Anatol Stefanowitsch

Table 13-5b. LIQUID metaphors for happiness and joy

H APP N ESS W ORD

HAPPINESS JOY Total

FULLNES Obs: 24 Obs: 16 40
HOR

S Exp: 16,45 Exp: 23,55

Obs: 20 Obs: 47 67
M ET AP

OTHER
LIQUID Exp: 27,55 Exp: 39,45
44 63 107

is case study demonstrates how central metaphors for a given target domain can be identified by
searching for general words describing (aspects o ) the domain in question. It also shows that
these metaphors can be associated to different degrees with different words within a given target
domain. It is unclear to what extent this lexeme specificity of metaphorical mappings supports or
contradicts current cognitive theories of metaphor, so it is potentially a highly productive area
of research.

13.2.3 Metapftor and text

13.2.3.2 Case Study: Metaphoricity signals. Although metaphorical expressions are pervasive in all
registers and speakers do not generally seem to draw special a ention to their occurrence, there
is a wide range of devices that mark non- literal language more or less explicitly (as in
metaphorically/figuratively speaking, picture NP as NP, so to speak/say):

(13.6a) …the princess held a gun to Charles 's head, figuratively speaking… [BNC CBF]
(13.6b) He pictures eternity as a filthy Russian bathhouse… [BNC A18]
(13.6c) …the only way they can deal with crime is tofightfire, so to speak,withfire. [BNC ABJ]

Wallington et al. (2003) investigate the extent to which these devices, which they call metaphoricity
signals, correlate systematically with the occurrence of metaphorical expressions in language use.
ey find no strong correlation, but asthey note, this may well be due to various aspects of their
design. First, they adopt a very broad view of what constitutes a metaphoricity signal, including

361
13 Metaphor and Metonymy

expressions like a kind/type/sort of, not so much NP as NP and even prefixes like super-, mini-, etc.
While some or all of these signals may have an affinity to certain types of non-literal language,
one would not really consider them to be metaphoricity signals in the same way as those in
(13.6a–c). Second, they investigate a carefully annotated, but very small corpus. ird, they do
notdistinguish between stronlgy conventionalized metaphors, which are found in almost every
u erance and are thus unlikely to be explicitly signaled, and weakly conventionalized metaphors, which
seems more likely to be signaled explicitly a priori).
A more restricted case study could determine whether the idea of metaphoricity signals is,
in principle, plausible. Let us look at what is intuitively the clearest case of such a signal on
Wallington et al.’s list: the sentence adverbials metaphorically speaking and figuratively speaking.
As a control, let us use the roughly equally frequent sentence adverbial technically speaking, which
does not signal metaphoricity but which can, of course, co-occur with (conventionalized)
metaphors and which can thus serve as a baseline.
ere are 22 cases of technically speaking in the BNC. Only four of these arepart of a clause that
arguably contains a metaphor:

(13.7a) Technically speaking, of course, she was off duty. [BNC JYB]

(13.7b) [D]irectors can, technically speaking, be held liable for negligence and consequently sued.
[BNC CBU]
(13.7c) [T]echnically speaking, you are no longer in a position to provide him with employment. [BNC
HWN]
(13.7d) As novelists, however, Orwell and Waugh evolve not towards each other but, technically
speaking, in opposite directions. [BNC CKN]

Note that we are taking a very generous view on what counts as a metaphor: (13.7a) contains the
spatial preposition off as part of the phrase off duty, which could be said to instantiate the metaphor “a
situation is a location”; (13.7b) uses hold as part of the phrase hold liable, instantiating a metaphor
like “believing something about someone is holding them” (cf. also hold s.o.
responsible/accountable, hold in high esteem); (13.7c) uses provide as part of the phrase provide
employment, which instantiates a metaphor like “causing someone to be in a state is transferring an
object to them” (cf. also provide s.o. with an opportunity/insight/power…). All three cases are highly
lexicalized. Even (13.7d), which uses the verb evolve metonymically to refer to a non-
evolutionary development and then uses the spatial expressions towards and opposite direction

362
 Anatol Stefanowitsch

metaphorically to describe the quality of the development, is a relatively conventionalized

expression.
ere are 19 cases of the sentence adverbial figuratively/metaphorically speaking in the
BNC. In stark contrast to technically speaking, all but two of these are part of a clause that
clearly contains a metaphor or metonymy. e metaphorical expressions are listed in Table 13-
6, the two exceptions are shown in (13.8a, b):

(13.8a) “Figuratively speaking”, he declared, “in case of need, Soviet artillerymen can support the
Cuban people with their rocket fire …”. [BNC G1R]
(13.8b) Let's pick someone completely at random, now we've had Tracey
figuratively speaking! [BNC F7U]

