that I carried out on keywords in Jane Austen's novels
could be used as a pre-reading activity or as a preliminary step in our, or in our
6
For further studies which use corpus approaches in the analysis of literary works, the reader is
referred to Krishnamurthy's (1995) article on the poem "Spring" by Philip Larkin, Stubbs' (2001)
chapter on "Eveline" by James Joyce (from Dubliners), his analysis of Conrad's Heart of Darkness
(this volume), and Sinclair's (forthcoming) paper on Alexander Pope's An Essay on Man.
7
JAC has an overall size of roughly 650,000 words (tokens) and includes the novels Emma,
Mansfield Park, Northanger Abbey, Persuasion, Pride and Prejudice, and Sense and Sensibility.
The texts were downloaded from the Project Gutenberg online archive (see
http://www.gutenberg.net/; consulted: Oct. 3rd 2004).
8
I would like to thank Katrin Oltmann for granting me permission to use JAC and NRC for this
study.
9
The texts in the Novels Reference Corpus (NRC) were also taken from Project Gutenberg. The
NRC has a size of roughly 4.4 million words.
96
students', interpretation of any literary work (on this issue see also Kettemann &
Marko 2004).
Figure 11: Keywords in JAC (items 1-20), sorted according to keyness
Figure 11 displays the first 20 items in a keywordlist based on JAC, sorted
by keyness values, with NRC used as reference corpus. A number of mainly
female first names (e.g. Fanny, Emma, and Elinor) top this list, which – given the
composition of the underlying corpus – is hardly surprising. In fact, if we scroll
further down the JAC keywordlist, we note that the majority of the items listed are
names of novel characters or places. This observation tells us something about
who Austen's heroines and heroes are and how often they are referred to in the
texts, but it does not reveal much about what her novels are actually concerned
with. In a further analytic step I have, therefore, extracted all keywords that are
not names and listed the first 40 items in table 1 (again in order of keyness). A
large number of these keywords can be summarised under four headings:
'female' items (e.g. her, she, Miss, herself, sisters),
97
modal verbs (could, must, might, would),
emphasisers/boosters (e.g. very, such, really, certainly, perfectly), and
'emotion' nouns (feelings, engagement, acquaintance, attachment, attentions).
From these categories it is possible to infer (still keeping in mind that this might
be a pre-reading activity) that Austen's novels deal with the world of women, with
issues related to obligation, permission, and probability, and with human
emotions.
1. her 11. must 21. soon 31. have
2. she 12. been 22. much 32. always
3. be 13. every 23. nothing 33. would
4. very 14. such 24. really 34. exactly
5. Miss 15. feelings 25. am 35. attachment
6. Mrs 16. do 26. think 36. certainly
7. not 17. being 27. might 37. had
8. could 18. quite 28. engagement 38. attentions
9. herself 19. any 29. sisters 39. sister's
10. it 20. was 30. acquaintance 40. perfectly
Table 1: Keywords in JAC, excluding names of characters
Particularly striking here is the high frequency of different kinds of
boosters (as opposed to hedges) and emphatic expressions. Since such expressions
often mark humour or instances of ironic language use (cf. Barbe 1995: 22, Kreuz
& Roberts 1995: 24, or Seto 1998: 241ff.), I will now briefly address this topic
and present some results from a corpus-driven study on irony in JAC (cf. Oltmann
2004).
98
Austen and Irony – Collocations, Semantic Prosodies, and Deviations
from the Norm
Instances of irony in texts are certainly not easy to spot, neither manually in a
traditional reading approach, nor by means of any computer software, and
especially not in older texts which lie outside our personal scope of experience:
Since irony is bound to intertextuality and social and stylistic conventions
of language use, it is often difficult to detect instances and correctly
interpret allusions and connotations, especially in historical periods (where
readers may also be misled by anachronistic interpretations based on
contemporary conditions). (Görlach 1999: 152)
Oltmann (2004: 45) also stresses this problem and notes that ironic instances,
which come in many different shapes, "are hard to find using the tools of corpus
linguistics because they follow few patterns, they are usually highly specific and
creative, infrequent and context-based". Nevertheless, Oltmann in her corpusdriven
study manages to establish several ironic devices and traces a number of
items that often function as the source of irony or humour.
