The function of metaphor: Developing a corpus-based perspective

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.1 (53-121)

The function of metaphor*

Developing a corpus-based perspective

Anatol StefanowitschUniversity of Bremen

This paper presents a corpus-based approach to investigating the function ofmetaphor, specifically to the question whether the use of metaphoricallanguage is motivated primarily by stylistic considerations or by cognitiveprinciples. The paper focuses on concepts that can be expressed alternativelyby a literal or a metaphorical linguistic expression with the same structuralproperties. Such expressions can be individual words (such as grasp andunderstand, which can both encode the concept understand), or fixedphrases (such as in the heart of and in the center of, which can both encodethe concept in the center of). It turns out that a comparison of thedistinctive collocates of the literal and the metaphorical variant in each caseprovides clear evidence for the hypothesis that metaphorical language has acognitive function.

Keywords: metaphor, function of, cognitive linguistics, corpus linguistics,distinctive collocates, collostructional analysis

. Introduction

There are currently two main hypotheses in the language sciences concerningthe function of metaphorical language: one that assumes the use of metaphor-ical language to be motivated primarily by stylistic considerations, and that Iwill therefore refer to as the stylistic hypothesis; and one that assumes the useof metaphorical language to be a reflection of a general cognitive mechanism,and that I will therefore refer to as the cognitive hypothesis.1

In this paper, I develop a corpus-based method for investigating a partic-ular type of metaphorical expression, with the dual aim of providing initialanswers concerning the plausibility of these two hypotheses and demonstrat-

International Journal of Corpus Linguistics : (), ‒.

‒ ⁄ - ‒ © John Benjamins Publishing Company

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.2 (121-183)

Anatol Stefanowitsch

ing the feasibility of a corpus-based approach to the function(s) of metaphorin general, thus adding to the small but growing body of corpus-linguistic liter-ature on metaphor.2 I show that, in principle, a corpus-based approach allowsus to decide between hypotheses such as the ones just mentioned.

The approach is based on the fact that there are concepts that can be ex-pressed alternatively by a literal linguistic expression or by a metaphorical ex-pression with the same structural properties. Such expressions can be individ-ual words (like grasp and understand, which can both encode the concept un-derstand), or fixed phrases (like in the heart of and in the center of, which canboth encode the concept in the center of). Basically, the approach I developand demonstrate is based on the idea that if we identify the distinctive collo-cates of the literal and the metaphorical variant of five such pairs of expressionsand investigate the semantic differences between them, the results should allowus to decide between the stylistic and the cognitive hypothesis. I will show thatin all cases, the comparison of distinctive collocates provides strong supportfor the cognitive hypothesis.

This paper is structured as follows: in Section 2, I provide the theoreticaland methodological background for the study, i.e. I introduce the two hypothe-ses under investigation, demonstrate the methodology, and formulate some ex-pectations following from the two hypotheses with respect to the methodology.Section 3 presents five case studies of literal/metaphorical pairs of expressionsdisplaying a range of properties relevant in the context of this study. Finally,Section 4 summarizes the results, discusses their repercussions for the corpus-based investigation of metaphor in general, and points out some directions forfurther research.

. A corpus-based approach to the function of metaphor

This section provides the theoretical and methodological background for thepresent study; Section 2.1 provides a brief sketch of the two hypotheses andSection 2.2 introduces the methodology in detail and discusses the expecta-tions that we might derive from each of the two hypotheses with respect to thismethodology.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.3 (183-214)

