28

Corpus Linguistics and

Metaphor
Elena Semino

28.1 Introduction: The Relevance of Corpus Linguistics

to the Study of Metaphor

Research on metaphor increasingly exploits, in a variety of ways and for

a variety of purposes, the computer-aided methods developed within cor-
pus linguistics. In this chapter, I begin by briefly introducing corpus
linguistics and its relevance to the study of metaphor. In section 28.2,
I introduce the different types of corpora and corpus-linguistic methods
that can be applied to the study of metaphor. In section 28.3, I discuss the
different types of contributions that corpus methods have made to meta-
phor theory and analysis. In section 28.4, I provide a concrete example of
corpus-based research on metaphor, focusing on the use of metaphor to
express the experience of pain. In section 28.5, I provide some concluding
remarks.
Corpus linguistics involves the construction of large digital collections
of authentic texts (corpora) and their investigation through dedicated
software tools (e.g. McEnery and Hardie 2012). The methods of corpus
linguistics were initially primarily applied to the study of lexis and gram-
mar, but have recently been extended to a wider range of areas, including
discourse analysis, translation studies and (first and second) language
acquisition, as well as other branches of the humanities and social
sciences. Why and in what ways, however, are corpus methods relevant
to the study of metaphor?
In principle, the understanding of any linguistic phenomenon can ben-
efit from being systematically analyzed in large quantities of naturally
occurring data, that is, from the kind of analysis that corpus methods
make possible. This applies particularly to any phenomenon that is
claimed to be frequent in language, and that is given center stage in theory-
making at least in part because of its frequency. Metaphor is such
a phenomenon, particularly as it is viewed within cognitive linguistics.

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 464 SEMINO

Lakoff and Johnson’s (1980a) Conceptual Metaphor Theory (CMT) based

its claim that metaphor is central to thought on the pervasiveness and
conventionality of metaphor in language. In Lakoff and Johnson’s
Metaphors We Live By, and in many subsequent CMT studies, linguistic
examples are the main or sole type of evidence that is provided for the
existence of particular conceptual metaphors as mappings between
‘source’ and ‘target’ domains in conceptual structure. Subsequent devel-
opments of CMT, as well as some alternative theoretical accounts of
metaphor in cognition, are also founded, at least in part, on linguistic
evidence, such as Grady’s (1997a) theory of primary metaphor and
Fauconnier and Turner’s (2002) account of metaphor in Blending Theory.
The same applies to much work on metaphor across languages and cul-
tures, and to claims about the ‘universality’ or otherwise of particular
metaphors (e.g. Kövecses 2005).
The linguistic evidence provided in early work within CMT, however,
consisted of decontextualized examples invented or remembered by the
authors themselves. In contrast, over the last few decades, research on
conceptual metaphors has increasingly made use of authentic data, includ-
ing by studying patterns of metaphor use in a variety of electronic corpora.
Later in this chapter I describe in detail the theoretical contributions that
this change of methodology has made possible.
The CMT account of metaphor has also inspired, more or less directly,
a large amount of research on the frequencies, forms, and functions of
metaphor in particular texts, text types, or discourses. This kind of work
varies in the extent to which it explicitly contributes to metaphor theory,
but is firmly based in the analysis of language use in context. Here corpus
methods tend to play an important role, often in combination with quali-
tative analysis, as I show below.

28.2 Different Corpora and Tools for the Corpus-based

Study of Metaphor

Before discussing the contribution made by corpus-based studies to meta-

phor theory and analysis, in this section I discuss the different kinds of
corpora and tools that have been exploited in the study of metaphor.

28.2.1 Types of Corpora

In corpus linguistics, a corpus is generally defined as:
[a] set of machine-readable texts which is deemed an appropriate basis on
which to study a specific set of research questions. The set of texts . . . is
usually of a size which defies analysis by hand and eye alone within any
reasonable timeframe. (McEnery and Hardie 2012: 1–2)

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 Corpus Linguistics and Metaphor 465