Central in her analysis are concordances, and the key analytic step consists
in the comparison of the collocates and semantic prosodies of words in JAC on
the one hand, and the collocates and semantic prosodies of the same words in the
Novels Reference Corpus on the other hand. This comparison highlights how
Austen's language differs from the language of a larger group of her
contemporaries. In other words, Oltmann's analysis shows how Austen deviates
from the norm of eighteenth/nineteenth century novel writing and how she creates
meaning in her novels differently from contemporary authors.
The use of the word recover in JAC presents a nice example of this
deviation from the norm, or as Muka ovský (1970: 42) calls it, "violation of the
norm" (cf. also Barnbrook 1996, Louw 1993, and Louw 1997 on this issue).
While recover is normally used in rather negative contexts with unfavourable
semantic prosodies by Austen's contemporaries (see the concordance sample in
99
figure 12) – just as it is used in 'general' English today (people usually recover
from something unpleasant),10
Austen uses collocations such as, "the agitating
happiness of success", "the flutter of pleasure", or "a fit of good humour" in the
context of recover, collocations that run counter the general negative pattern and
thus create ironic or humorous effects.
Figure 12: The use of recover in NRC (a 10-line sample from the concordance)
Figure 13: The use of recover in JAC (a 10-line sample from the concordance)
Other unusual collocations that can be found in Austen's novels include "attack +
watch" as in example (1), "attack + gallantry and compliment" as in (2), and
"pleasure + mischief" as in (3).
10
Examples of such 'unpleasant' experiences in NRC are agony, delusion, depression, shock, and
horror. A concordance based on the written part of the British National corpus highlighted
collocates such as injury, strain, wounds, depression, and anorexia.
100
(1) "Oh! Do not attack me with your watch. A watch is always too fast or too
slow. I cannot be dictated to by a watch." (Mansfield Park)
(2) For a day or two after the affront was given, Henry Crawford had
endeavoured to do it away by the usual attack of gallantry and
compliment, but he had not cared enough about it to persevere against a
few repulses; (Mansfield Park)
(3) Indeed she had no taste for a garden; and if she gathered flowers at all, it
was chiefly for the pleasure of mischief – at least so it was conjectured
from her always preferring those which she was forbidden to take.
(Northanger Abbey)
Comparative cross-checks in JAC and NRC of randomly selected keywords from
the JAC list also helped to identify a group of words that almost always express
irony in Austen's novels, though usually not in the novels of her contemporaries.
The semantically related words fortune, glory, heroine, heroism, honour, and
bustle belong to this group (cf. (4) and (5) for illustrative examples from JAC).
(4) This was all very promising; and, but for such an unfortunate fancy for
having his hair cut, there was nothing to denote him unworthy of the
distinguished honour which her imagination had given him; the honour, if
not of being really in love with her, of being at least very near it, and saved
only by her own indifference – (for still her resolution held of never
marrying) – the honour, in short, of being marked out for her by all their
joint acquaintance. (Emma)
(5) But from fifteen to seventeen she was in training for a heroine; she read all
such works as heroines must read to supply their memories with those
quotations which are so serviceable and so soothing in the vicissitudes of
their eventful lives. (Northanger Abbey)
101
Louw (1997: 244) observes that "the difference between the norm and
features of the text is responsible for many of the 'devices' which give the reader
so much pleasure", and indeed, instances of irony and humour in Jane Austen's
novels certainly belong to these 'devices'. The corpus-driven identification of
norm deviations in literary texts, however, can do more for readers and text
analysts than indicate sources of reading pleasure or confirm intuitions about
peculiar passages of a text. I would argue that the type of corpus-analytic
exploration described in this essay can help us answer the central question how
meaning is constructed in literary texts, and in what ways this construction of
meaning differs from normal everyday language.
Surely, meanings that are created in literature are often rather subtle and
not directly visible to the naked eye of the literary scholar or the student of
literature (the proverbial 'reading between the lines' is not normally a
straightforward activity), or, quoting Louw again,
[t]he belief is gradually becoming firmer that the question of meaning can
no longer be settled at first sight during the act of reading, especially
where first sight refers to a reading which is 'unassisted' by data. (Louw
2004: 1, my emphasis)
This is where corpora and concordance programs can help. As I hope the study of
Austen's novels has shown, the skilful use of corpus-analytic techniques can
contribute to making the invisible visible, and subtleties in literary language more
noticeable.11
What we need to do in the analysis and interpretation of literary texts
is pay more attention to patterns and collocations in 'normal' natural language and
compare them to the collocations and patterns in the more specialised literary
language. That means that an investigation of the special or deviating patterns in a
literary text should ideally be preceded by "a clear and comprehensive account of
the established meaning-bearing patterns of the language" (Sinclair forthcoming).