Table 13-6. Metaphorical expressions signaled by metaphorically/literally speaking

(BNC)
EXPRESSION
bloody the nose of sb
take it on the chin
put the boot in
hold gun to sb’s head
a taste for excitement
balance the books
cash in
make a killing
transport sb across the world
bulwark of society
make sth serve one's aims
a frozen moment in time
superego be soluble in alcohol
burst into song
give one’s right arm to do sth
be within arm’s length
you tell me
MEANING METAPHOR SOURCE
be successful in court against sb “legal fight is physical fight” A3G
endure being criticized “argument is physical fight” ASA
treat sb cruelly “sports is physical fight” K25
coerce sb to act “power is physical force” CBF
experience “experience is taste” AC9
set things right “life is commercial transaction” KAE
be successful “life is commercial transaction” C9U
be financially successful “life is a hunt” BMR
makesb think ofa distant location “imagination is motion” ACX
defender of society “defense is a wall” B0Y
put sth to use in achieving sth “to be used is to serve” BMA
documentation of a particular state “time is a river” HPN
self-control disappears when drunk “character is a substance” KM7
take one's turn speaking “singing for speaking” B21
want sth very much “body part for personal value” B2D
be in close proximity “arm's length for short distance” HTP
your co-employee tell me “employee for company” HGY

We can now compare the literal and metaphorical contexts in which the expressions technically
speaking and metaphorically/figuraltively speaking occur. If the la er are a metaphoricity signal,
they should occur significantly more frequently in metaphorical contexts than the former. As Table
13-7 shows, this is indeed the case (χ2= 20.74 (df = 1), p < 0.001).

363
13 Metaphor and Metonymy

Table 13-7. Literal and figurative u erances containing the sentence adverbials
metaphorically/figuratively speaking and technically speaking in the BNC
S EN T ENCE ADV ERBIAL

MET APHORICALLY / TECHNICALLY Total

FIGURA TIVELY SPEAKING SPEAKING

Obs: 17 Obs: 4 21
YES
Exp: 9.73 Exp: 11.27
F IG U RA T I V E

Obs: 2 Obs: 18 20
NO
Exp: 9.27 Exp: 10.73
19 22 41

Of course, the question remains, why some metaphors should be explicitly signaled while the
majority is not. For example, we might suspect that metaphorical expressions are more likely
to be explicitly signaled in contexts in which they might be interpreted literally. is may be the case
for put the boot in,which occurs in a description of a rugby game where one could potentially
misread it for a statement that someone was actually kicked:

(13.9a) Guy Gregory, who was later to prove the match winner, easily converted . Gloucester needed a
kick start to get into the game. Three penalties helped put them in front. But not for long.
Gregory put the boot in… metaphorically speaking!

Alternatively (or additionally), a metaphor may be signaled explicitly if its specific phrasing is
more likely to be used in literal contexts. is may be the case with hold a gun to sb’s head, of which
there are ten examples in the BNC, only one of which is metaphorical:

(13.9b) ‘I'm not sure if the princess held a gun to Charles's head, figuratively speaking, but it
seems if she wanted something said.’

Again, which of these hypotheses (if any of them) is correct would have to be studied more
systematically.
is case study found a clear effect where the authors of the study it is based on did not. is
demonstrates the need to formulate specific predictions concerning the behavior of specific
linguistic items in such a way that they can

364
 Anatol Stefanowitsch

be tested systematically and the results be evaluated statistically. e study alsoshows that the area of
metaphoricity signals is worthy of further investigation.

13.2.3.3 Metaphor and ideology. Regardless of whether metaphor is a rhetorical device (as has
traditionally been assumed) or a cognitive device (as seems to be the majority view today), it is clear
that it can serve an ideological function, allowing authors suggest a particular perspective on a
given topic. us, ananalysis of the metaphors used in texts manifesting a particular ideology should
allow us to uncover those perspectives.
For example, Charteris-Black (2005) investigates a corpus of “right-wing communication and
media reporting” on immigration, containing speeches, political manifestos and articles from the
conservative newspapers Daily Mail and Daily Telegraph. He finds, among other things, that the
metaphor “immigration is a flood” is used heavily, arguing that this allows the right to portray
immigration as a disaster that must be contained, citing examples like a flood of refugees, the tide of
immigration, and the trickle of applicants has become a flood.
Charteris-Black’s findings are intriguing, but since he does not compare the findings from his
corpus of right-wing materials to a neutral or a corresponding le -wing corpus, it remains an open
question whether the use of these metaphors indicates a specifically right-wing perspective on
immigration. Let us therefore replicate his analysis more systematically. e BNC contains 1 232
966 words from the Daily Telegraph (all files whose names begin with AH, AJ and AK), which
will serve as our right-wing corpus, and 918 159 words from the Guardian (all files whose names
begin with A8, A9 or AA, except file AAY), which will serve as our corresponding le -wing
(or at least le -leaning) corpus. Since Charteris-Black’s examples all involve reference to target-
domain items such as refugee and immigration, a metaphorical-pa ern analysis (cf Section
13.2.2.1 above) suggests itself. Table 13-8a shows all concordance lines for the words
migrant(s), immigrant(s) and refugee(s) containing metaphorical pa erns instantiating the
metaphor “immigration is a mass of water”.
In terms of absolute frequencies, there is no great difference between the two subcorpora (10 vs.
11), but the overall number of hits for the words in question differs drastically: there are 136
instances of these words in the Guardian subcorpus but only half as many (68) in the Telegraph
subcorpus. is means tharelatively speaking, in the domain of migration liquid metaphor are
more frequent than expected in the Telegraph and less frequent than expected in the Guardian (see
Table 13-8b), which suggests that such metaphors are indeed typical of right-wing discourse. e
difference is only marginally significant,