A wide range of corpora are used by metaphor scholars, depending on the

goals of the research. Some studies make use of large pre-existing general-
purpose corpora, such as the 15 million-word Italian Reference Corpus
(Deignan and Potter 2004), the 100 million-word British National
Corpus (e.g. Stefanowitsch 2006b) and the much larger Bank of English
Corpus (e.g. Deignan 2005). These studies tend to make or test general-
izations about metaphor use in a whole (national) language, such as British
English, or across two languages, such as English and Italian.
In contrast, much research on metaphor is concerned with the forms,
functions, and implications of metaphor use in particular texts or text
types. In the former case, corpus methods are applied to digital versions of
complete existing texts, such as the Bible and the Koran in Charteris-Black
(2004) or Sylvia Plath’s Smith Journal in Demjén (2015). When the focus is
metaphor use in a particular genre or register, the analysts may either
study a subsection of a larger existing corpus, or, more frequently, con-
struct their own corpus or corpora. Skorczynska and Deignan (2006) for
example, compare the use of metaphor in two specially constructed cor-
pora of business research articles (e.g. Management Science) and articles from
business periodicals (e.g. The Economist). Demmen et al. (2015) analyze the
use of V I O L E N C E metaphors in a dedicated corpus of interviews with and
online forum posts by patients with cancer, family carers, and healthcare
professionals. Similarly, L’Hôte’s (2014) study of the metaphors and narra-
tives of New Labour in the UK involved the construction of three compar-
able corpora of texts produced by different UK political parties, including
manifestos and leaders’ speeches. These specially constructed corpora are
seldom larger than a million words, but are nonetheless big enough to
allow generalisations for the relevant text types.
Finally, the extent of variation in the size of corpora exploited in meta-
phor research can best be appreciated by considering studies that lie at
opposite ends of the continuum. Steen et al.’s (2010) work, for example, is
based on the manual analysis of four samples from different sections of the
British National Corpus that are small enough to be manually annotated
for metaphor (approximately 50,000 words each). In this case, corpus
methods were only used to search the annotations themselves. At the
other extreme, Veale (2012) creatively used the whole World Wide Web
as a corpus in order to study, among other things, the forms and uses of
similes in English.

28.2.2 Types of Corpus Tools

Variation can also be observed in the specific corpus tools that are
employed to find and analyze occurrences of metaphor in corpora of
different sizes (see also Stefanowitsch 2006a: 2–6). In spite of continued
progress in automatic metaphor identification (e.g. Mason 2004, Berber-
Sardinha 2010, Neuman et al. 2013), most corpus-based studies of

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 466 SEMINO

metaphor involve searching the data for words or phrases that are likely to
be used metaphorically, or to occur in close proximity to relevant uses of
metaphor. ‘Concordancing’ tools provide each instance of the search term
in the corpus on a separate line, accompanied by the immediately preced-
ing and following co-text. Some studies involve the concordancing of
expressions that are likely to be used metaphorically at least some of the
time in the data, that is, vocabulary associated with a particular source
domain or vehicle grouping. Deignan (2005), for example, concordanced
animal terms such as rabbit and squirrel in the Bank of English to study the
realization of the Animal source domain in English. Semino et al. (2015)
concordanced words such as journey and path in a corpus of online posts by
people with cancer, in order to study the use of J O U R N E Y metaphors by this
particular group of patients.
Where the focus of the research is the metaphorical construction of
a particular topic or experience, on the other hand, the selected search
terms realize the target domain of the potential metaphors. Stefanowitsch
(2006b), for example, concordanced in the British National Corpus a set of
emotion words such as happiness and sadness in order to test and refine the
claims made about metaphors for emotions in CMT. In both types of
approaches, the co-text of each occurrence of the search term has to be
carefully scrutinized, first, in order to see whether the search term or some
surrounding words are in fact used metaphorically, and second, to identify
any patterns that are relevant to the goals of the study. When specialized
corpora are used, it is not uncommon for a small representative sample of
the data to be analyzed manually for relevant metaphorical expressions
first, and then for those expressions to be concordanced in the whole
corpus (e.g. Charteris-Black 2004).
Different approaches can be used to identify more open-ended sets of
metaphorical expressions. Some studies involve the concordancing of
‘signalling expressions’ (Goatly 1997) or ‘tuning devices’ (Cameron and
Deignan 2003), that is, words or phrases that are often found in close
proximity to metaphorical, or generally figurative, uses of language.
Cameron and Deignan (2003), for example, concordance expressions
such as sort of and like in order to investigate metaphor use in corpora of
difference sizes. Veale (2012) employed standard search tools to look for
structures associated with similes in the World Wide Web, including, for
example, ‘as. . . as.’
A different technique involves the exploitation of semantic annotation
tools such as the USAS tagger, which is part of the online corpus compar-
ison software Wmatrix (Rayson 2008, ucrel.lancs.ac.uk/wmatrix/).
The USAS tool annotates each word in a corpus with a tag for the main
semantic domains it belongs to. For example, the word war is allocated to
the semantic domain ‘Warfare, defence and the army; weapons.’ Where
a semantic domain in USAS is likely to correspond to an important meta-
phorical source domain in a corpus, this tool can be used to search for

