Professional Documents
Culture Documents
in discourse
How to make the most of qualitative software
and find a good research design
Michael Kimmel
University of Vienna
Michael Kimmel
Michael Kimmel
publications keep the inevitable stumbling blocks in the dark and the applied rationales of coping to a minimum. It is common (and perhaps unavoidable) in academic writing to iron out the smaller creases and kinks of the research process
to present a more coherent picture. Difficulties are given short shrift, instead of
publishing project notes others may learn from. Such problems may concern the
sampling, the elicitation procedure (when interviews or focus groups are used),
the reliability of coder ratings, the disentangling of metaphor from related phenomena, and the grouping metaphors into coherent sets, amongst others (cf. Low
& Todd, 2010). Furthermore, studies frequently fail to make their data generation
process transparent enough. Hardly an article presents a corpus in its full original
complexity or talks about its metaphorically less systematic parts. This makes us
lose sight of how the coherence we see in a publication is inevitably also an outcome of the researchers interpretive endeavor and choice of material. By a similar
token, quantitative aspects of a corpus such as metaphor frequency, diversity (in
general and by target), and type/token ratio are not discussed for their implications
(or even mentioned) by many authors. Next, meta-reflexive evaluations of how
suitable metaphor analysis is for a given topic remain largely absent, at least to the
extent that comparisons with other methods are rare. Finally, issues of study design are given little explicit attention. Even well-versed researchers have to decide
technical issues, choose a level of granularity, and deploy their time-resources in
accordance with the scope of their aims. Alternative strategies are hardly ever contrasted, in order to give prospective researchers criteria for deciding what will suit
their needs. The root common to all of these shortcomings is that comprehensive
procedural standards for qualitative metaphor research are slow to come. If metaphor scholarship is to measure up against the best practice of qualitative research
at large it needs to emulate their explicit procedures, as well as being transparent
about the way theorizing is rooted in the data. Hence, a systematic approach should
guide scholars through a project step-by-step in their endeavor to reconstruct
(a) conceptual models or (b) discourse dynamics through a collection, categorization, and analysis of metaphors in a corpus,
provide checks and keep the moorings of the analysis in the data backwards
traceable (cf. audit trail, Cameron & Maslen, 2010b, p.99), and
reflexively explicate all strategic choices and the possible bias that results from it.
The aforementioned hermeneutic nature of the applied art of metaphor analysis is
certainly not a license for an anything goes or reliance on intuition. It is because
the qualitative research community recognizes interpretive expertise as something
irreducible that readers of a study should be able to reconstruct how claims came
about.
Michael Kimmel
be included. In the end I decided that all these as well as targets like politicians
and political action would benefit the study. Finally, one may also decide to exclude certain general types of metaphors like ontological metaphors (see 2.4) and
personifications.
Metaphor identification and unit size. Next, the scholars contextually and theoretically informed skills are needed to identify expressions that manifest the linguistic category of metaphor. Typically, one carefully reads a text looking for vehiclewords that signal a metaphor. The context is important here. In a sentence relating
to political institutions the word architecture signals a metaphor, whereas in
a context of urban planning it will probably be literal. Deciding what qualifies
as a metaphor is by no means trivial and requires considerable linguistic foreknowledge as well as sensitivity to context. Roughly speaking, we may identify as
a metaphor any expression in which a vehicle-word creates tension with a topicterm or an implied topic. In she is a rose the topic term she stands in semantic
tension with the underlined metaphor-indicating vehicle-word rose. Frequently,
the topic needs to be partly inferred (dirty-keeled swans for ships) or wholly
so (silly ass! for a dumb person and attach the mouse to the keyboard for an
electronic device) (Goatly, 1997, ch. 7). One of the difficulties is that most metaphors are not realized as copula constructions (A=B) like she is a rose, but in a
great many other syntactic forms (listed in Goatly, 1997). Most metaphors are not
even realized through noun vehicles, but verbs and prepositions (Cameron, 2003).
For this and other reasons,2 inferring an unstated topic is a skill in its own right
that metaphor researchers need to acquire. In the silence was slashed by a fierce
yell the implied topic is [hearing a] sudden acoustic quality, and contrary to the
superficial appearances, not the silence. Note also that the tension responsible
for metaphor can be purely contextual-pragmatic. The expression the Rottweiler
behind the bar may require the hearers knowledge of whether an actual dog or a
person is present to ascertain whether the expression is a metaphor and to figure
out its topic (cf. Steen, 1999). The same is true of get to the point, which in any
standard context is an injunction regarding communication, not physical motion.
Even more radically, recognizing no man is an island as being metaphorical requires the inference that the literal meaning is pragmatically irrelevant in all contexts (other than very far-fetched ones).
By a general definition in metaphors the vehicle words have a physical, sensorial (non-abstract), more precise, historically older, or otherwise more basic reference
than their meaning in the given context (Pragglejaz Group, 2007). Hence, (a) one
needs to establish what the contextual meaning of the expression is, (b) whether a
more basic reference can be found elsewhere in contemporary usage, especially
in a lexicon. If no such reference is found, the contextual expression is literal. If,
however, a more basic reference is found one decides in a last step (c) whether it
contrasts with the contextual reference, while also involving an element of comparison. This helps exclude phenomena of semantic/pragmatic tension that are
not strictly metaphorical. To differentiate metaphor from metonymy we need to
ensure that the contextual and basic meanings are not related by being part of
the same frame or domain. To differentiate metaphor from polysemy we need to
ensure that the contextual meaning is still somehow understood in relation to the
basic meaning, i.e. that some conceptual transfer occurs. The reader is advised to
consult the Metaphor Identification Procedure (MIP) developed by the Pragglejaz
Group (2007), as well as to read Steen (2002), Heywood et al. (2002) and Cameron
& Maslen (2010b), who discuss these issues and the numerous possible types of
borderline cases in detail.
Another vital decision concerns the unit size of analysis. Should the maximum
text span of a metaphor be single words only, multi-word, or even whole sentences? While the MIP advocates screening every single word for metaphoricity,
this atomistic strategy can be unwieldy and overly time-consuming, in addition
to being psycholinguistically implausible. Even simple multi-word metaphors like
get to the point need to be split into four separate analytic units of which three
vehicle words (get, to, point) may be recognized as metaphorical. The same is
true for we have a mountain to climb. Conversely, the drawback of using multiword units is that this strategy is less reducible to a simple rule of thumb and creates many boundary cases. An ideal solution does not exist. At the theoretical level,
the MIP recognizes that each single word of a sentence functions as a backdrop for
the others against which their contextual basic reference and thus metaphoricity
are decided. The approach thus has a certain degree of implicit holism. In any case,
scholars who study large corpora will often find the the afforded gains small compared to the added workload that the MIPs piece-meal procedure necessitates.
Grouping metaphors into coherent sets. After the identification stage, one can begin
to ask of what type a metaphor is and what it shares with others. As this is explained
later (2.3) I shall skip ahead one step and assume the coder has finished with tagging the metaphorical expressions in the corpus for source and target domain.
Now the task is to find systematicity in the data that reveals something about a
discourses key topics and their conceptualization. This is done by grouping metaphors into sets with a shared conceptual basis. Many, although not all authors
assume that systematicity results from culturally shared conceptual metaphors,
i.e. sets of correspondences between conceptual domains that drive discourse
production (Lakoff & Johnson, 1980). Goatly (1997, 2007) also calls these root
metaphors or themes. The discourse dynamics framework with its emphasis
on discursive context prefers to speak of systematic metaphor (Cameron, 2007;
Michael Kimmel
Cameron, 2010).3 Whatever the nomenclature, metaphors may have a shared basis
when they are (a) about the same or related targets and are (b) coherent in imagery, propositional content, and inferences (which can either originate in the source
domain alone or in its interaction with the target). Here is a famous example from
Lakoff & Johnson (1980, p.4) for a set of common lexical expressions conforming
to these criteria:
Table1. A lexical set with a shared conceptual basis
Your claims are indefensible. / He attacked every weak point in my argument. / His criticisms
were right on target. / I demolished his argument. / Ive never won an argument with him. /
You disagree? OK, shoot! / If you use that strategy, hell wipe you out. / He shot down all of my
arguments.
Next, the expressions of the set are grouped together under a metaphor formula
like argument is war. Finding such a summarizing formula is, for better or for
worse, guided by an intuition of what counts as conceptually similar. The researcher faces a tricky decision: How generalizing and broad should the formula be? One
set of metaphors could point to the formula the antagonists of a discussion
are combatants, another to entering the discussion is entering into battle, a third set to intense discussion is heavy fighting, and a fourth set to an
argumentative plan is a strategy. These four formulas may either be posited
to be separate or subsumed under the generic formula argument is war. What
is more, although the metaphors are logically all war-related, some aspects like
intensity or entering can potentially be subsumed under orthogonal sets as well
(see 4.2). While grouping expressions together remains a refined interpretive skill,
we shall later see that annotation software provides a natural way of collecting
similar metaphors via codes and ways of dealing with orthogonal sets.