The function of metaphor

. Two hypotheses concerning the function of metaphor

The view that I refer to as the stylistic hypothesis regards metaphor primar-ily as an extraordinary use of language, an embroidering figure of speech thatis intentionally employed in order to achieve particular aesthetic effects, andthus found mainly in poetry, literary language, public oratory and variousother registers prone to ‘ornamental’ uses of language. This view goes backat least to Aristotle, who states in his Art of Rhetoric that “[i]t is metaphorabove all that gives perspicuity, pleasure, and a foreign air” and then goes onto give advice about how to use metaphor appropriately in order to “ornamentour subject” (Aristotle, Rhetoric Book III, Chapter 2, §8ff.).3 It has since beenthe mainstream hypothesis in literary criticism and rhetoric, often simply as-sumed implicitly rather than being subject to discussion. Authors taking thisview typically either focus on the question whether individual cases of literaryuses of metaphor are appropriate or they proscribe rules for the proper use ofmetaphor (cf. Richards 1936, Ch. 5 for a critical discussion); thus it unfortu-nately remains somewhat unclear what precisely the ‘ornamental’ function ofmetaphor entails. The most succinct statement that I have been able to findconcerning this issue comes from the Encyclopedia Britannica:

Metaphor is the fundamental language of poetry, although it is common onall levels and in all kinds of language. . . . In poetry, a metaphor may performvaried functions from the mere noting of likeness to the evocation of a swarmof associations; it may exist as a minor beauty or it may be the central conceptand controlling image of the poem.

(Encyclopedia Britannica Micropædia, s.v. metaphor)

The view of metaphor that I refer to as the cognitive hypothesis is diametricallyopposed to the stylistic hypothesis. Instead of treating metaphorical expres-sions as ornamental stylistic devices, it regards them as a pervasive feature ofeveryday language More importantly, it views metaphor as a conceptual (ormental) phenomenon, whose “essence . . . is understanding and experiencingone kind of thing in terms of another” (Lakoff & Johnson 1980:5). In otherwords, the main function of metaphorical language is that of enabling us tounderstand abstract concepts (i.e. concepts that are not directly accessible viabodily (sensory) experience) in terms of concrete concepts (i.e. concepts thatare directly accessible to us). This view can also be traced back to antiquity (cf.Jäkel 1999 for a brief historical survey), but the most recent and most detailedaccount of it comes from Lakoff and his colleagues, (e.g. Lakoff & Johnson1980; Lakoff 1987; Grady 1997, 1999, and the contributions in Gibbs & Steen

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.4 (214-268)


1999; cf. Lakoff 1993, for a comprehensive overview). A succinct characteriza-tion is found in Lakoff (1993:244f.):

Metaphor is the main mechanism through which we comprehend abstractconcepts and perform abstract reasoning. . . . Metaphor is fundamentally con-ceptual, not linguistic, in nature. . . . Metaphor allows us to understand a rel-atively abstract or inherently unstructured subject matter in terms of a moreconcrete, or at least more highly structured subject matter.

It should be noted that there is some a priori evidence for this view. First, Lakoffand Johnson (1980) and many others following their approach have shownthat metaphorical language is pervasive in all registers and that it is highly sys-tematic, i.e. that metaphorical expressions can systematically be related to verygeneral mappings of concrete source domains to abstract target domains (e.g.emotions (Kövecses 1986); law (Winter 1989); economics, science, and men-tal activity (Jäkel 1997); religion (Biebuyck et al. 1998); politics (Lakoff 1996;Partington 2003); education (Cameron 2003), etc.) This pervasiveness and sys-tematicity of metaphor could not be accounted for if metaphor were simply astylistic phenomenon.

Second, there is often no literal way of expressing a given concept (for ex-ample, the expression at 5 o’clock is based on the metaphor time is space(cf. Radden 1997), but there is no literal paraphrase). If metaphor were sim-ply ornamental, there should be a literal expression corresponding to everymetaphorical one, since there should always be a literal statement capable ofbeing ‘ornamented’ in the first place.