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 Corpus Linguistics and Metaphor 467

open-ended sets of metaphor candidates. This enables the analyst to go

beyond the limitations of traditional lexical concordances (involving pre-
selected words or phrases; see Koller et al. 2008). For example, concordan-
cing the word war in a 500,000-word corpus of online writing by cancer
patients (Semino et al. 2015) provides all nine occurrences of that word in
the corpus, which can then be checked for metaphoricity. In contrast,
concordancing the ‘Warfare’ USAS tag in the same corpus results in 181
occurrences of words that share that tag. This leads to the identification of
a much wider set of war-related metaphors for cancer, including expres-
sions such as chemo veteran and invaders for cancer cells (see Demmen et al.
2015, Semino et al. 2015). In a study of metaphors for mental states in
Sylvia Plath’s Smith journal, Demjén (2015) uses the same technique to
search both for vocabulary that relates to potentially relevant source
domains (e.g. the USAS semantic domain ‘Darkness’) and for target-
domain vocabulary (e.g. the USAS semantic domain ‘Thought, belief’).
These kinds of studies also often tend to involve a first stage where a set
of relevant semantic domains is identified via a manual analysis of
a subsection of the corpus, and a second stage where these domains are
systematically investigated in the whole corpus by means of the semantic
tagger.
Some recent studies additionally incorporate ‘keyness’ analyses as part
of their methodology. In corpus linguistics, ‘keywords’ are words that are
used much more frequently (to a statistically significant extent) in one’s
corpus of data as compared with a (usually larger) reference corpus.
Similarly, for semantically annotated corpora, a key semantic domain is
a domain that is statistically significantly overused in one’s corpus of data
in comparison with a reference corpus (Rayson 2008). Particular expres-
sions or semantic domains do not have to be ‘key’ in a corpus to be of
interest for the purposes of metaphor analysis, as not all instances of that
expression/domain may be metaphorically used in the corpus under ana-
lysis, the reference corpus, or both (e.g. news data is always likely to include
frequent literal references to wars). However, when a metaphorically used
expression or domain is key in a corpus in statistical terms, strong claims
can be made about the dominance of that particular kind of metaphor in
that data. For example, in L’Hôte’s (2014) New Labour corpus, both the
words tough and strong and the USAS semantic tag ‘Tough/strong’ are key as
compared with a corpus of Labour texts from an earlier period. This
suggests that metaphors to do with physical strength are systematically
used in the New Labour corpus to counter the party’s earlier ‘soft’ image in
British politics. Similarly, some signaling devices for metaphoricity are
keywords in Partington, Duguid, and Taylor’s (2013) corpus of opinion
pieces in British newspapers, as compared with a corpus of reviews in
the Times Literary Supplement. This finding is used to support a more general
claim that humorous metaphors and similes are much more frequent in
the former corpus than the latter.

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 468 SEMINO

A few studies also make use of collocation tools for metaphor analysis.
The ‘collocates’ of a word are words that are used unusually frequently in
close proximity to that word. Collocations can be calculated on the basis of
different measures of statistical significance (see Brezina, McEnery, and
Wattam 2015). L’Hôte (2014), for example, considers the collocates of the
metaphorically used word tough in her New Labour corpus, in order to see
what policy areas and initiatives are described in these terms in her data
(e.g. the New Labour slogans tough on crime). Semino (2008) considers the
collocates of the adjective rich in the British National Corpus to test
Lakoff’s (1993) claim that the expression a rich life is evidence of the
existence of a conventional conceptual metaphor A P U R P O S E F U L L I F E I S
A BUSINESS.
Having discussed the types of corpora and tools that can be employed for
metaphor analysis, I now turn to the main contributions that this kind of
approach has made to the study of metaphor.

28.3 Contributions of Corpus-based Studies of Metaphor

Corpus linguistics is an approach to the study of language that is not

associated with any general or specific theory. However, the use of corpus
methods rests on the assumption that actual linguistic behavior is not only
worth studying systematically, but also needs to be accounted for by any
theoretical model of language. As such, corpus linguistics is particularly
consistent with the usage-based models of language that have been pro-
posed within cognitive linguistics. When it comes to metaphor in particu-
lar, corpus methods have been used to contribute to metaphor theory and
analysis in a number of ways. In this section, I start from some general
contributions, especially concerning CMT. I then consider the findings
that corpus methods have led to in studies that approach metaphor from
historical, cross-cultural, discoursal, and practical perspectives.