Analysis of functions. Optionally, the researcher can analyze the conceptual mappings for their discursive functions (cf. Goatly, 1997, ch. 5; Semino, 2008). This
means getting an idea why a specific type of metaphor is recurrently used and to
what extent the uses vary in cognitive or discursive function. Whether this is in texts
aiming at persuasion and explanation or in spoken discourse, potentially interesting
functions of metaphor may include how speakers (a) highlight and hide aspects of
their topic or reframe it, (b) compress inferences or create complex analogies, (c)
evoke emotions like pride, pathos, or contempt, (d) evoke vivid quasi-perceptual
imagery, (e) create argumentative impact and grab the audiences attention, (f) create common ground in discourse, (g) mark discourse boundaries, or achieve (h) ingroup marking and intimacy, (i) humor and hyperbole, and (j) euphemism. The required analytic task remains deeply hermeneutic in that one needs every metaphors
context, a good knowledge of the entire discourse and, in some cases, a guiding
framework (see 2.9). Yet, software can assist even here if so desired. Metaphors can
be additionally tagged with functional codes to later explore which metaphor type
goes with which functional types (i.e. through code co-occurrences).
Getting a quantitative overview. Although this is seldom realized, basic quantitative information is essential to qualitative research. The sufficient frequency of a
pattern is a prerequisite for postulating that a conceptual metaphor or any other
kind of systematicity applies to a discourse (cf. Goatly, 2007). Moreover, the scholar may want to select conceptual metaphors by their relative importance or get an
overview of how diverse mappings are in the corpus. She may also aim at comparing of metaphor across sub-corpora (e.g. left vs. right wing newspapers, male vs.
female speakers) or across sampling times (weeks, months, years). Such comparison can involve type and token frequencies, metaphor diversity for a given target
or across all targets, metaphor frequency per word or per analyzed document/
interview, or metaphors bursts (see below). As Schmitt (2005) suggests, metaphor
sets can also be compared to wider discourse trends either by using corpus tools
that access so-called reference corpora, by using metaphor databases, or simply
by comparison to previous metaphor studies. A comparative view can showcase
conspicuously absent patterns or weigh a metaphors relative import in the total
picture. More generally, qualitative researchers benefit from a basic grasp of what
corpus linguists do with software like WordSmith Tools (Deignan, 2005; Deignan
& Semino, 2010). A basic grasp of the logic of reference corpora and indicators
such as unusually frequent words (keyness) is helpful as well.
Textual cohesion between metaphors. Retrieving metaphor sets that manifest a
shared logic amounts to studying discourse coherence across a corpus. By contrast, in a cohesion-based analysis we probe for textual adjacency patterns between
metaphors, for example to identify metaphor bursts (Cameron & Stelma, 2004;
Corts, 2006), to explore interaction types of cohesive metaphors (Goatly, 1997,
ch.9) or to study cohesion devices that link metaphors in the same sentence or
argument (Kimmel, 2009b, 4.3). With the appropriate software one can also combine both perspectives. One may search for cohesive metaphors under the added
constraint that they be coherent, e.g. all path-related metaphors in a local cluster.
Or, one may study the dynamics of metaphor in spoken discourse regarding how
one speaker explicitly rejects, accepts, expands, or renegotiates metaphors by the
other (Liebert, 1997; Cameron, 2010). Both kinds of research imply that we look
at coherence (or disparity) together with cohesion (or distance).
10
Michael Kimmel
(3) aim to address complex research questions about metaphor coherence and
cohesion.
On the other hand, it is evident that the time manual coding requires limits it
to medium size corpora of, say, 1501000 newspaper articles, 540 interviews or
310 literary texts per researcher. This brings us to my next topic, the annotation
tools needed for applying a qualitative approach in that data range with reasonable
economy.
2. Coding and analysis with atlas.ti
Prospective metaphor researchers are faced with the tasks of text annotation, retrieval, filtering, data searches, perhaps some basic number-crunching, and, increasingly, the management of large projects in teams. State-of-the-art software
like atlas.ti, maxqda,4 or NVivo is an asset for all of these. In old-style studies
texts were annotated on the margin or cut-and-paste was used to compile quotes
from text editors. Qualitative tools now streamline the basic annotation process
through simple drag-and-drop from code lists. Later filters and complex data output options can be applied. Is software a matter of mere expediency then? The
answer is no. Software promotes an economic workflow, allows browsing vast
ranges of data, as well as sharing, merging or comparing sub-projects in a team.
Judiciously applied software also helps meet the criteria of transparency and explicit procedures (see 1.1 on validation checks and audit trails). Researchers with an
all-in-one grasp of the data and visualization aids structure their hermeneutic task
effectively and minimize error or oversight. Besides these metaphor-unspecific
virtues, software enables some of the specific procedures listed in Section1.2 and
allows us to implement others with unprecedented power. How this is done will
be the main topic of this section. Later I will illustrate how software can facilitate
quantitative checks, which may in turn help with the qualitative steps.
2.1 Software for studying metaphor in discourse
Let us begin with a brief tour dhorizon. One widely available option is to customize general-purpose software for metaphor research. For example, Cameron and
her colleagues rely on Excel functions (Cameron & Maslen, 2010a) for most tasks.
Excel has powerful sorting options and allows using handy output functions like
pivot tables (Maslen, 2010). A second option for researchers with programming
skills is to develop specialized software such as VisDis, a software package developed and used by Lynne Cameron earlier. This caters to special research needs,
11
12
Michael Kimmel
justification for Section4. First, researchers who have assembled their corpus feed
it into atlas.ti in an electronic format. This can be an image, a text-file or PDF,
a video, a sound-file, or a synchronized transcript (for listening to the sound-file
while reading the text). I will stick with simple texts here, such as interview transcripts or newspaper articles. Typically, the researcher will designate each text a
sub-unit of the project (given that data patterns can later be searched across subunits). Once the data have been fed into atlas.ti, the text to be currently worked
with is chosen and appears in the left panel on screen. The code system can be displayed as a list in the right panel (see Figure1). The area between them is reserved
as a text margin for the annotations.
2.3 Compositional coding
The most time-consuming and decisive stage in any project is the coding/annotation. What is the purpose of codes? Codes are tags attached to several text
units, which later function as data containers to retrieve theoretically equivalent
expressions. A useful metaphor for codes is to see them as shopping-carts for
text units that are used while running through a text, with the aim of collecting
similar metaphors. In the later analysis, the contents of each cart can be retrieved
and displayed either as a list or as a shortlist with the option to click back to the
quote in its context. To be able to do this, codes have to be assigned to the text in
one or several thorough work sessions. Usually the researcher begins by reading
the whole text once. Then, she goes through it more slowly to identify metaphors.
The metaphors are marked and coded, two steps I will look at in greater detail now.
Text units identified as metaphors are marked by highlighting with the mouse
cursor (left panel in Figure1). Then two strategies are possible. When a step-bystep procedure is chosen a bracket is assigned to the annotation space in the central
panel of the screen. It marks the text segment for later coding. When we choose
the (default) two-in-one procedure we immediately assign a code to the selected
text unit through drag-and-drop. The code we consider fitting is selected from
13
14
Michael Kimmel
the list on right and dragged onto the highlighted text. If we wish to code various
kinds of tropes, we may also choose to first assign a generic code like metaphor to
differentiate other general phenomena like idiom, pun, or metonymy and move
on to more specific codes only later. The quotation size is up to the researcher. It
can be changed after the first coding if necessary. The metaphor units in my case
study below were, for practical reasons, a clause or at most a sentence in length,
but seldom single words. Whenever a sentence involved several independent metaphors, these were tagged separately.
The next step is to assign a more precise ontological identity to every marked
metaphor unit. I recommend a specific coding logic here, rooted in a theoretical
consensus about what metaphor is. To describe a metaphor appropriately one must
identify (i) its wider source domain (ii) its wider target domain and (iii) the specific
amount of information that actually gets mapped between them (Lakoff & Johnson,
1980; Goatly, 1997). For the purposes of discourse analysis, metaphors belong to
the same type only when they share source and target, and by a yet narrower constraint, only when the same inferences or images get mapped between them. The
latter aspect is probably best dealt with more informally (see 2.8). However, the
system implemented in atlas.ti should allow us to systematically assign source
and target codes to each metaphor unit. For instance, the expression launched the
European project receives the codes source: paths and target: EU integration
which will later be subsumed under the conceptual metaphor formula EU integration is a path. The screenshot in Figure2 shows how the researcher chooses
source domain codes from the (in this example already fully populated) code list
on the right and drags them onto the marked quotation.
Usually the text is read meticulously and tagged piecemeal, often in more than
one sitting. Occasionally, however, a strategy of semi-automated coding can be
employed, as atlas.ti also supports searches for predefined word-lists. If we are
certain the vehicle words in our list exhaustively circumscribe the aimed at phenomenon hundreds of metaphors can be automatically marked-up with codes in a
matter of seconds. However, this simple corpus tool-like facility makes sense only
for the few metaphor types that can plausibly be restricted to a well-circumscribed
range of lemmas (see 1.3).
How does the code list itself come into existence? As to the source domain
codes, many researchers may want to predefine them by compiling common metaphors from the literature or prior knowledge of the field (= deductive approach).