365
13 Metaphor and Metonymy

however, so a larger corpus would be required to confirm the result (χ2=3.822, df=1, p = 0.0506).

Table 13-8a. Selected LIQUID metaphors with migrant(s), immigrant(s), refugee(s)

TELEGRAPH (n=68)
Britain would be ‘ swamped with [[immigrants]] ’ under a Labour Government s country
would be swamped with [[immigrants]] of every colour and race , support was due to the
flood of [[migrants]] and would-be asylum seekers . t increasing levels of economic
[[migrants]] and asylum seekers entering B nges to the constitution if the [[refugee]]
influx was to be curbed . Herr d open up Britain to a flood of [[immigrants]] and permit
the rise of fasc
the Gulf war and the influx of [[refugees]] from Afghanistan and Iraq . T for help to
deal with flood of [[refugees]] By Philip Sherwell in Tuzla T Sherwell in Tuzla THE FLOOD
of [[refugees]] fleeing the escalating confli an militia groups . Most of the
[[refugees]] flowing into Tuzla are escapi gling to cope with the flood of [[refugees]]
and have appealed to the inte
GUARDIAN (n=136)
mirroring this year 's flood of [[refugees]] . Watched by a demonstration litary security ,
migration and [[refugee]] flows on an vast scale . As We
cope with the current surge of [[refugees]] , her Foreign Secretary , Mr nd other
aspects of large-scale [[immigrant]] absorption . The bureaucracy he need to control an
influx of [[immigrants]] . Rebel troops end siege of control and manage the flows of
[[migrants]] that wars , disasters , and , greed that the flow of economic [[migrants]]
from Vietnam should be stoppe e form of an influx of Romanian [[refugees]] . In one case
in 1988 a Roman control and manage the flows of [[migrants]] that wars , disasters , and ,
greed that the flow of economic [[migrants]] from Vietnam should be stoppe

Table 13-8b. LIQUID pa erns with the words migrant(s), immigrant(s), refugee(s)
N E W S P AP ER

GUARDI AN (PROGRESSIVE) TELEGRAPH Total

(CONSERVATIVE)
LIQUID Obs: 10 Obs: 11 21
MET. Exp: 14 Exp: 7
P AT T E R N

Obs: 126 Obs: 57 183

OTHER
Exp: 122 Exp: 61
136 68 204

is case study demonstrates that even general metaphors such as “immigration is a mass of water”
may be associated with particular political ideologies. ere is a large research literature on the role
of metaphor in political discourse (see, for

366
 Anatol Stefanowitsch

example, Koller 2004, Charteris-Black 2004, cf. also Musolff 2012), although at least part of this
literature is not as systematic and quantitative as it should be, so this remains a promising area of
research). e case study also demonstrates theneed to include a control sample in corpus-linguistic
designs (in case that this still needed to be demonstrated at this point).

13.2.4 Metonymy
Following G. Lakoff and Johnson (1980:35), metonymy is defined in a broad sense here as “using one
entity to refer to another that is related to it” (this includes what is o en called synecdoche, see Seto
1999 for critical discussion). Text book examples are the following from (Lakoff and Johnson
1980: 35 and 39 respectively):

(13.10a) e ham sandwich is waiting for his check

(13.10b)Nixon bombed Hanoi.