Finally, it has been shown that metaphors always map more concrete do-mains onto more abstract domains, never the other way around (cf. Lakoff &Johnson 1980:112; Lakoff 1987:281; cf. Jäkel 1997:57–65). For example, wetalk about time in terms of money (as in saving, spending or wasting time), butwe do not talk about money in terms of time (for example, measuring moneyin seconds or minutes). This unidirectional transfer from concrete to abstractdomains has also been observed in processes of diachronic change, both inthe context of the conceptual theory of metaphor (cf. e.g. Sweetser 1990) andin the literature of semantic change and grammaticization in general (cf. e.g.Hopper & Traugott 2003). If metaphor served a purely stylistic function, thisunidirectionality would have to be regarded as an accident; in contrast, thecognitive hypothesis accommodates this fact naturally: since abstract conceptsare more difficult to understand than concrete aspects of the environment, it

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.5 (268-318)


makes sense that the direction of transfer is from the concrete to the abstract, asthe opposite direction would inhibit understanding rather than facilitating it.

. Method

One of the fundamental methods in corpus-linguistic approaches to (lexical)semantics is the analysis of a word’s collocates, i.e. of the words that occur (withabove-chance frequency) in a user-defined span around the word in question(Sinclair 1991:170, 175; cf. Barnbrook 1996:87–106 for discussion). One as-pect of lexical semantics that is particularly suitable to this approach is theinvestigation of the (sometimes very subtle) semantic differences between nearsynonyms (cf. e.g. Kennedy 1991). A sophisticated method for this task is sug-gested by Church et al. (1991), who analyze the near-synonyms strong and pow-erful, taking into account not simply the collocates of each of these adjectives,but actually calculating the degree to which each collocate is associated with theword in question as opposed to its near synonym. In this approach, a ‘distinc-tive’ collocate (Gries 2003) is a word that occurs more frequently than chancewith one of the members of a pair of synonyms given its overall frequency withboth members of the pair.

More recently, Gries and Stefanowitsch (2004a) have suggested that the ap-plication of this general procedure is not limited to the investigation of lexicalsemantics. Instead, they demonstrate that it can be extended to the domain ofgrammatical constructions, where it can shed light on semantic factors influ-encing the choice between near-synonymous constructions (so-called ‘argu-ment structure alternations’, such as the ‘dative shift’, but also other kinds ofalternation, such as that between the s-genitive and the of -construction or thewill-future and the going-to-future).

In the following, I will demonstrate that this procedure can also be uti-lized in investigating the function of metaphor. Note that while in many casesa metaphorical expression is the only choice available to a speaker for ex-pressing a given meaning (cf. Section 2.1 above), in many cases there are lit-eral alternatives, as in the following examples (the metaphorical examples arefrom the Collins Cobuild English Dictionary (CCED), the literal counterpartsare constructed):4

(1) a. [She has] a busy dentist’s practice in the heart of London’s West End≈ She has a busy dentist’s practice in the center of London’s West End

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.6 (318-402)


b. The cost has gone up to $1.95 a minute≈ The cost has increased to $1.95 a minute

c. The government has not yet grasped the seriousness of the situation≈ The government has not yet understood the seriousness of the situ-ation

Given such pairs of literal and metaphorical patterns, we can determine whichlexemes occurring in a given slot best distinguish between the two membersof a pair, i.e. which lexemes are most strongly associated with one of the al-ternatives in comparison to the other. These lists should then allow us to un-cover systematic differences between the literal and the metaphorical expres-sion, and to determine whether the motivation for these differences is morelikely a cognitive one or a stylistic one.

Following the terminology introduced in earlier work (Stefanowitsch &Gries 2003; Gries & Stefanowitsch 2004a, b; cf. also Stefanowitsch & Gries2005), I will refer to a lexeme occurring in a particular slot in a pattern as acollexeme, and to its statistically significant association with one member of apair of patterns as its distinctiveness.5 In order to calculate the distinctivenessof a potential collexeme, one needs four frequencies: the lemma frequency forthe collexeme in the first pattern (in this case, the literal one), the lemma fre-quency for the collexeme in the second pattern (in this case, the metaphoricalone), and the frequencies of both patterns with words other than the poten-tial collexeme in question.6 These are then entered into a two-by-two table andsubmitted to a distributional statistic of one’s choice. Take the pair of patternsin example (1) above; both of its members occur frequently with the noun city.The information needed to determine whether city is a distinctive collexemefor one of the two patterns is shown in Table 1 (for expository purposes, italso gives the expected frequencies for each combination of noun and patternin parentheses). The figures in italics are derived directly from the corpus (theBritish National Corpus, Release 1), the other figures are the results of additionsand subtractions.