Alternatively, codes can be created on the fly while exploring the text (= inductive approach). It makes sense to combine both strategies by starting with a list of
frequent sources, but allowing for augmentations. Even if experienced researchers
will have usual suspects in mind it is unwise to limit the range of source codes
based on intuition, because subjective bias will enter.
The target domain code list depends on the kind of project one has in mind.
There are two possible scenarios here: If one aims to code all metaphors independently of what they are about, i.e. if all target domains remain eligible, the list has
to remain open and will grow through inductive work until some saturation point
is reached. If, by contrast, the project has a deliberate thematic focus on particular
domains say, metaphors for religion, economics, law, or politics the researcher
will delimit in advance which targets are eligible. Targets absent from the researchers predefined list will thus be left uncoded. Such off-topic metaphors may be highly frequent both in written and spoken discourse. Even in a corpus with a narrow
thematic focus off-topic metaphors may go up to a margin of 30% or so, among
other things because a certain percentage of metaphors always serves discourseorganizing rather than content-related functions, but also because of asides, etc.
2.4 Two-tier coding of source domains
Regarding source domain coding, a particular complexity highlighted by CMT
is commonly overlooked in practice. Many kinds of metaphors have been found
to involve image schemas as primary scaffolds for conceptual structure (Lakoff &
Johnson 1999). Metaphors with similar underlying image schemas (e.g. path) can
share meanings even when their rich domains (e.g. boat travel) differ, and vice
versa. More generally, each metaphor can be described from two viewpoints, with
two cognitive layers that inform metaphor processing. The expression the state
ship confronted an iceberg invites both path and collision image schemas that
are shared with non-nautical metaphors such as running into a wall of silence.
Parallel to that, our example calls up knowledge about ship navigation, crews, and
captains shared with any ship metaphor, but quite independently of collisions or
paths. One layer is the image-schematic core representation that carries the ontology of a mapping (cf. Invariance Principle, Turner 1991), while the other layer,
the cultural exemplar, piggybacks on it by adding richer knowledge and inferential
entailments. As a matter of principle, metaphors should be coded at both levels,
as different similarities with other metaphors are brought out by each layer. A bias
in favor of one way of grouping metaphors limits the quality of any study (see
15
16
Michael Kimmel
Table2. Co-occurrences for a single target: The source domains are sorted by frequency
(in brackets) to help us find the more important patterns.
EU-integration is ...
Image-schematic sources
Movements, paths and object transport
Forces
Center-periphery
Up-down
Together-apart
[65]
[11]
[6]
[5]
[3]
Rich sources
Vehicles, drivers, and journeys
Buildings
War and aggression
Social relations and groups
Sports, games and play
[8]
[8]
[3]
[3]
[3]
17
18
Michael Kimmel
Once we have inferred the fitting metaphor formulas, our aim is to compile
the expressions for each in order to start the interpretive analysis for a write-up.
For this we need the atlas.ti query tool, a data output machine that allows us
to retrieve sets of theoretically equivalent text units. Text retrieval happens in a
hypothesis-driven fashion. The printed co-occurrence table tells us what to look
for in the query tool. We simply enter source and target codes, connect them by
Boolean AND (see below), and let the machine produce all expressions belonging
to one set. Figure4 shows both the query tool and an output report.
Each query procedure results in a list of theoretically equivalent metaphors.
This Boolean co-occurrence search needs to be repeated for every source-target
pairing we want to investigate (see 3.4).6 Although this is the mainstay of the
method, an almost unlimited range of complex hypothesis testing can be done on
top of this. Basically, the atlas.ti query tool offers three separate, but combinable
search modes:
(1) We can search for text units belonging to single codes, code combinations (A
and B), or complex patterns of Boolean logic (e.g. A or B, but not C; A and
B or C and D, all but A and B, A or B or C). We can also pick out text units
belonging to a code family created for that purpose (e.g. a combination of the
different kinds of force and path related image schemas).
Figure4. Query tool and quotation output for the metaphor set [EU + vehicles/drivers/
journeys]
(2) We can search for spatial distribution patterns of metaphors in a text or multimedia file. The query tool can locate all units that overlap or rub-shoulders,
i.e., occur within a chosen text distance. For instance, we might ask if metaphor units typically are textually adjacent or have non-metaphor words between them. We might explore if metaphor clusters are distributed equally
across a longer text or occur in special slots only, such as the introduction. We
might explore how metaphor units are embedded in units coded by qualitative context analysis and explore the relative scope of two qualitative research
methods (see 4.5). Or we might explore how ontological metaphors are slotted
in larger discourse metaphors.
(3) We can search for expressions for all codes that occupy a parent, child or
sibling position in a hierarchy of codes once we have predefined this hierarchy for this purpose. This is done in a visual tool which is akin to drawing
a mind-map on a piece of paper. It allows positioning codes in a virtual space
and defining relational ontologies between them, e.g. supports, is a, exemplifies, or contradicts. Any such hierarchy is reflected in the code window on
our screen.
We can easily combine all three search options and ask which container metaphors follow immediately upon a metaphor cluster of three metaphors referring to
the target domain power, but none of which has the source domain balance? This
might sound far-fetched, but similar queries do occur. Hence, the query tool encourages exploring the data in whatever ways fit our research question and can even help
us discover new ones. All in all, atlas.ti provides a solid and variegated basis for doing interpretive work, which will be further explored in the Sections 4.3 through 4.5.
2.6 Excursus: The benefits of compositional coding
This flexibility is enhanced by combining the tool itself with the compositional coding strategy. To sum up, compositional coding has three main stages: (1) The coder
browses the text for the first time and marks text units with a (still undefined) bracket or with a generic code like METAPHOR. (2) In a second sitting text units are
tagged with target and source domains codes. (3) Identifying the precise mapping
is deferred to the final analysis; as will be explained later, this is done through creating a panoramic view of the data, counting the number of hits, and then running
a co-occurrence search for each pattern to access the expressions for the write-up.
Critics will ask this: Why dont we take the more direct route of applying full
metaphor formulas like EU integration is a path to the expressions? Foremost,
our indirect coding strategy keeps a complex study manageable. By the law of
combinatorics a few dozen sources and targets are capable of covering hundreds
19
20 Michael Kimmel
note right away that creating special codes for the following tasks does not work
too well, unless the corpus is very limited. We had best leave this to the researchers
interpretive skills after having created the code output in order to keep the code
system manageable and attention focused on one level of analysis at a time.
As probably the most important criterion for sub-divisions we may now pick
up on a claim made earlier. It was said that identifying the precise mapping relation
is a key aspect besides the source and target descriptions. Researchers should thus
be keenly aware of the mapped attributes that underlie a metaphoric expression,
i.e. what Goatly (1997) calls the grounds of a metaphor. Grounds pertain to the
analogy or similarity a mapping is based on. In the past is a foreign country; they
do everything differently there and in a faint trickle of smoke the italicized part
of the sentence expresses the basis for the mapping or at least provides a clue. The
grounds may or may not be linguistically expressed. Many metaphors are conventional and need no spelling out of the grounds, or they are novel and creative (such
as these two) and leave it to the reader to attribute the ground to the expression.
Importantly, mappings from a single set in our output may go with quite varying grounds. Which ground is intended can usually be inferred from the context,
although sometimes several readings remain open. For example, the metaphor
The EUs common house, depending on the context, might have been created
because of the inference that the EU is well-designed and solid, that it protects its
owners, that the latter share responsibilities and belong together, or that one enters
it only by being allowed to do so. Since we usually want to study metaphors with
respect to the inferences they create, the grounds constitute a highly relevant criterion for sub-grouping metaphors from a set. Thus, from the inferential viewpoint,
the formula EU=house proves to be overly abstract and requires sub-divisions
in the applied analysis.7
A still more complex way to achieve stratification is to check whether a metaphor is story-like and if speakers narrate through metaphor (Johnson, 1993,
ch. 7; Eubanks, 2000). When this is the case we can search for similar narrativization patterns of a basic mapping. In this respect, a key distinction runs between
central/core mappings and metaphor scenarios spinning-off from these (Musolff,
2004; cf. Semino, 2008, pp.219222; Kimmel, 2009a, pp.8992). A core mapping
is a generic structure such as the EU is a family. It may be linguistically explicit
in expressions like the EU family, but often remains an inferred background
structure underlying several expressions. It often consists of a simple role ascription, but need not specify a concrete action and is thus not story-like. A metaphor
scenario, by contrast, is a more contingent, dynamic characterization building on
a core mapping, such as is triggered by the expression the EU parents are getting
angry with their most recent offspring. The scenario usually specifies what happens between several roles in a small metaphorical story and is essential to the
21
22
Michael Kimmel
argumentative value of the expression. These specific mappings are relational and
put several metaphorical entities in their respective slots. The logic of metaphor
scenarios creates added ways of sorting the data into sub-sets. A core mapping
like the EU is a family entertains a one-to-many relation with the metaphor
scenarios altercation is splitting, compromising is marrying, producing
an idea is giving birth to offspring, etc. Likewise, the EU is a house spawns
various creative scenarios like entering the EU is entering a house without
an exit or institution building is construction work. Identifying such subsets makes us see the different implied inferences (and grounds), while keeping the
common higher level in evidence.