28.3.1 General Implications for Metaphor Theory

In the first book-length study of metaphor from a corpus linguistic per-
spective, Deignan (2005) provided evidence from the Bank of English
corpus to support the most general claims made in CMT, especially con-
cerning the frequency of conventional metaphorical expressions, and the
ways in which they form patterns that can be taken as evidence for
particular conventional conceptual metaphors. However, Deignan also
showed systematically how the de-contextualized linguistic examples
used in CMT are often not sufficiently representative of actual language
use to be cited as evidence for claims about conceptual metaphors.
In addition, Deignan used corpus methods to reveal patterns of metaphor
use that were not considered or accounted for in (conceptual) metaphor

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 Corpus Linguistics and Metaphor 469

theory, at least up to that point. These include, for example, some systema-
tic associations between particular source domains and linguistic meta-
phors belonging to different word classes, as well as the fact that some
words have different conventional metaphorical senses for different mor-
phological inflections (e.g. rock as a singular noun vs. rocks as a plural noun).
Deignan also found evidence for a greater influence of target domains in
metaphorical mappings than suggested by Lakoff’s (1993) Invariance
Hypothesis.
Some subsequent corpus-based studies have questioned or refined spe-
cific claims about conceptual metaphors. Semino (2008) finds no evidence
in the British National Corpus for Lakoff’s (1993) claim that the expression
rich life is a realization of the conceptual metaphor A P U R P O S E F U L L I F E I S
A B U S I N E S S . Rather, corpus evidence is provided for a different conven-
tional pattern that accounts for the expression rich life alongside other
metaphorical uses of rich, such as rich soil and rich culture. Stefanowitsch
(2006b) conducts a systematic analysis of the metaphorical expressions
that co-occur with emotion words in the British National Corpus. He
provides broad support for earlier claims about emotion metaphors in
CMT (Kövecses 2000). However, he also reveals patterns that were not
discussed in previous studies, and provides information about the frequen-
cies of different metaphors in British English, as represented in the corpus.
Stefanowitsch’s (2006b) study also empirically addresses the question of
whether there are emotion-specific metaphors, and what metaphors can
be described as such.
Studies such as Musolff (2006), in contrast, use corpus evidence to make
suggestions about the level of generality at which claims about conceptual
metaphors are best made. Musolff analyzes the use of metaphor in
a bilingual English–German corpus of news reports about the EU. He
finds that the use of metaphor in the corpus is best accounted for not in
terms of mappings involving broad conceptual domains such as Marriage,
but rather in terms of more specific ‘scenarios,’ such as End-of-honeymoon
and Adultery.
More generally, corpus-based studies of metaphor tend to reveal linguis-
tic patterns that are difficult to account for in terms of CMT, and of
metaphor theory more generally. For example, as mentioned above, the
singular noun rock tends to have a positive meaning when used metaphori-
cally (the rock on which society is built), while the plural rocks tends to have
a negative meaning (The marriage has been on the rocks for a while) (Deignan
2005: 158–59). Furthermore, some A N I M A L metaphors seem to occur only
as nouns (e.g. cow as a derogatory term); others only occur as verbs (e.g.
horsing around); yet others occur both as nouns and verbs, but not necessa-
rily with the same meaning (e.g. racist pigs and pigging out on food) (Deignan
2005: 153). In German, Boot (‘boat’) and Schiff (‘ship’) have similar literal
meanings but different conventional metaphorical uses: Boot is used in
expressions such as being ‘in the same boat’ or to describe a place as