Larger reference corpora can help us identify such patterns with some precision.
11
See also Stubbs' argumentation, this volume.
102
3.3 Corpus Analysis and Language Pedagogy
Having exemplified the use of corpora and corpus-analytic tools in linguistics and
literary studies, I will now demonstrate how language pedagogy can also profit
greatly from corpus work. I argue that students of English language didactics,
most of them future teachers of English as a foreign language, should be
acquainted with the basic steps in corpus analysis and that they should be familiar
with some central direct and indirect applications of corpora in language teaching.
By 'direct application' I mean the use of corpora and concordances in the
classroom. This approach is usually referred to as 'data-driven learning' (DDL). In
a nutshell, the idea behind DDL is that learners are presented with corpus data, the
exploration of which will lead them to discover regularities (and oddities) about
the language they are learning. In the following I shall not, however, elaborate on
this direct approach, which features learners as researchers, but provide an
example of the indirect use of corpora in language teaching, which focuses on the
pedagogical application of corpus research findings.
I made extensive use of the tools and methods of corpus analysis in a study
of if-clauses in spoken British English and so-called 'school' English (cf. Römer
2004b). Starting from the observation that their use often presents a significant
problem to learners, even on an advanced level, I examined whether if-clauses are
treated differently in EFL teaching materials than they are actually used in natural
spoken English. Of particular interest in this analysis was the sequence of tense
forms, since this appears to cause particular difficulties for learners. I, therefore,
determined the respective tense form sequence in altogether 211 if-clauses in a
small and specialised corpus of spoken-type texts (dialogues, interviews, etc.)
taken from two best-selling German EFL coursebook series and compared my
findings to two if-clause datasets of similar size retrieved from the spoken part of
the British National Corpus (BNC_spoken).
103
23.7%
39.0%
25.1%
8.6%
17.1%
6.6%
0.0%
52.6%
8.6%
18.7%
0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 55%
SimplePres - SimplePres
SimplePres - PresTenseModal+Inf.
SimplePast - PastTenseModal+Inf.
PastPerfect - would+have+PastPart
Other Combinations
BNC_spoken EFL coursebook corpus
Figure 14: The distribution of if-clauses across tense form sequences in an EFL
coursebook corpus and in BNC_spoken
Figure 14 illustrates the distribution of the four most common tense form
combinations in the coursebook if-clauses and compares the determined shares to
the respective results based on BNC_spoken data. We note some significant
distributional differences between the two types of corpora. There is a clear
overuse in EFL coursebook if-clauses of the three combinations that are usually
introduced in the following way (on the basis of invented examples) and typically
referred to as if-clauses 'type 1', 'type 2', and 'type 3' in the teaching materials:
simple present (if part) – present tense modal (usually will) + infinitive (main
part), as in "If I win the lottery, I will buy a house in Tuscany."
simple past (if part) – past tense modal (usually would) + infinitive (main
part), as in "If I won the lottery, I would buy a house in Tuscany."
past perfect (if part) – past tense modal (would) + have + past participle (main
part), as in "If I had won the lottery, I would have bought a house in Tuscany."
On the other hand, we observe a significant underuse of the 'simple present –
simple present' sequence and of a number of other combinations that frequently
occur in if-clauses in natural speech.
104
On the basis of these and other attested differences between real spoken
English and 'school' English, I made a number of suggestions in what ways
teaching materials could be changed so that they mirror more closely how
language is really used, and not how writers' materials imagine it to be.12
This
study shows that corpus analysis can contribute to an improvement of teaching
materials and that corpora can play a vital role in language pedagogy.
4 Conclusion: The Importance of Corpus Literacy in English
Studies
In the introduction to this chapter I formulated the questions, "Can corpus analysis
help to improve interdisciplinarity in English Studies?" and "Can corpora
contribute to a reconciliation of literature, linguistics, and pedagogy?" By means
of applying some central corpus-analytic steps in these different branches of our
discipline, I hope to have shown that the answer to both questions has to be a
definite "Yes."
I regard corpus-analytic techniques as multi-purpose strategies with an
immense potential to enhance all sorts of textual analysis and to confirm or
contradict our intuitions about patterns and meanings in literary and non-literary
language. Since we all work with texts in some way or other, no matter whether
we teach if-clauses, interpret Jane Austen's novels, or investigate modal verbs, it
appears only natural to share some of the strategies we use, thus assigning
universal status to them. Figure 15 aims to illustrate the universality of corpusanalytic
text exploration in the different intersecting sub-branches of English
Studies. I see corpus analysis as a key skill that ought to reach all parts of the
discipline as it can be equally applied in linguistics, literary studies, and language
pedagogy.