Due to different grounds, it is quite possible that a set of metaphors from a
single core mapping does not share the same image-schematic basis. The solid
build of the EU house is different from its entry condition, for instance. This brings
us to a yet finer stratification option. We may differentiate diverging role ascriptions and interaction patterns of otherwise similar image-schematic scenarios.
Take, for instance, metaphors that concern the relationship of Britain to the EU.
In one sub-set the EU engulfs the UK and in another the metaphors talk about
its entering the EU. Evidently we may group them together, because both realize the container image schemas with the same two agents. Yet, the UKs role
shifts from a passive nation surrounded by an expanding container to an active
nation deciding to join a static container. So the image-schematic logic is slightly
different. For one thing, entering lacks the emotional connotation of fear that is
present in being engulfed, and for another, the implied agency totally differs. For
this reason, the researcher may not want to base the analysis on image schemas
at the canonical abstraction level, when a discourses intrinsic logic systematically
distinguishes sub-variants like entering and surrounding.
2.9 Discourse functions
Finally, researchers may take interest in the cognitive, rhetorical, and discursive
functions of each metaphor set (see 1.2 for a list). Some key questions are: What
effect does this metaphor have that its literal counterpart (if any exists) does not?
How does the metaphor frame its topic in comparison to possible other metaphors? And what role does it take on vis--vis a texts overall purpose? When
we look at individual metaphoric expressions in co-text and context we realize
that metaphors under the same general formula may not always fulfill the same
rhetoric functions (due to different grounds but also due to negation, etc.). Thus,
even when we have generated valid metaphor sets these need not be functionally
monolithic. And conversely, when we think through the data diagonally we may
find similar functions in different metaphor sets.
23
24
Michael Kimmel
Guardian
Sun
Coded
articles
501
174
Word
count
321411
41704
Words per
article
642
240
Coded
metaphors
1588
986
Metaphor codes
per article
3,2
5,7
Metaphor codes
per word
0,005
0,024
This table shows some interesting differences between the two newspapers concerning metaphor density (but not yet diversity, see 3.4) against the backdrop of
the length difference between the quality and the yellow-press paper. This striking
disparity may be discussed if desired, but remains beyond my present scope.
3.3 Quantitative survey of targets
A further option I recommend is to generate an only-by-target survey of all metaphors. This yields a somewhat more differentiated birds-eye-view of the project.
Although it says nothing about specific mappings yet, it allows us to compare frequency trends in both newspapers in a heuristic fashion. To compute the table, the
sum of all co-occurring source codes for a given target code are added up in a line.
This is done for all targets. Let us look at the set of targets from my study concerned
with the EU itself in as depicted Table4. It becomes clear that the EU (understood
as an entity) is by far more frequent than any of its more specific sub-topics, although EU integration is also an important topic. Evidently, comparing the topics
that receive metaphoric recognition in each newspaper independently of the actual mappings may become a key heuristic before starting the interpretive analysis.
Although it conflates metaphors of different sorts, this data output option offers
orientation about a corpus.
EU is
Body
Buildings
Center-periphery
Containment, engulfment, breach
Machinery and technology
Movements, paths, object transport
Personification
Structure-lack of structure
Superstate
Up-down
Vehicles, drivers, and journeys
Animal
Animate being / agent
Business
Creation / monster
Crime and conspiracy
Body
[13]
Buildings
[16]
Center-periphery
[8]
Containment, engulfment, breach [9]
Machinery and technology
[11]
Movements, paths, object transport[7]
Personification
[6]
Structure-lack of structure
[6]
Superstate
[45]
Up-down
[6]
Vehicles, drivers, and journeys
[8]
25
26 Michael Kimmel
Table5. (continued)
EU targets
Co-occurring sources Guardian
Forces
Health and disease
Plants and growth
Religion, ritual and sacrifice
Social relations and groups
Sports, games and play
Together-apart
Streamline
Part-whole
Natural forces
War and aggression
Forces
EU functions
are
EU idea / project Life and death
is
Intact objects and destruction
Structure-lack of structure
EU economy is Forces
EU enlargement Movements, paths, object transport
is
[6]
[5]
[5]
[8]
[4]
[5]
[5]
[4]
[4]
Near-far
[6]
[61] Movements, paths, object transport[24]
[8]
[6]
[5]
[11]
[6]
[5]
[8]
[5]
[7]
Near-far
Forces
[6]
[4]
This tabular overview is a multiple asset. Almost at a glance the trained eye can
glean important metaphor formulas like the EU is a body, EU is a building, or
the EU has a center and periphery and discern the patterns that quantitatively
dominate. The table thereby suggests a preference order for the analysis. To develop methodological reflexivity, we can also pinpoint codes that seem too broad,
e.g. the frequency of force metaphors may inspire a check to see if the category is
too global. A data clean-up is equally supported. For instance, the low number of
hits for the targets EU functions and EU economy suggest that they might be
subsumed elsewhere. Finally, the list allows a quick comparison between sub-corpora in terms of their metaphor diversity. We may ask why the Sun has strikingly
fewer metaphor types than the Guardian. It can be especially instructive to subject
to closer scrutiny source domains that appear exclusively in one newspaper and
inquire into the journalistic purposes for not using metaphors used by other writers. Note, however, that a meaningful comparison of metaphor diversity and of
differences in metaphor frequency between sub-corpora necessitates taking into
account sample sizes (see Table3).9
A similar kind of tabular grouping by source domains may be created as a
countercheck. This reveals the range/semantic variety of targets onto which a particular source gets mapped.Targets that exclusively occur under the same source
rubric may be combined in the write-up. In my case study, the targets EU and EU
integration were frequently coded with path, a clue that they were inherently
relatively close and might be collapsed. Detecting such overlaps therefore helps
avoid excessive analyticity.
3.5 Commented write-up excerpt
To illustrate how the table assists a systematic write-up, I have taken from my
original study a passage that structures the text both by sources (such as path and
journey) and all seven EU related targets. Basically each relevant cell of Table5
informs a part of the text, around which I built an argument structure. In addition,
rhetoric functions are differentiated. I added comments in bold-faced brackets to
explicate the proposed methodology and to connect the discussion to Table5.
The largely overlapping source domains paths and journeys create the single largest metaphor group in both papers, being slightly more dominant in the
Guardian. [general role of section in the corpus] Because of further meaning
overlaps in the target, I decided to draw the twin-targets EU and the EU integration together (the numbers in parentheses will refer to both). The dominant mapping found here is EU integration is a journey. [summary formula] A number
of different evaluations are couched in this metaphor concerning the process, its
aims, and its status, especially after the French and Dutch No votes [discourse
27
28
Michael Kimmel
context stated]. We find two closely related mappings here: the EU is an entity
moving on a path and EU integration is movement on a path. [sub-variants] The latter seems to presuppose an implicit moving entity, most likely the EU.
Regardless of a speakers convictions, journeys are useful.
In the Sun (N=7+24), [from cells EU is + EU integration is] the pro-EU
camp speaks of a brave new course for Europe, Europe moving forward, a perfectly sensible way forward, and that the constitution does not go far enough
(Jean Luc Dehaene) and shouldnt stop there (Rocco Buttiglione). The critics say
that the EU has gone so far down the dangerous route, warn about travel[ing]
one inch further down the slippery slope of European integration. They criticize
that Brussels is carrying on regardless or scold the clowns who have driven
The Project to the brink of disaster. [scenario variants depending upon speaker viewpoint] Tony Blair is reported to call for a huge change in direction, while
foreign minister Jack Straw limits Britains involvement by saying This far and no
further. A commentator mentions that Blair has a unique opportunity to drive
the EU in the direction that is best for Britain.
A look at the Guardian (N=28+61) [from cells EU is + EU integration is]
demonstrates systematically that path attributes are used to reason with: The
EU travels in a particular direction, at a particular speed, and driven by some
force [overview of mapped aspects]. The EU is thought to be stopping, going on, moving at a certain pace, or being propelled forward too quickly and
ambitiously. It must be kept going or need not proceed forever. While for
its supporters the EU is in permanent onward flux and moving on, its critics
aim at rendering it immobile, and some skeptics expect that it will encounter
large road blocks [scenario variants depending upon speaker viewpoint]. As
to its pace, integration happens in leaps and bounds or two steps backwards
followed by almighty leaps forward. The EU can be static or dynamic; e.g. it can
be overtaken by the world economy. Hence, the EU can metaphorically vie with
other kinds of forceful movement. Or, Europe can move at two speeds concerning
rumours that Paris and Berlin planned to form a political union leading to a twotrack Europe within the EU, leaving behind recalcitrant states such as Britain.