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 470 SEMINO

having no space for newcomers; in contrast, Schiff tends to be used as part

of metaphorical descriptions of difficult enterprises (Zinken 2007).
Such findings from corpus-based studies have considerable theoretical
implications. In her 2005 book, Deignan comments that “[t]he data dis-
cussed so far suggest that each linguistic metaphor has a life of its own”
(Deignan 2005: 166). On the basis of patterns such as Boot vs. Schiff in
German, Zinken (2007) proposes the notion of ‘discourse metaphors’ –
form-meaning pairings that develop in communication as part of
a ‘conceptual pacts’ among interlocutors, and that constitute a mid-point
between novel metaphors and fully conventionalized ones. Other meta-
phor scholars explain these findings in terms of Dynamics Systems Theory
(e.g. Cameron 2011). From a dynamic systems perspective, the meanings
and functions of metaphorical expressions arise from the dynamic inter-
action of lexico-grammatical, semantic, cognitive, pragmatic, and affec-
tive factors in actual contexts of communication (e.g. Cameron and
Deignan 2006, Gibbs 2016a). Within this account, conceptual metaphors
are only one of the factors involved in the development of metaphorical
meanings, including conventional ones. Johansson Falck and Gibbs (2012)
specifically show the relevance of embodied simulations associated with
particular words as one of these interacting factors. Like Zinken (2007),
they consider a pair of words (road and path) that have similar literal
meanings, but different metaphorical meanings. Corpus evidence from
the British National Corpus shows that path is typically used metaphori-
cally to refer to ways of living, and often to suggest (potential) difficulties.
In contrast, road tends to be used metaphorically to describe purposeful
activities. Johansson Falck and Gibbs compare the corpus findings with the
mental imagery that people reported in a questionnaire about their experi-
ences with paths and roads. The triangulation of participant responses and
corpus findings suggest that people’s embodied experiences with the
entities referred to by path and road can explain the (different) metapho-
rical meanings that the two words have in the corpus.

28.3.2 Corpus-based Findings on Historical, Cultural, and Register

Variation in Metaphor Use
Corpus methods are also contributing to the study of variation in meta-
phor use in the history of a language, across languages, and in different
types of texts.
The availability of historical corpora of English in particular has led to
several diachronic studies of metaphor. As Tissari (2001: 238) puts it, these
studies show both “stability and change” in metaphor use over time.
Tissari (2001) compares some of the central metaphors for love as an
emotion in corpora representing, respectively, Early Modern English and
Present-Day English. An analysis of concordances for the string ‘love’ in
both corpora showed that the main broad source domains that are applied

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 Corpus Linguistics and Metaphor 471

to love appear to be the same (C O N T A I N M E N T , A M O U N T , and E X C H A N G E ).

However, some differences can be observed in how these source domains
are realized. For example, E X C H A N G E metaphors seem to be realized more
often by words associated with agriculture in the Early Modern English
corpus than in the Present-Day English corpus. This reflects a decrease in
the cultural centrality of agriculture over time. Similar findings arise
from studies that exploit the Historical Thesaurus of English to study
conceptual mappings across the history of the language. Alexander and
Bramwell (2014) show, for example, how the metaphorical description of
wealth in terms of large body mass (e.g. fat and the now archaic cob) has
decreased over time, as the positive association of looking well-fed have
waned. These studies often have broader theoretical implications. In a
diachronic analysis of metaphors for intelligence in the Historical
Thesaurus, for example, Allan (2008) finds evidence to support the impor-
tance of the metonymic basis of many conventional metaphors. Some of
her findings also question the claim that metaphors tend to map concrete
experiences onto abstract ones. For example, folk etymology may suggest
that dull as a description of lack of intelligence involves a concrete-to-
abstract mapping from the source domain of (lack of) physical sharpness.
However, historical evidence reveals that the word was first used for lack
of intelligence, and thus suggests that it is the more concrete physical use
that was derived from the one that relates to intellectual abilities (Allan
2008: 186).
Many corpus-based cross-linguistic studies of metaphor have built on
and extended the work of Kövecses and others on the relationship between
metaphor and culture, broadly conceived (e.g. Charteris-Black 2003,
Deignan and Potter 2004, Chung 2008). These studies also tend to reveal
both similarities and differences in the metaphors used in languages
associated with different cultures. The similarities often confirm the
embodied basis of many metaphors. The differences emphasize cultural
relativity, not least in perceptions of the body itself. Simó (2011), for
example, builds on Charteris-Black’s (2001) research on English to com-
pare the metaphorical use of blood in corpora of Hungarian and American
English. She finds that the main target themes for blood metaphors are the
same: E M O T I O N and E S S E N C E . However, there are differences in the con-
notations and frequencies of different metaphorical expressions involving
blood. For example, the expression in cold blood/hidegvérrel is always negative
in American English but equally split between positive and negative uses
in Hungarian. In Hungarian, hidegvérrel is also particularly frequent in
sports reports. More generally, Simó finds evidence of within-culture var-
iation by comparing blood metaphors in each section of each corpus.
Indeed, a large number of studies have used corpus methods to study the
forms and functions of metaphor in different text-types, including news
reports (e.g. Koller 2004, Musolff 2004), political speeches (Charteris-Black
2005), religious texts (Charteris-Black 2004), and so on. These studies often