12
A similar comparative analysis, though on a considerably larger set of data, was also carried out
on English progressive forms (cf. Römer 2005 and forthcoming).
105
Figure 15: Corpus analysis in English Studies
Getting back to the perspective of our students, I think we have to consider
what they need in order to succeed in their studies. What should they be equipped
with so that they will be able to handle problems from different parts of the
subject? My suggestion would be that we equip our students with a tool box,
containing skills that are transferable from problem to problem across subdisciplines.
One of these skills should relate to the knowledge of making use of
corpora and software for corpus analysis. I would like to call this particular skill
'corpus literacy' and regard it as a key literacy in English Studies.13
The acquisition
of a corpus literacy including a certain amount of 'hands-on' experience in
computer-aided text analysis may open new perspectives for students. It can make
them more independent and show them that, even at an early stage in their studies,
they can be researchers themselves. Also, as convincingly discussed by
13
In the context of English language teaching, Mukherjee (2002: 153) also mentions "corpus
literacy" as an important skill that pupils have to acquire if they are to profit from the use of
corpora and corpus-analytic software in the foreign language teaching classroom.
106
Kettemann and Marko (2004), the use of corpora and corpus-analytic techniques
in English Studies helps to strengthen students' language awareness, in particular
their discourse awareness, i.e. it enables them to distinguish more clearly between
different types of texts or discourses within the discipline.
Figure 16: The use of reconciling in written English (sample from a
BNC_written concordance)
If we look at the concordance sample in figure 16, we note that in general
English language use the word reconciling carries a rather negative semantic
prosody and that "difficulty in reconciling" and "problem of reconciling" count
among the main collocations. Despite the unfavourable context in which
reconciling apparently tends to occur, I believe that there are possible ways of
reconciling English Studies, and, in fact, we find evidence for this in lines 16 to
18 of the displayed concordance. Like Mair (2002: 126) I am convinced that
corpora, defined in a wide sense as computerised text collections, and corpus tools
can be beneficially used in all parts of our discipline which would then "pave the
107
way for philology, the scholarly analysis of texts, to re-establish itself in the
digital age". If this can be achieved, a reconciliation of the different facets of
English Studies cannot be far.
References
Barbe, Katharina. Irony in Context. Amsterdam: Benjamins, 1995.
Barlow, Michael. MonoConc Pro. Houston: Athelstan, 2000.
––. "Software for Corpus Access and Analysis" in How to Use Corpora in
Language Teaching, ed. John McH. Sinclair. Amsterdam: Benjamins, 2004.
pp. 205-21.
Barnbrook, Geoff. Language and Computers. Edinburgh: Edinburgh University
Press, 1996.
Biber, Douglas, Susan Conrad & Randi Reppen. Corpus Linguistics. Investigating
Language Structure and Use. Cambridge: Cambridge University Press,
1998.
Bublitz, Wolfram. "Semantic Prosody and Cohesive Company: 'Somewhat
Predictable'," Leuvense Bijdragen (Leuven Contributions in Linguistics and
Philology) 85/1-2 (1996), 1-32.
Esser, Jürgen. "Collocation, Colligation, Semantic Preference and Semantic
Prosody: New Developments in the Study of Syntagmatic Word Relations"
in Words, Lexemes, Concepts – Approaches to the Lexicon: Studies in
Honour of Leonhard Lipka, ed. Wolfgang Falkner & Hans-Jörg Schmid.
Tübingen: Narr, 1999. pp. 155-65.
Görlach, Manfred. English in Nineteenth-Century England. An Introduction.
Cambridge: Cambridge University Press, 1999.
Hunston, Susan. Corpora in Applied Linguistics. Cambridge: Cambridge
University Press, 2002.
Hunston, Susan & Gill Francis. Pattern Grammar. A Corpus-driven Approach to
the Lexical Grammar of English. Amsterdam: Benjamins, 2000.
108
Kennedy, Graeme. An Introduction to Corpus Linguistics. London: Longman,
1998.
Kettemann, Bernhard & Georg Marko. "Can the L in TALC Stand for Literature?"
in Corpora and Language Learners, ed. Guy Aston et al. Amsterdam:
Benjamins, 2004. pp. 169-93.