As to direction, the EU is heading in a wrong direction, the direction may be
unclear, can be cooperative or should be changed. Of course, the end point
may differ depending on whom one asks. The EU may be on the road to ruin if it
lacks vision. The required force for traveling is equally elaborated on. The founding fathers and important nations are its driving forces. Politicians try to take
it or lead it on a journey for which a pace is set, milestones exist, and fellow
travellers are sought. The EU is being forced along in Gordon Brown [] is
brimming with ideas and determined to drag the EU kicking and screaming into
the 21st Century. Causality does not only issue from the politicians, but also from
the constitutions own driving force. Thus, in the view of supporters, rejecting the
constitution will bring the EU to a grinding halt or set the European project
back by 15 years. The constitution is the causal determinant of the direction of the
EU, such as when it turns Europe away from the path of solidarity and into that
of neo-liberalism. The path scenario provides a flexible common ground for discourse [reflection on discourse function]. It easily accommodates the crisis after
the referendum Noes in France and the Netherlands, when the future turns out to
be a rollercoaster ride, creates road blocks and blocks the path of new nations
queuing to join. Now, Europe is new territory, at a crossroads, or it can neither
go forward nor stay the same, it is shuddering to a halt with integration and
enlargement stopped in their tracks. At best, the way ahead is far from clear
and the momentum [for deepening, for widening] is broken. It can stall in the
present impasse or stumble along. Others ask from every member state to put
a shoulder to the wheel in order to move on.
Less frequent target domains inherit the path logic. [relation between targets
specified] In the Guardian, the target European enlargement (N=7) has a pace,
it may be a long and winding road or pose no hurdles, and, depending on
the viewpoint, is a process that had not been derailed or stopped in its tracks.
Paths for the target EU institutions (N=5) result in the endless, grinding pace of
institutional change, the European institutions inveigling their way into every
nook and cranny of life, and the rotating presidency will plough on. The Suns
path metaphors are too few to be considered in these categories [comment on
below threshold data].
In both newspapers, rich images of vehicles, drivers and journeys specify the
paths for the targets EU and EU integration. [reflection on the relation between
image schemas and rich sources] In the Guardian (N=29+8), the EU is a bicycle:
you keep pedaling or you fall off , a ship without a clearly defined course, a train
stopped in its tracks or it can be streamlined. In the crisis Britain can seize the
steering wheel or the helm of the EU. Vehicles are used to specify the speed and
safety in bicycling vs. an amble. In the Sun vehicles are found for the EU target
(N= 8). Blair takes the helm or sets a brave new course or has the unique
opportunity to drive the EU in the direction thats best for Britain, when Britain
assumes the EU presidency. On the whole, the vehicle code overlaps with the path
code almost totally and is near-redundant.
At this point a full analysis would turn to the next source domain from Table5,
e.g. machinery. Space limitations preclude this here, although I shall present an
abbreviated analysis of machinery towards the end of the next synoptic section.
3.6 Synopsis of metaphor formulas and discursive functions
I shall now further condense the data for better global insight. My synopsis will
distill the argumentative thrust from the metaphor formulas, highlight the actually mapped aspects/grounds, and define them relative to their discourse functions.
It is here that we apply stratification options from Section2.8. The above excerpt
29
30
Michael Kimmel
was developed along the lines of the image schema code paths, which more or
less coincided with the rich domain code of vehicles, drivers, and journeys. As
I shall show now, this data segment is easily given structure by splitting a single
core mapping into its mapped sub-aspects and scenario variants. I shall also list
a second group below, which I did not have the space to present in more detail
above. It concerns machinery metaphors, a rich domain of many overlaps with
the rich domain buildings and the image-schematic domain structure. This example points to an important further purpose of the synopsis, which is to make us
take explicit notice of image-schematic variants within a rich domain and discuss
them, for instance the relationship between the EUs motor, a force metaphor
and the idea to dismantle the EU, a structure metaphor. Finally, note the interesting abstract overlap between the journey and machinery sets in the metaphor
gridlock, where process and function meet in a single notion.
EU integration is a journey [generally highlights processes and causes]
the modality of EU integration is a chosen pace emphasizes progress as being continuous, erratic, rushed, too inactive, or too slow.
EU policy is a chosen direction / decision-makers are leaders emphasizes decisions about the EUs future being (un)clear or wrong, and that
the integration process has powerful advocates.
difficulties in integration are impediments to motion emphasizes
the No vote as obstacle to further integration; converges with gridlock of
institutional functioning and streamlining to avoid it (Guardian only).
a pause for reflection is stepping back brought up in warnings against
rushing decisions after the No vote.
a political agent / the constitution is the integrations driving
force emphasizes the causal role of the constitution or particular politicians for integration.
the EU is a machine [generally highlights functionality and its prerequisites]
The EU is a motor The European institutions and, by implication the
constitution, drive integration (either smooth running or the No votes put
a spanner into the works, create gridlock, dismantle it.)
important nations are the EUs motor / control room used to indicate
that some nations are more powerful or more willing to shape European politics.
EU member states interact as cogs of a machine emphasizes that the
EU consists of countries with complementary functional roles.
the EU is a structure conceived at the drawing board emphasizes
that the EU is a complex idea, has a sound structure/needs to be redesigned,
and that the constitution shapes it (as its blueprint).
[Further source domains from the table are summarized in a similar way]
At the end of a project we have thus arrived at an executive summary that boils
down the data, presents sets of core mappings, and applies the discussed stratification rules to differentiate them internally. With a summary of a whole corpus
at hand, it is possible to discuss the focus, internal variety, and boundaries of a
discourse, to compare newspapers globally, and to explore discourse coherence
from a birds eye view.
4. Analytical potential for discourse research on metaphor
We are now in a position for a broader evaluation of the proposed strategies. To
begin with, what justifies the compositional and two-tier coding strategies in view
of their time-intensiveness? I shall argue that they make up for this drawback
through improved workflow management, methodological precision, and analytic
scope (4.1 and 4.2). The subsequent subsections will illustrate the rich analytic
possibilities for reconstructing metaphor cohesion, narrative linkages, and for
multi-method comparisons (4.34.5).
4.1 Requirements of a good coding strategy
Annotation is beset by two interconnected problems: First, we need to find code
names that capture what the expressions subsumed share. Second, we need to decide on a general strategy, i.e. a coding design that is calibrated to the aimed at
analysis (see 2.3). Let us briefly contrast the two available main options here. The
most intuitive strategy is to assign a full mapping formula to a metaphor the moment we hit upon it. This one metaphor-one code procedure will later make each
type of mapping retrievable by clicking a code. We may call it one-shot coding.
By contrast, compositional coding, the strategy I recommend, applies two codes to
each metaphor and uses a co-occurrence search to retrieve the mapping later on.
Is this slightly roundabout strategy really preferable to the simplicity of one-shot
coding? Although the latter works well enough for small corpora, compositional
coding gains appeal with all larger and metaphorically diverse corpora for three
reasons. One has to do with code list size, one with workflow, and a third one with
the superior possibilities in the later analysis.
The first drawback occurs when many new metaphorical phenomena keep
popping up. In one-shot coding this makes the code list explode in size. One may
have to add new formulas far beyond the 2030% of a corpus considered sensible. In compositional coding many combinatorial possibilities can be generated
out of a circumscribed set of basic codes. Second, one-shot coding tends to force
researchers to repeatedly adapt a code name to make it an adequate summary
31
32
Michael Kimmel
33
34
Michael Kimmel
the constitution. Based on rich knowledge, the metaphor coheres with all punishment-, trial-, justice-, or death-related metaphors, depending on our focus. Thus,
each tier can afford a potentially relevant way of grouping. In sum, two-tier coding
does not resolve all classificatory problems, but it provides a basis for scanning the
data multi-directionally, thus generating a layered and rich analysis.
4.3 Tracing the textual interaction of metaphors (i.e. cohesion patterns)
Discourse research has emphasized the importance of addressing cohesion relations between metaphors (Goatly, 1997:ch. 9., Liebert, 1997, Cameron 2010, pp.
89ff). Metaphors in clusters may repeat, elaborate, or negate their neighbors or
otherwise interact with them. How interaction happens has theoretical repercussions on CMT. It has been asked whether conceptual metaphors govern discourse
production or whether higher-level constraints simply enlist conceptual metaphors (Quinn, 1991; Shen & Balaban, 1999; Kimmel, 2009a). It has been argued
that if a conceptual metaphor motivates one metaphoric expression then further
ones in the same text unit will tend to derive from the selfsame conceptual metaphor or at least cohere with it in part. If one accepts this hypothesis, the issue can
be addressed empirically by measuring how frequently textually cohesive metaphors overlap in their source domains, their target domains, or in both. A count
of ontological similarity between adjacent metaphors elucidates how integrated a
discourse is conceptually. A code system should allow tracing and surveying textual
interaction between cohesive metaphors in a way that captures ontological similarity. Compositional coding provides for this. A first option is to search metaphor
clusters in a hypothesis-driven fashion. Lets assume that we hypothesize a recurrence of cohesive and, at the same time, ontologically coherent path metaphors.
We simply feed the atlas.ti query tool with instructions to look for (a) passages
with metaphor quadruples within a sentence range that (b) have path as a source,
and (c) political action, here by the former British Prime Minister, as a target. This
results in the following hit.
This would, of course, represent a startling volte-face for Tony Blair. And yet
the prime ministers particular skill is the performance of the graceful U-turn,
couched in the language of the moral imperative. His current crusade is to make
African poverty history. Let him start by withdrawing from the two commitments
most harmful to that continent: the EU common policies on overseas aid and
agriculture. (Guardian PD 97)
Possibly we could also run a search for all metaphor quadruples with totally different sources, but a shared EU related target, and get the following hit:
Over the past few years, the EU has become something of a juggernaut and I fear
it has become so concerned with navel-gazing that it has actually lost touch with
the populations of Europe. (Guardian PD 214)
The mappings the EU is a monster, the EU is a body seeing itself and the EU is
an entity moving away from another entity are combined to shed light on
various aspects of the target domain.