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 472 SEMINO

tend to aim to answer questions that pertain to those text-types and their
contexts of use, rather than or in addition to furthering metaphor theory. For
example, Koller (2004) shows how the dominant S P O R T S and W A R meta-
phors in reports about the economy are both closely linked to each other and
a reflection of a sexist bias. Charteris-Black (2012) uses corpus methods to
analyze metaphorical expressions in a large corpus of interviews with people
with depression. Among other things, he shows that men and women tend
to use similar metaphors, but suggests that the few differences that do
emerge may have relevance in the therapeutic process. More generally,
these studies reveal the potential pitfalls involved in making generalizations
about metaphor in language and cognition without taking into account
variation at the level of speakers/writers, registers, genres, and discourses
(see also O’Halloran 2007, Deignan, Littlemore, and Semino 2013).
A more specific contribution of corpus-based studies of metaphor in
particular text-types is the systematic investigation of the functions of
metaphors in discourse. L’Hôte (2014) shows the way in which metaphor
is used to project a political party’s new identity in a corpus of New Labour
texts in the UK from 1997 to 2007. Partington, Duguid, and Taylor (2013:
131ff) concordance tuning devices in a corpus of reviews in British broad-
sheets and show the ways in which creative, humorous metaphors and
similes are used to entertain the reader. These expressions tend to involve
incongruous comparisons and proper names of various kinds, and to rely
heavily on the readers’ cultural knowledge (e.g. a sort of Goldilocks on crack
and an unexpected, juicy cross between meaty liquorice and Noggin the Nog’s bum).
Veale’s (2012) study of similes on the World Wide Web reveals, among
other things, how certain simile structures tend to be used creatively and
humorously. For example, the structure ‘about as. . . as. . .’ is shared by
many ironic similes, especially when the adjective following the first as is
positive (e.g. about as useful as a chocolate teapot).
Corpus-based approaches to particular figurative phenomena within spe-
cific genres can also have broad relevance for metaphor theory. This is the
case with Dorst’s (2015) work on similes in fiction and two recent studies on
metaphor clusters and mixed metaphors (Kimmel 2010, Semino 2016).
In particular, Semino (2016) shows how people often use the label ‘mixed
metaphor’ to describe the use, in close proximity, of metaphorical expres-
sions that actually involve the same broad source domain, such as the source
domain of Music in: to keep things on a different note, instead of the same record
getting played over and over again. This suggests that people are particularly
sensitive to contrasts between nearby metaphors when those metaphors are
similar enough to each other for their differences to be consciously noticed.

28.3.3 Corpus-based Approaches to Metaphor and Practice

Finally, some corpus-based approaches to metaphor have implications for
practice in different areas. Skorczynska and Deignan’s (2006) comparison

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 Corpus Linguistics and Metaphor 473

of a corpus of business-research articles and a corpus of business-

periodicals articles revealed differences that are relevant to the choice of
teaching materials for non-native speakers who are preparing for higher
education courses in English. Skorczynska and Deignan suggest that the
use of magazines such as The Economist is not sufficient to prepare students
for reading academic papers on business and economics. Skorczynska
(2010) similarly found that the metaphors in a Business English textbook
differ considerably, both in kind and in frequency, from those found in
a corpus of specialist business articles. These studies have implications for
the choice of teaching materials in English for specific purposes, as well as
for a further appreciation of variation in metaphor use depending on
audience and text-type.
A corpus-based study of metaphors for cancer and the end of life at
Lancaster University, in contrast, has implications for communication in
healthcare. The study involved a combination of manual and computer-
aided analysis of a 1.5 million-word corpus of interviews with and online
posts by patients with advanced cancer, family carers, and healthcare
professionals. A variety of patterns of metaphor use were identified by
means of a combination of lexical and semantic concordances. Among
other things, the analysis showed differences among patients and health-
care professionals in the use of V I O L E N C E metaphors such as fighting can-
cer, with patients using them significantly more frequently than
healthcare professionals. The study also showed that, as many have argued
(e.g. Sontag 1979), violence-related metaphors can be detrimental to
patients’ morale and self-esteem, as when a patient with a terminal diag-
nosis says I feel such a failure that I am not winning this battle. However, for
a substantial minority of patients, these metaphors can be empowering, as
when a patient says Cancer and the fighting of it is something to be proud of (see
Demmen et al. 2015, Semino et al. 2015). These findings suggest a degree of
individual variation in the use of metaphor for illness that is relevant to
communication training and practice in healthcare (see Demjén and
Semino 2017).
Overall, this section has shown the variety of contributions that corpus-
based approaches can make to the understanding of metaphor in language
and cognition. In the next section I provide a more extensive example of
corpus-based analyses of metaphor in relation to physical illness.