Table 1. The distribution of city in in the {center/heart} of NP (BNC-1)

city Other nouns Row totals

in the center of NP 27 (48) 1,075 (1,054) 1,102in the heart of NP 40 (19) 404 (425) 444Column totals 67 1,479 1,546

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.7 (402-457)


If we submit these figures to a distributional statistic such as the Fisher-Yates exact test or the Chi-square test, we find that the the probability of errorfor this distribution is exceptionally small (Fisher-Yates exact: p = 6.24E-08;Chi-square: χ2 = 32.84 (df = 1), p = 1.00E-08).7 This tells us that the noun cityis indeed highly significantly distinctive for one of the two patterns. It does nottell us for which one, however, since any distributional statistic can determinewhether the observed frequencies in one or more cells of a table deviate sig-nificantly from the expected frequencies, but cannot determine the directionof deviation. In order to determine which of the two patterns city is distinc-tive for, we need to determine the direction in which the observed frequenciesof occurrence deviate from the expected ones. In this case, city occurs abouttwice as frequently as expected in the metaphorical pattern, and roughly halfas frequently as expected in the literal one. Thus, city is highly distinctive forthe pattern in the heart of NP.8 In isolation, this finding does not tell us muchabout the function of the metaphorical pattern as opposed to the literal one,but once we repeat the procedure just described for all nouns occurring in theslot in question in either or both of the two patterns, we get complete listsof distinctive collexemes for each of the two constructions, ranked by associ-ation strength. These lists then allow us to investigate the issue in the mannerdescribed above.

A brief comment on measures of association strength is in order at thispoint. In principle, any association measure can be used to calculate distinc-tiveness. In the example above, both the Fisher-Yates exact test and the Chi-square statistic were used, and they yielded results on the same order of mag-nitude. Other statistics that could be directly applied or adapted to the taskinclude mutual information (MI, cf. Church & Hanks 1990), and the log-likelihood coefficient (cf. Dunning 1993). However, given the extremely sparsedata we frequently have to deal with in the kind of procedure outlined above, itis highly desirable to use the exact test, since the distributional conditions forthe approximate tests are frequently not met (cf. the discussion in Stefanow-itsch & Gries 2003:217f.; Gries & Stefanowitsch 2004a). For this reason, allanalyses presented below are based on the Fisher-Yates exact test, whose p-value is interpreted directly as a measure of association strength (cf. Pedersen(1996), Pedersen, Banerjee and Purandare (2003); cf. also Stefanowitsch andGries (2003:238f., Note 6), for more detailed justification).

In order to apply distinctive-collexeme analysis to the issue at hand, weneed a formal context that straightforwardly allows us to compare the collex-emes of the literal and the metaphorical alternative systematically. For example,

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.8 (457-490)


it would be highly problematic simply to compare the collexemes of the expres-sions go up and increase from (1b) above, since these verbs differ in the rangeof complementation patterns they allow: increase can be used transitively (sothat the agent or ‘increaser’ appears as the subject and the ‘thing-increased’ asthe direct object), or intransitively (with the ‘thing increased’ as the subject).For go up, only the latter option is available. If we compared all subject collex-emes of the two expressions, the list for increase would contain many agents forpurely syntactic reasons. If, on the other hand, we compare all ‘thing-increased’collexemes, we would miss potential differences between the transitive and theintransitive use of increase (for example crime can increaseitr, but you cannotincreasetr crime). A similar problem arises with understand and grasp in (1c)above: both may occur with the ‘content understood’ as a direct object or athat-clause, but in addition, understand, but not grasp, may occur with the‘person understood’ as a direct object (cf. Rusty nodded as though she under-stood the old woman (CCED), but not *Rusty nodded as though she grasped theold woman). The problems posed by such formal differences are clearly notinsurmountable; a sufficiently fine-grained analysis paying attention to all for-mal differences and their potential repercussions could certainly be devised.9