Compositional coding can also become the basis for powerful algorithms employed for a (non-hypothesis directed) survey search. In Kimmel (2009b) I calculated the ratio of mixed versus coherent metaphors over the whole corpus, in
order to get a summary measure of how logically integrated the metaphors are on
average. (Technically, this required multiple calculations by hand and inserting
their results into a formula.) The basic idea was to run a search for adjacent metaphor pairs among the 2754 metaphors, while considering all possible similarities
between them. To do this I grouped all codes into similarity based sets (= ontologies). Based on these sets, all adjacent metaphors were checked for image-schematic similarities and for inferential similarities via rich source domains. Then the
pairs were checked for similarities in the targets. Third, all these constraints were
linked with an OR operator to find just any kind of similarity. The study yielded
somewhat unexpected insights on metaphor mixing. Among adjacent metaphors,
mixed metaphors dominated (although by different margins in the two newspapers)! More expectedly, source domain differences were more likely than target
domain differences to account for this fact.
4.4 Implicit narrative linkages and other coherence chains
Metaphors of a corpus can be quasi-narratively connected or related at some other
higher plane. By piecing several data segments together mini-stories enacted in
the source-domain world can be explicated (cf. Kimmel, 2009a; Eubanks, 2000).
For example, the EUs difficulties in integration is stalling and the constitution
being needed to streamline the bulky EU can be intuitively grasped to contribute
to a single mental frame based on metaphoric motion. Hence, an analytic tool
should allow surveying quasi-narrative coherence patterns. When narratively compatible metaphors occur in a single discourse passage the connection is rather
evident. Frequently however we need to infer coherence across non-adjacent text
units. This requires us to recognize as logically compatible, say, two agents in
complementary roles, agents and objects useful to their role, actors and actions,
or triggering actions and outcomes. Researchers frequently aim at reconstructing
multiple chains and overlaps, typically after having completed the executive summary of metaphor formulas (see 3.6). Quinn (1991) sees such overlaps as the royal
35
36
Michael Kimmel
road to analyzing complex cultural schemas. She found that metaphors contribute
to cultural schemas or themes, but also that metaphor types cross-cut discourse
themes. This in turn led Quinn to question key assumptions of CMT about the
discourse shaping role of conceptual metaphors. For quite different conclusions
drawn from a similar set of data, see my abovementioned study on argumentation
with mixed metaphors (Kimmel, 2009b). In any case, this kind of analytic angle is
a powerful and neglected theory building asset.
Let me illustrate such a chain of narrative overlaps from my EU study. Here,
politics is war overlaps with the politicians are boxers (via a force schema
shared by the two targets), which overlaps with the referendal no vote is
a forceful blow (via belonging to a common scenario). This, in turn, overlaps
with the referendal no vote is a natural catastrophe (via a force schema
applied to the same targets), which, finally, overlaps with the no vote is a loud
event (via the metonymical link between sharp movements and sudden noises).
Finding such contingency relations is difficult. However, a valuable starting point
is to browse the synoptic co-occurrence lists for patterns that might stand in a
metonymic relation, scenario relation, etc. Notably by selecting those text units to
which two metaphorical targets attach or paragraphs in which they are adjacent
we can piece together implied chains with more than two constituents. For example, I found that the referendum No is typically the metaphorical death of the
EU constitution brought about by the angry citizens to hurt the elites. We can thus
start from the logic the target items display. Once we have found evidence that
some speakers think of the targets as connected at the argumentative plane, we can
check to see if the source domain logic further reflects this connection.
37
38
Michael Kimmel
the atlas.ti search functions as an entry point for finding metaphor clusters, as
in Maslens study. Even by looking at several dozens of examples only, it became
evident that metaphors commonly mix in highly complex ways and contribute to
larger argument units in a non-predictable fashion. In sum, these analyses create a meta-methodological perspective, which can help prospective users evaluate
whether a second method affords surplus value.
4.6 Benefits of the proposed approach an overview
We have now looked at some methodological desiderata and how my approach
responds to them.
(1) The difficult decision about metaphor formulas is facilitated by deferring it
until after coding. This allows us to capture the relevant inferences and imagery only and safeguards against an over-inclusiveness bias.
(2) Several layers of metaphor logic are reflected in the multi-tier code system, including the layers of image schemas and richer inferential knowledge and, if
desired, abstraction levels as well. This is the basis for later multiple data sorting.
(3) A compositional coding approach allows the researcher to explore how textually cohesive metaphors relate to each other conceptually, be this interaction
source-based, target-based, or rooted in both principles at the same time.
(4) Software assists the exploration of higher-level coherence phenomena, such as
implied narrative chains.
(5) With software we can compare coding methods applied in parallel; I have suggested possible quantitative indicators for this.
5. Summary and conclusion
atlas.ti and related qualitative annotation software enhances the interpretive
skills of the scholar in many ways. While the tool itself decides nothing for you, it
systematizes the procedure via checks and good workflow management, provides
utilities for making decisions (e.g. how do you want your data tabled?), facilitates
multi-level categorization, and provides leverage on large corpora. Software makes
hermeneutic skills reflexive and encourages the systematic presentation of what
sometimes appears as research alchemy.
With the specific aim of analyzing discursive metaphor, the tool adds to each
stage of the process (see Annex II for a summary). During coding itself software
allows complex tagging and facilitates navigating the corpus. At the next stage the
software provides tabulated or filtered data and complex search queries. Finally,
the interpretive analysis benefits in various ways. Search facilities help us crosscheck and generate valid data sets. And, tabulation procedures, in addition to providing quantitative survey data, can be used to prioritize and structure the writeup of a metaphor study. The tables also provide audit trails for tracing where
scholars categorize, select, discard, arrange, or combine data.
Furthermore, I have addressed the important issue of research design and argued for specific strategies to avoid some pitfalls that typically accompany larger
studies. The incremental strategy of compositional coding renders the analysis
flexible while safeguarding against premature decisions about the appropriate
metaphor formulas. The multi-tier coding strategy allows for grouping metaphoric expressions into sets in more than one way. Finally, various nice-to-haves
are supported by software, e.g. comparing metaphor analysis to other qualitative
methods, combining textual cohesion and conceptual coherence of metaphor, and
other complex ways of looking at the data.
In closing, I hope to have contributed to reflexivity in the nascent method
canon of metaphor research and to the visibility of software-enhanced qualitative
approaches.
Acknowledgements
I wish to thank Lynne Cameron and her work group, Christine Irran, Priscilla Hill, Andrew
Goatly, Rudolf Schmitt, Andrzej Pawelec, and two anonymous reviewers for their detailed comments and suggestions to the paper. My thanks also goes to Ray Gibbs and Steve Terrell for a
critique of two earlier versions of the paper.
Notes
1. These approaches are lexical, i.e. they do not aim at the annotation of continuous natural
discourse. Birte Lnneker-Rodman, Astrid Reining, and Wolfgang Settekorn have developed
the Hamburg Metaphor Database and applied it to metaphors collected from a corpus-analytic
search across large German and French corpora. Their analysis tags expressions identified as
metaphorical for sources and targets as defined in the Master Metaphor List by Lakoff and his
associates, allowing for further types. It covers 434 metaphor types and 1657 analyzed tokens.
Goatlys database METALUDE with about 9000 English metaphors has been tested in standard
corpora and can be used for data mining. All included metaphor tokens are tagged in an information-rich way both for source and targets and subsumed under conceptual metaphor types
with not more than 50 tokens (see references).
2. Also note that only some metaphors are explicitly signalled by general markers like, a kind
of , as it were, metaphorically speaking and so forth, as well as by so-called ground terms that
give a clue about the basis of (real or perceived) resemblance between vehicle and topic.
39
40 Michael Kimmel
3. This viewpoint is representative of a more agnostic stance concerning the relation between
language and thought. Its adherents rightly point out that stable conceptual metaphors are one
possibility among several for generating systematic metaphors. This insight converges with the
mixed picture of cognitive mechanisms we can glean from psycholinguistic studies, a mix which,
among other things, depends on whether a metaphor is conceptually creative or novel and
whether it is lexically fixed (Gibbs, 1994; Allbritton et al., 1995; Shen & Balaban, 1999; Keysar et
al., 2000). In any event, discourse research cannot adjudicate such comprehension related issues.
4. Concerning the use of maxqda, readers of German may refer to the brief papers by Marsch
(2007) and Schmieder (2010), which are both rooted in Schmitts (2005) procedure of metaphor
analysis.
5. This applies unless we are researching what Goatly (1997) calls metaphor diversification and
multivalency, i.e. we want to get an overview of targets that go with several different sources or
vice versa.
6. However, if desired, we may store complex search queries as Supercodes that we can later
run again via a click. These allow us to transfer old queries to new projects, create snapshots
of different stages within a single project, or simply help us to access a complex query without
having to type it each time.