28.4 Corpus Approaches to Metaphor and Pain

The experience of physical pain is well known to be difficult to articulate

in language (e.g. Scarry 1987). English, for example, has relatively few
words that specialize in the description of pain sensations, and all of
them are quite general (e.g. pain, painful, hurt, ache, aching, sore). This poses
a challenge in healthcare, particularly when pain becomes chronic: in that

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 474 SEMINO

context, diagnosis, and treatment rely to a significant extent on the suf-

ferer’s ability to express their pain, and on healthcare professionals’ sen-
sitivity to patients’ descriptions. It is also well known, including among
clinicians, that figurative language is an important resource in articulating
pain sensations. In the words of a consultant at the National Hospital for
Neurology and Neurosurgery in London:
Attempts to truly describe pain indeed appear as difficult as they are
frustrating, yet the need to communicate is overwhelming, and I suggest
that the only option available is the resort to analogy . . . (w)hether by
means of metaphor or simile. (Schott 2004: 210)

Kövecses (2008) discusses the main metaphors used in English to convey

pain experiences, from the perspective of CMT. He observes that “pain is
conceptualized metaphorically in terms of its potential causes” (Kövecses
2008: 28). More precisely, pain sensations tend to be expressed metonymi-
cally or metaphorically in terms of causes of damage to the body. For
example, the expression a burning pain is used metonymically when the
pain is caused by skin contact with a flame, and metaphorically when the
pain does not result from contact with sources of heat, but can be com-
pared with the sensation caused by such contact (e.g. a burning pain caused
by acidity in the stomach). Kövecses (2008) proposes a number of specific
conceptual metaphors that are all part of this broad pattern:
PAIN IS A SHARP OBJECT A sharp stab of pain made her sit back down.
PAIN IS A TORMENTING ANIMAL A massive killing pain came over my right
eye . . . I clawed at my head trying to uproot the fiendish talons from their iron
grip.
P A I N I S F I R E Pain is fire that can devour the whole body.
(Kövecses 2008: 28; emphasis in original)
The connection with metonymy emphasizes the main motivation for such
metaphors. The most prototypical and intersubjectively accessible kind of
pain results from damage to bodily tissues, as in the case of cuts, burns, etc.
Other types of more subjective and invisible pain (e.g. migraine) are
described metaphorically in terms of properties or processes that cause
damage to the body (e.g. a splitting headache).
In Semino (2010) I discuss these metaphors from the perspective of
theoretical and empirical work on embodied simulation. I also point out
that language-based diagnostic tools for the diagnosis of pain also heavily
rely on linguistic metaphors that draw from a wide range of causes of
bodily damage. The McGill Pain Questionnaire (Melzack 1975), for exam-
ple, requires patients to choose appropriate descriptors from a list that
includes words such as stabbing, burning, drilling. In the same study, I also
provide corpus evidence to support and refine Kövecses’s (2008) claims
about conventional conceptual metaphors in English. The top sixty-five
collocates of the word pain in the British National Corpus include eight
expressions that have literal, basic meanings to do with causes of bodily

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 Corpus Linguistics and Metaphor 475