However, for the present study, the first of its kind and thus partly intended totest the feasibility of the method in the first place, I will limit the investigationto synonyms that share the same (set of) interpretations in the same formalpattern, as in the case of in the center/heart of NP in (1a) above.

After surveying a range of potential pairs of expressions, I selected five pairsconforming to the grammatical pattern [(Preposition) [(Det) N [PP Preposi-tion NP]]], exemplified by the patterns in (1a) above. The patterns were chosenso that they differ from each other according to a range of factors described indetail in the next section in order to make the study as broad as possible giventhe formal limitations imposed. The five pairs selected are shown in (2a–e):

(2) a. [PP in [NP the [Nom {center/heart} [PP of NP]]]]b. [PP at [NP the [Nom {center/heart} [PP of NP]]]]c. [NP the [Nom {beginning/dawn} [PP of NP]]]d. [NP (Det) {increase/growth/rise} [PP in NPthing-increased]]e. [NP (Det) (Adj) [Nom {understanding/grasp} [PP of NPcontent]]]

This study is largely exploratory, and will therefore refrain from formulatingspecific predictions following from the two hypotheses outlined in the preced-ing section. Instead, the aim is to generate potentially relevant data sets andattempt to relate them to the broad distinction between stylistic and cognitive

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.9 (490-536)


functions. This may in turn lead to more specific hypotheses, which can thenbe tested more rigorously. However, even an exploratory study should be ex-plicit about its expectations with regard to the data at least at a very generallevel, so as to avoid completely subjective post-hoc interpretations that wouldbe unsuitable for developing more specific hypotheses. Let us therefore brieflyoutline these expectations.

Recall that if the stylistic hypothesis is correct, the function of metaphoris to make a text more pleasurable, vivid, beautiful, eloquent or evocative. Itis difficult to derive specific expectations from such a function with regard tosystematic differences in the distinctive collexemes of a literal/non-literal pairof expressions. In the worst case, it might be the case that there are no system-atic differences at all; what is pleasurable, vivid, beautiful, eloquent or evocativemay depend so strongly on the general context in which an expression is usedthat a primarily stylistic motivation cannot show up in the list of collexemes atall. If this were the case, the stylistic hypothesis would be unfalsifiable. How-ever, it seems plausible to assume that, if the stylistic hypothesis is correct, thestylistic motivation for the metaphorical variant will show up in the form of atleast some distinctive collexemes with a clearly literary or poetic flavor, for ex-ample, words referring to emotionally charged concepts, archaic words, wordsfrom a highly literary register, etc.

In contrast, if the cognitive hypothesis is correct, the function of metaphoris to enable us to understand a text in the first place. This function wouldlead us to expect a very specific systematic difference between the conceptsreferred to by the distinctive collexemes of the two expressions: the collex-emes of the metaphorical expression should refer to more complex (i.e. lessdirectly experiencable and thus less easily understandable) concepts than thecollexemes of the literal expression. This prediction may at first glance seemcounterintuitive: we might expect metaphorical expressions to create an addi-tional processing problem for speaker and hearer, and this assumption wouldlead to the opposite prediction, namely that contexts that are already complexshould prefer literal expressions. However, according to the cognitive hypothe-sis, metaphor does not increase processing difficulties (and there is psycholin-guistic evidence that backs up this view, cf. Gibbs (1994)). In fact, the wholepoint about the cognitive hypothesis is that the use of metaphorical languagemay reduce processing difficulties. In other words, the function of metaphorcould be described as one of giving ‘conceptual support’ to our understandingof complex concepts.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.10 (536-591)


Of course, ‘complexity’ is not an uncontroversial and unambiguous tech-nical term, and thus the issue of which concepts are more or less complex thanothers is open to debate; the feasibility of the method depends on the plausibil-ity of the judgments about complexity made in the interpretation of the resultsit yields.