7. This does not mean the simple formula is a mere abstraction and not cognitively active for the
discourse participants. It may well be that metaphoric knowledge is ordered at various cognitive
layers in our cultural repository of rhetoric forms.
8. This routine affords efficiency gains and helps to discard noise in the system, but also incurs
risks when the researcher is not operating at the ideal abstraction level. Overly specific code
definitions decrease the number of hits that make it above the threshold. For example, a search
for the more generic politics is war scenario gets 12, but for a sub-mapping politicians are
generals only 2 hits in the Guardian, a result below my threshold. Hence, we should always
check whether we overlook metaphor sets that have been split into several below-threshold
codes without need. When a conflation makes sense for some reason the codes can be merged
or Supercodes used to combine them temporarily (see footnote 6).
9. The seeming absence of several counterpart mappings in the Sun may not only be due to
its highly repetitive journalistic style, but reflect an artifact of method. The combined effects
of smaller sample size and shorter average article length (i.e. a smaller word total) can produce
sub-threshold metaphors that might otherwise have made the threshold in a corpus of equal size
as the Guardians and thus have changed to overall picture.
10. An attractive alternative would be to tag expressions with a finite set of primary metaphors
(Grady, 1997, 1999), of which about 100 have been described. Grady found that many complex metaphors combine two or more primary mappings, e.g. structure is organization
and persisting is remaining erect together define the conceptual metaphor theories are
buildings. Primary metaphors overcome our problem of over-inclusive metaphor formulas,
because they inherently capture only those aspects that get mapped. The potential for creating a
code system is evident: Primary metaphors can coalesce into complex metaphors in shifting alliances and thereby describe a wide range of patterns. If all metaphors could be thus decomposed,
a finite list of basic constituents would be at hand and render a separate coding of sources and
targets dispensable. Instead, two or more primary mappings would be compositionally applied.
This approach shares with one-shot coding that full mappings describe a complex metaphorical
expression at the code level. The difference is that a single click on a code will not retrieve all
hits, but that a co-occurrence search of codes is still needed to describe a complex metaphor
(i.e. a composite of primary mappings). Unfortunately, Grady himself (1997, ch. 7) admits that
a large number of metaphors cannot be described via primary mappings. It is thus no accident
that his approach has not been empirically implemented and that no one has compiled a list of
discourse-tested primary metaphors.
11. Note that a frequent reason why researchers are faced with the question of abstractness is
that metaphor scenarios unfold in a more specific way than their respective core mappings. For
example, the negotiators went to war can be tagged with the core mapping argument is war
but also with the more specific entering into an argument is entering a battlefield.
References
Allbritton, D.W., McKoon, G. & Gerrig, R. (1995). Metaphor-based schemas and text representations: Making connections through conceptual metaphors. Journal of Experimental
Psychology: Learning, Memory and Cognition, 21(3), 612625.
Cameron, L. (2003). Metaphor in Educational Discourse. London: Continuum.
Cameron, L. (2007). Confrontation or complementarity: Metaphor in language use and cognitive metaphor theory. Annual Review of Cognitive Linguistics, 5, 107135.
Cameron, L. (2010). The discourse dynamics framework for metaphor. In L. Cameron & R.
Maslen (Eds.), Metaphor Analysis. Research Practice in Applied Linguistics, Social Science
and the Humanities (pp.7794). London: Equinox.
Cameron, L. & Low, G. (Eds.). (1999). Researching and Applying Metaphor. Cambridge:
Cambridge University Press.
Cameron, L. & Maslen, R. (Eds.). (2010a). Metaphor Analysis. Research Practice in Applied
Linguistics, Social Science and the Humanities. London: Equinox.
Cameron, L. & Maslen, R. (2010b). Identifying metaphors in discourse data. In L. Cameron &
R. Maslen (Eds.), Metaphor Analysis. Research Practice in Applied Linguistics, Social Science
and the Humanities (pp.97115). London: Equinox.
Cameron, L., Maslen, R. & Low, G. (2010). Finding systematicty in metaphor use. In L. Cameron
& R. Maslen (Eds.), Metaphor Analysis. Research Practice in Applied Linguistics, Social
Science and the Humanities (pp.116146). London: Equinox.
Cameron, L. & Stelma, J. (2004). Metaphor clusters in discourse. Journal of Applied Linguistics,
1(2), 736.
Charteris-Black, J. (2004). Corpus Approaches to Critical Metaphor Analysis. Basingstoke:
Palgrave Macmillan.
Chilton, P. (2004). Analysing Political Discourse. London/New York: Routledge.
Cienki, A.J. & Mller, C. (2008). Metaphor and Gesture. Philadelphia: John Benjamins.
Clausner, T.C. & Croft, W. (1997). Productivity and schematicity in metaphors. Cognitive
Science, 21(3), 247282.
Corts, D. (2006). Factors characterizing bursts of figurative language and gesture in college lectures. Discourse Studies, 8(2), 211233.
Deignan, A. (2005). Metaphor and Corpus Linguistics. Amsterdam/Philadelphia: John Benjamins
41
42
Michael Kimmel
Deignan, A. & Semino, E. (2010). Corpus techniques for metaphor analysis. In L. Cameron &
R. Maslen (Eds.), Metaphor Analysis: Research Practice in Applied Linguistics, Social Science
and the Humanities (pp.161179). London: Equinox.
Eubanks, P. (2000). A War of Words in the Discourse of Trade. Carbondale/Edwardsville:
Southern Illinois University Press.
Fauconnier, G. & Turner, M. (2002). The Way We Think: Conceptual Blending and the Minds
Hidden Complexities. New York: Basic Books.
Forceville, C. & Urios-Aparisi, E. (Eds.). (2009). Multimodal Metaphor. Berlin: Mouton de Gruyter.
Friese, S. (2012). Qualitative Data Analysis with atlas.ti. Thousand Oaks: Sage Publications.
Gibbs, R.W. (1994). The Poetics of Mind: Figurative Thought, Language, and Understanding.
Cambridge/New York: Cambridge University Press.
Goatly, A. (1997). The Language of Metaphors. London: Routledge.
Goatly, A. (2007). Washing the Brain: Metaphor and Hidden Ideology. Amsterdam/Philadelphia:
John Benjamins.
Grady, J. (1997). Foundations of Meaning: Primary Metaphors and Primary Scenes. Unpublished
Ph.D. Universiy of California, Berkeley.
Grady, J. (1999). A typology of motivation for conceptual metaphor: Correlation vs. resemblance. In R.W. Gibbs Jr. & G. Steen (Eds.), Metaphor in Cognitive Linguistics (pp.79100).
Amsterdam/Philadelphia: John Benjamins.
Gugutzer, R. (2002). Leib, Krper und Identitt. Eine phnomenologisch-soziologische
Untersuchung zur personalen Identitt. Wiesbaden: Westdeutscher Verlag.
Hellwig, B., Van Uytvanck, D. & Hulsbosch, M. (2010). ELAN Linguistic Annotator, version
3.9.0., http://www.lat-mpi.eu/tools/elan/
Heywood, J., Semino, E. & Short, M. (2002). Linguistic metaphor identification in two extracts
from novels. Language and Literature, 11(1), 3554.
Johnson, M. (1993). Moral Imagination: Implications of Cognitive Science for Ethics. Chicago:
University of Chicago Press.
Keysar, B., Shen, Y., Glucksberg, S. & Horton, W.S. (2000). Conventional language: How metaphorical is it? Journal of Memory and Language, 43, 576593.
Kimmel, M. (2009a). Metaphors of the EU Constitutional Debate Ways of charting discourse
coherence in a complex metaphor field. Metaphorik.de, 17, 49100.
Kimmel, M. (2009b). Why we mix metaphors (and mix them well): Discourse coherence, conceptual metaphor, and beyond. Journal of Pragmatics, 42(1), 97115.
Kimmel, M. (2009c). The EU constitution in a stereoscopic view: Qualitative content analysis
and metaphor analysis compared. In M. Pausch & M. Mokre (Eds.), The Contestation of
Europe (pp.119162). Frankfurt/New York: Campus.
Koller, V. (2003). Metaphor Clusters in Business Media Discourse: A Social Cognition Approach.
Unpublished Dissertation, Vienna: Vienna University of Economics and Business
Administration.
Koller, V., Hardie, A., Rayson, P. & Semino, E. (2008). Using a semantic annotation tool for the
analysis of metaphor in discourse. Metaphorik.de, 15, 141160.
Lakoff, G. (1993). The contemporary theory of metaphor. In A. Ortony (Ed.), Metaphor and
Thought (pp.202251). Cambridge: Cambridge University Press.
Lakoff, G. & Johnson, M. (1980). Metaphors We Live By. Chicago: University of Chicago Press.
Lakoff, G. & Johnson, M. (1999). Philosophy in the Flesh: The Embodied Mind and Its Challenge
to Western Thought. New York: Basic Books.
Lewins, A. & Silver, C. (2007). Using Software in Qualitative Research. A Step-by-Step Guide.
London: SAGE.