In the (13.10a), the metonym ham sandwich stands for the target expression “the person who ordered
the ham sandwich”, in (13.10b) the metonym Nixon stands for the target expression “the air-force
pilots controlled by Nixon” (at least at first glance, cf. Case Study 13.2.4.3 below).
us, metonymy differs from metaphor in that it does not mix vocabulary from two domains,
which has consequences for a transfer of the methods introduced for the study of metaphor
in Section 13.1.
e source-domain oriented approach can be transferred relatively straightforwardly –
we can query an item (or set of items) that we suspect may be used as metonyms then identify the
actual metonymic uses (see Case Study 13.2.4.1). e main difficulty with this approach is
choosing promising items for investigation. For example, the word sandwich occurs almost 900
times in the BNC, but unless I have overlooked one, it is is not used as a metonym even once.
A straightforward analogue to the target-domain oriented approach (i.e., metaphorical pa ern
analysis) is more difficult to devise, as metonymies do notcombine vocabulary from different semantic
domains. One possibility would be to search for verbs that we know or suspect to be used with
metonymicsubjects
and/or objects. For example, a Google search for 〈"is waiting for (his|her|their|the)
check"〉 turns up about 20 unique hits; most of these have people as subjects and
none of them have meals as subjects, but there are three cases that have table as
subject, as in (13.11):

367
13 Metaphor and Metonymy

(13.11) Table 12 is waiting for their check. [articles.baltimoresun.com]

A more systematic but still somewhat informal investigation of this type will be presented in Case
Study 13.2.4.2).
Finally, note that analyses of potential metonyms or target frames are not the only way in which
corpus data can shed light on metonymical relationships; a particularly interesting, much more
complex study is discussed in Case Study 13.2.4.3).

13.2.4.2 Case study: Metonymy and grammar. As mentioned above, a quasi-source- domain-oriented
approach, i.e. one starting from the potential metonym, is straightforward to implement, although
it depends on the choice of a promising candidate for metonymic usage. Body parts are known to be
such promising candidates (see, e.g., Deignan 1999 on shoulder, Levin and Lindquist 2007 on nose,
Lindquist and Levin 2008 on foot and mouth and Niemeier 2008 on head and heart).
Hilpert (2006) looks at metonymic uses of eye with the aim of determining the extent to which they
represent conventionalized pa erns (fixed or semi-fixed phrases like those in [13.11a,b]) as opposed
to productive uses (of which Hilpert provides no examples, but [13.11c], a variant of the
conventionalized have an eye for, might be acandidate):

(13.11a) [H]er brother-in-law … had kept an eye on her finances from the beginning
[BNC FPM]
(13.11b)A detail on the screen had caught his eye. [BNC FR0]
(13.11c) Joan had an eye which knew what to do with apparently unpromising houses [BNC A68]

In a ten-million-word sample from the BNC, Hilpert finds 443 metonymic uses of eye. Of these, 323
(i.e. 72.9 percent) represent highly conventionalized pa erns, which are listed in Table 13-9 (with
some minor differences from Hilpert’s original classification).

368
 Anatol Stefanowitsch

Table 13-9. Metonymic pa erns with eye in a 10 percent sample of the BNC
Pa ern Meaning Metonymic Links Freq.
i. eye contact 46
ii. catch POSS eye catch 34
‘visual contact’
the eye (of NP) eye for watching 7
eye-catching 2
iii. VDET/POSS eye over NP ‘scan NP’ 5
iv. keep an eye on NP keep 56
a ADJ eye on NP ‘pay a ention to NP’ 10
keep POSS eye on NP 11
v. one eye on NP ‘pay some a ention to’ eye for watching + 5
vi. PREP the public eye ‘PREP the public a ention’ watching for a ention 12
PublicEye (TV series) 7
vii. with an eye on NP (cf. xiii) ‘pay a ention to NP’ 2
viii. turn a blind eye (to NP) ‘disregard something/NP’ 17
ix. under the eye of NP eye for watching + 6
‘under observationof NP’
under POSS eye watching for supervision 1
eye for watching +
x. with an eye to NP ‘with concern for NP’ 5
watching for concern
xi. ‘with the intention of V-ing eye for watching + watching for
with aneye to V-ing NP 3
NP’ intention
eye for watching +
xii. have an eye for NP (cf xvii) ‘have interest in NP’ 3
watching for interest
xiii. with an eye on NP (cf. vii) “eye for watching + 2
‘want NP’
xiv. have POSSeye on NP watching for wanting” 6
xv. in POSS/the mind’s eye ‘in POSS/the imagination’ eye for vision 18
there BE more to NP than meets ‘there BE more to NP than is readily eye for vision + vision for
xvi. 3
the eye perceivable’ perception
xvii. have an eye for NP (cf. xii) ‘have good perception of NP’ POSS 5
ADJ eye for NP ‘good perception of NP’ eye for vision + vision for
good perception 2
with an eye for NP ‘with good perception of NP’ 2
eye for vision +
xviii. private eye ‘private investigator’ 1
vision for investigation
Private Eye (Magazine) 11
the Eye (Magazine) 6
xix. NP in POSS eye ‘NP in POSS expression’ 8
eye for expression
NP COME into POSS eye ‘NP enter POSS expression’ 3
xx. N to the eye ‘N to the beholder’ 3
xxi. ADJ to the eye ‘ADJ to the beholder’ eye for beholder 2
xxii. to the ADJ eye ‘to the ADJ beholder’ 4
xxiii. see eye to eye ‘agree’ understanding is seeing 7
xxiv. the apple of DET/POSS eye ‘cherished object’ eye for thing seen 5
xxv. black eye ‘discolored eye region’ part for part 5