Klages, Monika & Ute Römer. "Translating Modal Meanings in the EFL
Classroom" in Language: Context and Cognition: Papers in Honour of
Wolf-Dietrich Bald's 60th
Birthday, ed. Sybil Scholz et al. Munich:
Langenscheidt-Longman, 2002. pp. 201-16.
Kreuz, Roger J. & Richard M. Roberts. "Two Cues for Verbal Irony: Hyperbole
and the Ironic Tone of Voice," Metaphor and Symbolic Activity 10/1 (1995),
21-31.
Krishnamurthy, Ramesh. "The Macrocosm and the Microcosm: The Corpus and
The Text" in Linguistic Approaches to Literature: Papers in Literary
Stylistics [Discourse Analysis Monograph 17], ed. Jonathan Payne.
Birmingham: The University of Birmingham, 1995. pp. 1-16.
Louw, Bill. "Irony in the Text or Insincerity in the Writer? The Diagnostic
Potential of Semantic Prosodies" in Text and Technology. In Honour of John
Sinclair, ed. Mona Baker et al. Amsterdam: Benjamins, 1993. pp. 152-76.
––. "The Role of Corpora in Critical Literary Appreciation" in Teaching and
Language Corpora, ed. Anne Wichmann et al. London: Longman, 1997. pp.
240-51.
––. "Truth, Literary Worlds and Devices as Collocation." Presentation abstract
of a keynote lecture held at the 6th
Teaching and Language Corpora (TaLC
6) conference, 6-9 July 2004, Granada, Spain.
Mair, Christian. "Empowering Non-native Speakers: The Hidden Surplus Value
of Corpora in Continental English Departments" in Teaching and Learning
by Doing Corpus Analysis, ed. Bernhard Kettemann & Georg Marko.
Amsterdam: Rodopi, 2002. pp. 119-30.
McEnery, Tony & Andrew Wilson. Corpus Linguistics. 2nd
ed. Edinburgh:
Edinburgh University Press, 2001.
Meyer, Charles F. English Corpus Linguistic: An Introduction. Cambridge:
Cambridge University Press, 2002.
109
Muka ovský, Jan. "Standard Language and Poetic Language" in Linguistics and
Literary Style, ed. Donald C. Freeman. New York: Holt, Rinehart and
Winston, 1970. pp. 40-56.
Mukherjee, Joybrato. Korpuslinguistik und Englischunterricht. Eine Einführung.
Frankfurt: Lang, 2002.
Oltmann, Katrin. "The Pleasure of Mischief: A Corpus-Driven Approach to Irony
and Humour in the Novels of Jane Austen." Unpublished research paper,
University of Cologne, Germany, 2004.
Partington, Alan. Patterns and Meanings: Using Corpora for English Language
Research and Teaching. Amsterdam: Benjamins, 1998.
––. "'Utterly Content in Each Other's Company': Semantic Prosody and
Semantic Preference," IJCL 9/1 (2004), 131-56.
Reppen, Randi. "Review of MonoConc Pro and WordSmith Tools," Language
Learning & Technology 5/3 (2001), 32-6.
Römer, Ute. "A Corpus-driven Approach to Modal Auxiliaries and their
Didactics" in How to Use Corpora in Language Teaching, ed. John McH.
Sinclair. Amsterdam: Benjamins, 2004a. pp. 185-99.
––. "Comparing Real and Ideal Language Learner Input: The Use of an EFL
Textbook Corpus in Corpus Linguistics and Language Teaching" in
Corpora and Language Learners, ed. Guy Aston et al. Amsterdam:
Benjamins, 2004b. pp. 151-68.
––. Progressives, Patterns, Pedagogy. A Corpus-driven Approach to
Progressive Forms, Functions, Contexts and Didactics. Amsterdam:
Benjamins, 2005.
––. "Looking at looking: Functions and Contexts of Progressives in Spoken
English and 'School' English" in The Changing Face of Corpus Linguistics,
ed. Antoinette Renouf & Andrew Kehoe. Amsterdam: Rodopi, forthcoming.
Scott, Michael. WordSmith Tools (Version 3.0). Oxford: Oxford University Press,
1999.
Seto, Ken-ichi. "On Non-echoic Irony" in Relevance Theory: Applications and
Implications, ed. Robyn Carston & Seiji Uchida. Amsterdam: Benjamins,
1998. pp. 239-55.