Liebert,W.-A. (1997). Stop making sense! Metaphor and perspective in creative thinking sessions
of scientists and scientific radio broadcasts. In W.-A. Liebert, G. Redeker & L.R. Waugh
(Eds.), Discourse and Perspective in Cognitive Linguistics. Amsterdam: John Benjamins.
Low, G. & Todd, Z. (2010). Good practice in metaphor analysis: Guidelines and pitfalls. In L.
Cameron & G. Low (Eds.), Metaphor Analysis. Research Practice in Applied Linguistics,
Social Science and the Humanities (pp.217229). London: Equinox.
Luke, D. & Hart, C. (2007). Cognitive Linguistics in Critical Discourse Analysis. Newcastle:
Cambridge Scholars Publishing.
Marsch, S. (2007). Metaphern des Lehrens und Lernens Metaphernanalyse mit MAXQDA.
In U. Kukartz (Ed.): CAQD 2007. Computergesttzte Analyse Qualitativer Daten. MAXQDA
Anwenderkonferenz (Philipps-Universitt Marburg, 7. bis 9. Mrz 2007) (pp.3443).
Conference Proceedings
Maslen, R. (2010). Working with large amounts of data. In L. Cameron & R. Maslen (Eds.),
Metaphor Analysis. Research Practice in Applied Linguistics, Social Science and the
Humanities (pp.180194). London: Equinox.
Musolff, A. (2004). Metaphor and Political Discourse: Analogical Reasoning in Debates about
Europe. Basingstoke: Palgrave Macmillan.
Pragglejaz Group (2007). MIP: A method for identifying metaphorically used words in discourse. Metaphor & Symbol, 22(1), 139.
Quinn, N. (1991). The cultural basis of metaphor. In J. Fernandez (Ed.), Beyond Metaphor. The
Theory of Tropes in Anthropology (pp.5693). Stanford: Stanford University Press.
Ruiz de Mendoza, F.J. (1998). On the nature of blending as a cognitive phenomenon. Journal of
Pragmatics, 30(3), 259274.
Schmitt, R. (1995). Metaphern des Helfens. Weinheim: Beltz.
Schmitt, R. (2005). Systematic metaphor analysis as a method of qualitative research. The
Qualitative Report, 10(2), 358394.
Schmieder, Ch. (2010). Handwerkliche Schritte der Metaphernanalyse in MAXqda, http://
squaremethodology.com/materials/August/Handwerkliche%20Schritte%20der%20
Metaphernanalyse%20in%20MAXqda.pdf, accessed May 16th 2011
Semino, E. (2008). Metaphor in Discourse. Cambridge: Cambridge University Press.
Semino, E., Heywood, J. & Short, M. (2004). Methodological problems in the analysis of metaphors in a corpus of conversations about cancer. Journal of Pragmatics, 36(7), 12711294.
Shen, Y. & Balaban, N. (1999). Metaphorical (in)coherence in discourse. Discourse Processes,
28, 139153.
Steen, G.J. (1999). Metaphor and discourse: Towards a linguistic checklist for metaphor analysis. In L. Cameron & G. Low (Eds.), Researching and Applying Metaphor (pp.81104).
Cambridge: Cambridge University Press.
Steen, G. (2002). Identifying metaphor in language. A cognitive approach. Style, 36(3), 386407.
Turner, M. (1991). Reading Minds: The Study of English in the Age of Cognitive Science. Princeton,
NJ: Princeton University Press.
Vervaeke, J. & Kennedy, J. (1996). Metaphors in language and thought. Falsification and multiple
meanings. Metaphor & Symbol, 11(4), 273284.
Wengeler, M. (2003). Topos und Diskurs. Tbingen: Niemeyer.
43
44 Michael Kimmel
Zinken, J., Hellsten, I. & Nerlich, B. (2008). Discourse metaphors. In R. Frank, R. Dirven,
T. Ziemke & E. Bernrdez (Eds.), Body, Language and Mind: Volume 2. Sociocultural
Situatedness (pp.363385). Berlin/New York: Mouton de Gruyter.
45
46 Michael Kimmel
encompasses other tropes such as metonymy and irony or if it differentiates between general
types of metaphors (conventional, creative, idioms, etc.) an additional generic-level code can
be attached to each unit, which will serve to filter data segments for a separate analysis of each.
a3. Detailed coding: A code list is created in atlas.ti, usually by generating a logical hierarchy,
i.e. a code forest. (In inductive designs this forest can grow up to a certain point.) Each code
item is specified with coding rules and anchor examples in the atlas.ti commentary function.
Using drag-and-drop, all identified metaphor units are tagged with source and target codes. At
least one code from each sub-hierarchy of the list is applied, including both image-schematic
and rich source domains. This process is guided by the researchers hermeneutic and theoretical
knowledge.
a4. Automated tagging (limited alternative): The atlas.ti auto-coding tool offers an alternative. However, it can be used only when all the possible lexical patterns for a systematic/conceptual metaphor are known in advance and can therefore be compiled into finite a list of target
words. The tool either tags these target words automatically or asks for the researchers confirmation for every token.
B. Identification of mappings via data tables
b1. Heuristic overview of targets: When the coding is completed, a target domain overview is
plotted by atlas.ti, if desired. The relative frequencies of the different targets help the researcher
find a global focus for the ensuing analysis.
b2. Items for write-up /quantitative selection: Printable co-occurrence tables are generated for
three purposes: First, unsystematic mappings can be discarded by choosing a frequency threshold. Second, the reconstruction of systematic/conceptual metaphors [from source-target code
pairs] can start with the largest sets. Third, the table allows tracing connections across similar
metaphors, e.g. to evaluate how aptly the codes differentiate phenomena (see c4).
b3. Data check: Sets of text quotations for unique pairs of source-target codes (= mapping) are
retrieved in the atlas.ti query tool. This can be done systematically or as a spot-check to inspect
if the codes have been deployed in a contextually valid fashion. Some quotations may have to be
regrouped. When a set of quotations is split into two sub-threshold sets, one or both may have
to be discarded as too unsystematic. Conversely, the researcher may merge two sets and thereby
undo overly fine earlier distinctions.
C. Basic interpretive analysis
c1. Compilation of quotations for write-up: A list of all metaphoric expressions for a given
mapping is compiled in the atlas.ti query tool by combing the relevant source and target codes.
The retrieved quotation set is pasted into the write-up either with all meta-data or as a shortlist.
This procedure is repeated for all mappings, following the intended structure of the write-up.
c2. Assignment of a metaphor formula to each set: Expressions that conform to a single mapping (i.e. a systematic/conceptual metaphor) are subsumed under a fitting formula like the EU
is a family. This formula must not be over-inclusive, i.e. it should reflect the actually occurring
quotes only. Typically, anchor examples are selected that epitomize each formula.
c3. Internal differentiation of sets: Some broadly conceived mappings may be grouped into
subsets. This can happen (i) by identifying different scenarios of a core mapping, (ii) by identifying different mapped attributes belonging to a source domain, or (iii) by identifying variants
of an image schema, e.g. container content vs. boundary. New formulas for sub-sets can be
generated if desired.
c4. Executive summary and exploration of cross-linkages: A summary of all metaphor formulas is created that lists all sub-variants under their respective main headings. Similarities
(or not) in entailments or imagery between these sub-variants are now explicated for a critical
examination of discourse coherence. Further similarities across the main headings can equally
be looked into. In both cases, atlas.ti co-occurrence tables can be used to explore how imageschematic and rich source domains cross-cut each other.
c5. Analysis of rhetoric/discourse functions: The researcher works out cognitive, emotive, discursive, and argumentative functions that apply to each formula. In the write-up this typically
follows the descriptive explication of what a given quotation set has in common (before being
added to the executive summary). Per se different metaphors may show some equivalence due
to a common function, e.g. calling someone an animal and a clown are both disparaging. By
a similar token, functional differences under the same set may occur.
D. Optional steps for advanced analysis
d1. Cohesion-based analysis: The query tool locates adjacent metaphor units, i.e. clusters, for
further analysis. In order to select for coherent metaphor clusters searches can be run under additional constraints, e.g. by looking for units belonging to the same rich source, image-schematic source, or target domain code. The results can be fed into a qualitative analysis of metaphor
interaction or used to quantitatively compare coherent vs. mixed metaphors clusters.
d2. Quantitative analysis: Various quantitative output tables are available in atlas.ti. These are
used to display basic data like a word count, metaphor frequency, and metaphor diversity in the
corpus. Quantitative output can also be used to compare sub-corpora, e.g. different newspapers/
interviews or diachronic segments, and discuss why differences arise. All sorts of tabular output
can be fed into Excel or SPSS for advanced work.
d3. Meta-comparison of coding methods: In cases where two qualitative methods have been
applied to a corpus, overlap related queries can be run to investigate the focus and possible
complementarity of the methods.
Authors address
Schallergasse 39/30
A-1120 Vienna
Austria
michael.kimmel@univie.ac.at
47
48 Michael Kimmel
Dr. Kimmel is the team leader of a cognitive phenomenological project that investigates bodily
interaction skills in Tango argentino, Aikido, Feldenkrais, and Shiatsu. Dr. Kimmel is a trainer
for software-assisted qualitative research.
Homepage: www.michaelkimmel.at