369
13 Metaphor and Metonymy

is clear dominance of conventionalized metonymical expressions is interesting, as it suggests that

metonymies (or at least body-part metonymies) have a much more limited productivity than one
would expect if they were a genuinely cognitive process. is is not to suggest, of course, that
metonymy is a purely linguistic phenomenon; in addition to the 323 conventionalized
expressions, there are 120 expressions (27.1 percent) that only occur once in Hilpert’s sample and that
he therefore classifies as productive (although this estimate is probably too generous, as a larger
corpus would show that many of these pa erns do reoccur frequently enough to be classified as
(semi-)fixed phrases).
It would be interesting to look more closely at the nature of these productive uses, for example,
to see whether they are largely minimal extensions of conventionalized pa erns (like example
13.11c), or whether there are also truly creative uses, as are commonly found in the domain of
metaphor. Interestingly, these pa erns Hilpert identifies as productive uses
overwhelmingly represent the most frequently instantiated metonymies “eye for a ention”, “eye for
watching” and “eye for perception”. is may appear to suggest that these are indeed productive
metonymies, while mappings like “eye for expression” or “eye for beholder” are “dead”
metonymies limited to a few fixed phrases; however, it may also be due to the fact that the
expressions instantiating them are more entrenched and thus more amenable to extension.
is case study demonstrates the metonym-oriented approach to the corpus- based analysis of
metonymy. It also shows that such an approach need not be limited to description but can be used
to answer questions about the nature of metonymy. Metonymy has been studied far less
intensively than metaphor from a corpus-based perspective (and in general), so this remains an
exciting area of study.

13.2.4.3 Case Study: Subjects of bomb. As suggested in the introduction to this section, it may be
interesting to explore the possibility of target-domain-oriented studies of metonymy, i.e., to start from a
verb (or entire predicate) that we expect might have metonymic subjects (or objects). One such study
is Stefanowitsch (2015), which studies the verb bomb (as in 13.10b), Nixon bombed Hanoi.
According to Lakoff and Johnson, this sentence instantiates what they call the “controller for
controlled” metonymy, i.e. Nixon would be a metonym for the air
force pilots controlled by Nixon.39 us, a search for 〈 [pos: NOUN] [lemma:
39 Alternatively, as argued by Stallard (1993), it is the predicate rather than the subject

370
 Anatol Stefanowitsch

bomb]〉 should allow us to asses, for example, the importance of this metonymy in relation to other
metonymies and literal uses.
erying the BNC for 〈 [pos: NOUN] [lemma: bomb, pos: VERB] 〉 yields 31
hits referring to the dropping of bombs. Of these, only a single one has the
ultimate decision maker as a subject (cf. 13.12a). Somewhat more frequent in subject position are
countries or inhabitants of countries (5 cases) (cf. 13.12b, c). Even more frequently, the organization
responsible for carrying out the bombing
– e.g. an air force, or part of an air force – is chosen as the subject (9 cases) (cf. 13.12d,e). e most
frequent case (14 hits) mentions the aircra carrying thebombs in subject position, o en
accompanied by an adjective referring to thecountry whose military operates the planes (cf
13.12 ) or some other responsible group (cf. 13.12g). Finally, there are two cases where the
bombs themselves occupy the subject position (cf. 13.12h).

(13.12a)Mussolini bombed and gassed the Abyssinians into subjection. (13.12b)On the
day on which Iraq bombed Larak…
(13.12c) Seven years a er the Americans bombed Libya…
(13.12d)[T]he school was blasted by an explosion, louder than anything heard there since the Lu waffe
bombed it in 1944.
(13.12e)…Germany, whose Condor Legion bombed the working classes in Guernica…
(13.12 ) … on Jan. 24 French aircra bombed Iraq for the first time …
(13.12g)… Rebel
jets bombed the Miraflores presidential palace … (13.12h)… Watching our bombs
bomb your people …