damage: searing, sharp, stabbing, lanced, seared, stabbed, stinging, burning.1

The corpus data provides evidence for a conventional conceptual meta-
phor that can be expressed as P A I N I S C A U S E O F P H Y S I C A L D A M A G E , which
can in fact be seen as a ‘primary’ metaphor, in the terms used by Grady
(1997a). The metaphors proposed by Kövecses (2008) for different kinds of
bodily damage can be seen as specific-level variants of this more general
and basic metaphor (see also Lascaratou 2007 for metaphors and metony-
mies for pain in Greek). In addition, an analysis of concordance lines for
pain shows that some expressions that are included in the McGill Pain
Questionnaire are never or seldom used in the corpus to describe pain
sensations (e.g. taut). This may cast some doubts on the decision to include
such expressions in the questionnaire.
Corpus methods can be further exploited to investigate how chronic
pain sufferers use figurative language creatively to express particularly
intense and distressing experiences. As an example, I searched for the
structure ‘feels like’ in a 2.9 million-word corpus of contributions to an
English language online forum for sufferers of Trigeminal Neuralgia.2
Trigeminal Neuralgia (TN) is a rare but distressing condition that causes
unpredictable pain in the face of such intensity that all aspects of the
sufferer’s life can be affected. The search for ‘feels like’ in the corpus
returned 254 occurrences of the expression. As I expected, most
instances function as a tuning device, that is, they introduce a simile
that describes the writer’s experience of a TN attack, such as the follow-
ing (NB: original spellings have been retained but proper names have
been changed):
(1) feels like needles poking my eyes and ears
(2) sometimes it feels like I am covered in insects and this can be more
distressing than the pain
(3) At the moment, it feels like someone of considerable weight (not you
jenny lol) is standing on the side of my face, wearing stilletto’s whilst
pouring red hot liquid into my ear. When I get the bolts it feels like
someone has decided that I deserve to be cattle prodded!

Example (1) is a relatively creative instantiation of the conventional tendency

to describe pain in terms of the insertion of sharp objects into the body (cf.
Kövecses’s 2008 P A I N I S A S H A R P O B J E C T ). Example (2) does not involve pain as
such, but an uncomfortable symptom of TN that is described as more distres-
sing than the pain itself. The expression covered in insects primarily conveys
a physical sensation, but also has associations of repulsion and disgust, which
can account for the distressing nature of the experience. Example (3) is one of
many examples in the corpus where different causes of physical damage

1
Collocates of pain were computed on the basis of log-likelihood (Dunning 1993) and within a window span of one
word to the left and one word to the right of the search string.
2
I am grateful to the UK Trigeminal Neuralgia Association for allowing me access to their online forum.

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press

 476 SEMINO

(pressure, penetration with a pointed object, heat, electric shock) are com-
bined into a rich, dynamic and creative metaphorical scenario (see also
Deignan, Littlemore, and Semino 2013: 279ff.). This kind of description goes
well beyond the single-word options included in language-based diagnostic
tools such as the McGill Pain Questionnaire.
A systematic analysis of conventional and creative figurative descrip-
tions of pain, in both general and specialized corpora, can therefore be
used both to assess the validity of existing pain questionnaires and to
inform the creation of future language-based diagnostic tools.

28.5 Conclusions

Corpus approaches have made and are continuing to make a variety of

important contributions to the study of metaphor as a linguistic and
cognitive phenomenon. These contributions involve not just metaphor
theory, but also the understanding of communication in a variety of
contexts, as well as practice in areas such as education and healthcare.
The chapter has also shown the ingenuity and methodological eclecti-
cism involved in corpus-based studies of metaphors. Researchers do
not just use a variety of corpus tools, but do so creatively, in order to
identify the widest possible variety of potential metaphorical expres-
sions in their data. An initial manual analysis often provides the
springboard for computer-aided analysis. Even more importantly, deci-
sions about metaphoricity and about the meanings and functions of
metaphors in the data involve detailed qualitative analyses of the
output of corpus tools.
Two final points are in order, which I have not been able to do justice to in
the course of the chapter. First, experimental approaches to the study of
metaphor (e.g. Bowdle and Gentner 2005, Casasanto 2008b) can benefit
from using as stimulus materials authentic and, ideally, frequent linguistic
expressions: corpora are a potentially useful source of such materials. Second,
any large-scale linguistic investigation will come across metaphorical uses of
language. Therefore, corpus-based studies of language generally can benefit
from, or possibly even require, the findings and insights that are developed by
metaphor scholars. It is no coincidence, for example, that metaphor is the
topic of one of the guides that followed the creation of the first corpus-based
dictionary of English, the Collins Cobuild dictionary (Deignan 1995).
No systematic corpus-based investigation of word meanings and discourse
patterns in any language can avoid dealing with the role of metaphor in the
lexicon and in language use.
Future corpus linguistic research on metaphor is likely to involve the
use of ever more sophisticated corpus and computational tools, the analy-
sis of larger and more varied corpora and further theoretical, analytical,
and practical advances.

https://doi.org/10.1017/9781316339732.029 Published online by Cambridge University Press