Cases with pronouns in subject position have a similar distribution, again, there is only one hit with a
human controller in subject position. All hits (whether with pronouns, common nouns or proper
names), interestingly, have metonymic subjects – i.e., not a single example has the bomber pilot in
the subject position.
is is unexpected, since literal uses should be more frequent than figurative use(it leads Stefanowitsch
2015 to reject an analysis of such sentences as metonymies altogether). On the other hand, there are
cases that are plausibly analyzed as metonymies here, such as examples (13.12d–e), which seem to
instantiate a metonymy like “military unit for member of unit” (i.e. “whole for part”) and (13.f– h),
which instantiate “plane for pilot”, i.e. “instrument for controller”.
More systematic study of such metonymies by target domain could uncover

that is used metonymically in this sentence, which would make this a metonym- oriented case study.

371
13 Metaphor and Metonymy

more such facts as well as contributing to a general picture of how important particular
metonymies are in a particular language.
is case study sketches out a potential target-oriented approach to thecorpus-based study of
metonymy, along with some general questions that we might investigate using it (most obviously,
the question of how central a given metonymy is in the language under investigation). Again,
metonymy is a vastly under-researched area of linguistics, so much work remains to be done.

13.2.4.3 Case Study: Logical metonymies. Lapata and Lascarides (2003) present an innovative
approach to what they call “logical” metonymy – clauses, where the object of a verb is “logically” the
object of some other verb that does not occur in the u erance. For example, we would typically
interpret the u erance M aryenjoyed the book as “Mary enjoyed reading the book”, or perhaps “Mary
enjoyed writing the book, but definitely not as “Mary enjoyed buying the book” – even though
buying a book is a perfectly normal thing to do.
Lapata and Lascarides argue that the interpretation of the “missing” verb depends both on the
verb that is actually present (e.g. enjoy) and on the object (e.g. book) – it is determined, in our
examples, on the basis of our knowledge of what people typically enjoy and our knowledge of what
people do with films. By identifying verbal collocates shared by the verb (when used as a matrix verb
with a dependent clause) and the object, it should be possible to recover the most plausible
interpretations for any given logical metonymy.
For example, the ten most frequent dependent verbs of enjoy as a matrix verb in the BNC are (in that
order) be, work, read, play, have, go, watch, meet, make, and see; the verbs most frequently taking book
as an object (or predicate noun) are be, read, write, have, get, publish, buy, find, produce and make. Taking
into account the overall frequency of these verbs and their frequencies with enjoy and film, it should be
possible to find the best match. Lascarides and Lapata use the following formula, where Vmet is the
metonymically used verb (e.g. enjoy), Vimp is the potentially implied verb (e.g. read, write, buy)
and Obj is the object (e.g. book):

f (V , V imp) × f (Obj , V imp)

(13.13a) P (V imp ,V met
, Obj) = met

N × f (V imp)

Buy occurs 25,827 times in the BNC in general, and 125 times with the object [(Det) (Adj) book];
enjoy buying occurs 3 times, and there are 11,292,844 verbs in the version of the BNC used here.
us, we can calculate the hypothetical

372
 Anatol Stefanowitsch

probability of the co-occurrence of buy, enjoy and book as follows:

3 × 125
(13.13b) P (V ,V ,Obj) = = 0.000000001286
imp met
11292844 × 25827

If we repeat this calculation for all verbs occurring with enjoy and/or book and order the results by
probability of co-occurrence, we find that the most likely interpretation is, indeed, read, followed
by write (see Table 13-10).

Table 13-10. Most probable interpretations for enjoy & book.

VMET OBJ & V IMP V &V
MET IMP V
IMP P(V ,V , OBJ)
IMP MET

read 615 68 22702 1.63E-07

write 604 20 39230 2.73E-08
cook 18 13 3933 5.27E-09
be 1147 150 4122614 3.70E-09
get 191 19 213378 1.51E-09
travel 18 9 9890 1.45E-09
have 399 52 1317706 1.39E-09
buy 125 3 25827 1.29E-09
make 83 35 210554 1.22E-09
compile 9 2 1366 1.17E-09

Of course, the method is not flawless. First, it obviously depends on the size and quality of the corpus
(note the occurrence in third place of the “verb” cook – this is entirely due to the questionable
tagging of cook as a verb in the compound Cooking Book (British English for cookbook).
However, while it generally yields excellent results, there are cases where it ranks plausible
interpretations at the top of the list that are simply not the most expected, suggesting that pa erns
like enjoy + Object at least in some cases receive conventionalized interpretations that do not reflect
an intersection of the use of [enjoy + Verb] and [Verb + Object]. Compare, for example, the results
for food and meal in Table 13-11.
Obviously, I enjoyed the meal/food can both be interpreted to mean “I enjoyed cooking the
meal/food” in a specific context, but in both cases the default interpretation would be I enjoyed
eating the meal/food. However, based on the frequencies of [enjoy + Verb] and [Verb + meal], the most
likely interpretation for enjoy the meal should be enjoy cooking the meal. It is possible, that a more
fine- grained corpus analysis (for example, one that distinguishes between transitive

373
13 Metaphor and Metonymy

and intransitive uses of cook (and other verbs) in [enjoy + Verb]) would yield more accurate
results, but it is also possible, that usage does not fully reflect our assumptions about what people
enjoy and or typically do with meals.

Table 13-11. Most probable interpretations for enjoy & food and enjoy & meal.
VMET OBJ & V IMP V &V
MET IMP V IMP P(V ,V , OBJ)
IMP MET

FOOD
eat 210 14 13791 1.89E-08
cook 39 13 3933 1.14E-08
buy 129 3 25827 1.33E-09
share 20 9 12803 1.24E-09
be 354 150 4122614 1.14E-09
MEAL
cook 83 13 3933 2.43E-08
eat 112 14 13791 1.01E-08
make 93 35 210554 1.37E-09
share 15 9 12803 9.34E-10
have 216 52 1317706 7.55E-10

Finally, as a demonstration that the method does not simply pick the verb most frequently used with
the object, consider Table 13-12, which shows the most probable interpretations for [plan (Det)
(Adj) book]. Even though read is the most frequent verb with book, the method correctly picks write,
followed by the almost equally likely publish as the best interpretation (read is the eighth-best
interpretation).

Table 13-12. Most probable interpretations for plan & book.

VMET OBJ & V IMP V IMP V &V
MET IMP P(V ,V , OBJ)
IMP MET

write 604 27 39230 3,68E-08

publish 151 19 12697 2,00E-08
buy 125 45 25827 1,93E-08
sell 59 62 21088 1,54E-08
close 52 37 12983 1,31E-08
open 62 51 22345 1,25E-08
use 69 129 110838 7,11E-09
read 615 2 22702 4,80E-09

is study demonstrates a complex innovative research design for a highly interesting

research question. It is difficult to place in terms of “inductive” or

374
 Anatol Stefanowitsch

“deductive” and in terms of what exactly the variables are. is is because it does not use corpus
data primarily as a way of testing a hypothesis or answering a research question, but as a
model of how humans (or other language-learning systems) may acquire the correct
interpretations of logical metonymies. is model itself may then be tested, for example (as
Lapata and Lascarides do), by comparing its results with experimentally collected native-speaker
judgments. Here, the hypothesis would be that the model correctly predicts the majority of
interpretations, and the variable would be PREDICTION OF THE MODEL with the values
CORRECT and INCORRECT.

Furtfter Reading
Deignan 2005 is a comprehensive a empt to apply corpus-linguistic methods to arange of theoretically
informed research questions concerning metaphor. econtributions in Stefanowisch & Gries
(2006) demonstrate a range of methodological approaches by many leading researchers
applying corpus methods to the investigation of metaphor.
• Deignan, Alice. 2005. Metaphor and corpus linguistics. Amsterdam & Philadelphia:
John Benjamins.
• Stefanowitsch, Anatol & Stefan omas Gries. 2006. Corpus-based
approaches to metaphor and metonymy.

375
Epilogue
In this book, I have focused on corpus linguistics as a methodology, more precisely, as an
application of a general observational scientific procedure to large samples of linguistic usage. I have
refrained from placing this method in a particular theoretical framework for two reasons.
e first reason is that I'm not convinced that linguistics should be focusing quite as much on
theoretical frameworks, but rather on linguistic description based on data. Edward Sapir famously
said that “unfortunately, or luckily no language is tyrannically consistent. All grammars leak”
(1921: 39). is is all the more true of formal models, that a empt to achieve tyrannical
consistency by pretending those leaks do not exist or, if they do exist, are someone else's
problem. To me, and to many others whose studies I discussed in this book, the ways grammars leak
are simply more interesting than the formalisms that help us ignore this.
e second reason is because I believe that corpus linguistics has a place ni any theoretical
linguistic framework, as long as that framework has some commitment to modeling linguistic
reality. Obviously, the precise place, or rather, the distance from the data analyzed using this method
and the consequences of this analysis for the model depend on the type of linguistic reality that is
being modeled. If it is language use, as, for example, in historically or sociolinguistically oriented
studies, the distance is relatively short, requiring the researcher to discover the systematicity
behind the usage pa erns observed in the data. If it isthe mental representation of language, the length of
the distance depends onyour assumptions about those representations.
Traditionally, those representations have been argued to be something fundamentally
different from linguistic usage – that they are an ephemeral “competence” based on a “universal”
grammar that may be a “mental organ” (Chomsky 1980) or an evolved biological instinct
(Pinker 1994), but that is dependent on and responsible for linguistic usage only in the most indirect
ways imaginable. As I have argued in Chapters 1 and 2, even those frameworks have no alternative to
corpus data that does not suffer from the same drawbacks, without

376
 Anatol Stefanowitsch

offering any of the advantages.

However, more recent models do not draw as strict a line between usage and mental representations.
e “Usage-Based Model” (Langacker 1990) is a model of linguistic knowledge based on the
assumption that speakers initially learn language as a set of unanalyzed chunks of various sizes
(“established units”), from which they derive linguistic representations of varying degrees of
abstractness and complexity based on formal and semantic correspondences across these units (cf.
Langacker 1991: 266f). Hopper's “Emergent Grammar” is based on similar assumptions but is
skeptical even of abstractness, viewing language, instead, as “built up out of combinations of …
prefabricated parts. Language is, in other words, to be viewed as a kind of pastiche, pasted together
in an improvised way out of ready-made elements” (Hopper 1987: 144).
In these models, the corpus becomes more than just a research tool, it becomes part of a model of
linguistic competence itself (see further Stefanowitsch 2011). In fact, in the most radical version,
Hoey’s (2005) notion of “lexical priming”, the corpus essentially is the model of linguistic
competence:

e notion of priming as here outlined assumes that the mind has a mental concordance of
every word it has encountered, a concordance that has been richly glossed for social,
physical, discoursal, generic and interpersonal context. is mental concordance is
accessible and can beprocessed in much the same way that a computer concordance is, so that
all kinds of pa erns, including collocational pa erns, are available for use.It simultaneously
serves as a part, at least, of our knowledge base. (Hoey 2005: 11)

Obviously, this mental concordance would not correspond exactly to any concordance
derived form an actual linguistic corpus. First, because – as discussed in Chapters 1 and 2 –
no linguistic corpus captures the linguistic experience of a given individual speaker or the
“average” speaker in a speech community; second, because the concordance that Hoey
envisions is not a concordance of linguistic forms, but of contextualized linguistic signs – i.e., it
contains all the semantic and pragmatic information that corpus linguists have to reconstruct
laboriously in their analyses. Still, an appropriately annotated concordance from a balanced
corpus would be a reasonable operationalization of this mental concordance (cf. also Taylor 2012).
In less radical usage-based models of language, such as Langacker’s, the corpus is not a model
of linguistic competence, which is seen as a consequence of

377
Epilogue

linguistic input perceived and organized by human minds with a particular structure (for
example, the capacity for figure-ground categorization). It is, however, a reasonable model (or at
least an operationalization) of this input. Many of the properties of language that guide the
storage of units and the abstraction of schemas over these stored units can be derived from
corpora (frequencies, associations between units of linguistic structure, distributions of these units
across grammatical and textual contexts, the internal variability of these units, etc., cf
Stefanowitsch and Flach 2016 for discussion).
is view is explicitly taken in language acquisition research conducted w
ithnthe Usage-Based Model
(e.g. Tomasello 2003, Dabrowska 2001, Diessel 2004), where children's expanding grammatical
abilities (as reflected in acquisition corpora) are investigated against the input that they get from their
caretakers. It is not always shared by the major theoretical proponents of the Usage-Based Model,
who connect the notion of usage to the notion of linguistic corpora only in theory. However, it is a
view that offers a tremendous potential to bring together two broad strands of research –
cognitive-functional linguistics (including some versions of construction grammar) and
corpus linguistics (including a empts to build theoretical models on corpus data, such as
Pa ern Grammar [Hunston and Francis 2000] and Lexical Priming [Hoey 2005]). ese strands
have developed more or less independently and their proponents are sometimes mildly hostile
toward each other over minor, but fundamental differences in perspective (see McEnery and
Hardie 2012, Section 8.3 for discussion), but they could complement each other in many ways,
cognitive linguistics providing a more explicitly psychological framework than most corpus
linguists adopt, and corpus linguistics providing a methodology that cognitive linguists
serious about usage urgently need.
Finally, in such usage-based models, as in models of language in general, corpora can also be
seen as models (or operationalizations) of the typical linguistic output of the members of a
speech community, i.e. the language produced based on their internalized linguistic knowledge.
is is the least controversial view, and the one that I have essentially adopted throughout this book.
Even under this view, corpus data remain one of the best sources of linguistic data we have, one
that can only keep growing, providing us with ever deeper insights into the leaky, intricate, ever-
changing signature activity of our species.
I hope this book has inspired you and I hope it enables you to produce research that
inspires all of us.

378