In the case studies presented below, I will base my judgments explicitly ona set of assumptions which hopefully are commonsensical enough to be un-controversial. They are largely based on well-known principles of gestalt per-ception (cf. Wertheimer (1938) for the first formulation of these principles ande.g. Garner (1974) for evidence that percepts that adhere to these principles areeasier to process than those that do not). The assumptions are the following:

Assumption 1: At a particular level of granularity, concepts representingsmall entities are less complex than those representing large entities (be-cause they are easier to conceptualize in their entirety). This follows fromthe gestalt principle of figure-ground perception.

Assumption 2: Concepts representing entities that have a simple shapeand/or have a clear boundary are less complex than those representingentities with complex shapes or fuzzy boundaries (because they are moreeasily delineable). This follows from the gestalt principles of closure andsimplicity.

Assumption 3: Concepts constituting holistically perceived wholes are sim-pler than concepts constituting to details or component parts of wholes.This follows from the gestalt principle that configurations are simpler toperceive than their parts.

Assumption 4: Concepts representing individually attributable human ac-tions are simpler than concepts representing non-attributable processes(because they are closer to our everyday experience of ourselves and ourinteraction with the environment). A more general assumption that mayinclude this one is Assumption 5.

Assumption 5: Concepts representing concrete objects are less complexthan those representing abstractions (because they are more accessible todirect experience).10

Of course, at least in the context of the present study (but probably in general),complexity is not an absolute notion; after all, we will investigate precisely thosecases where metaphorical expressions are not the only way to express a givenconcept, but where a literal alternative is available. Thus, complexity is best

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.11 (591-665)


understood as a gradual phenomenon: some concepts are more directly acces-sible to experience and easier to understand and thus less in need of conceptualsupport by metaphor, than others.

Finally, note that in the case studies below I will focus on the questionwhether the distinctive collexemes show differences in terms of style (in thesense in which the term is used in the present study) or complexity (in terms ofthe five assumptions just stated). This is not meant to suggest that there are noother semantic differences that could potentially be observed and discussed.On the contrary, there clearly are such differences, and in a few places I willpoint these out. However, a detailed analysis is not the aim of this paper andwill have to await a different study.

. Case studies

For each of the patterns, all instances were retrieved from the British NationalCorpus (Release 1) by concordancing the head noun (center, heart, beginningetc.) followed by the respective preposition. False hits, i.e. hits not correspond-ing to the pattern under investigation, were manually removed. This searchstrategy ensured that all instances of each pattern were exhaustively identified.For each pair of patterns, the collexemes (the noun heading the NP slot, andin one case, the adjective modifying the head noun) were manually identified,lemmatized, and frequency lists were generated. These were then submitted tothe distinctive collexeme analysis as described in the preceding section.

. The pattern in the center/heart of NP

The first pattern that I will investigate is the one in (3); the examples in (4a,b) are evidence that the alternatives can function as rough paraphrases of eachother in at least some contexts:

(3) in the {center/heart} of NP

(4) a. [The station] was provided with platform canopies and flowers toenhance its attractive position beside the lake in the centre of the city.

b. The princely town house stands beside a broad pond in the heart ofthe city.

The metaphorical use of heart in (4b) is motivated by the metaphorical map-ping the center is the heart. This mapping is part of a more general

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.12 (665-801)


object-as-organism metaphor, which motivates a number of expressions in avariety of domains, including locative and partitive expressions (e.g. the headof the line/stairs/chart/table, the mouth of a river/cave/hole/bottle, the teeth of acomb/saw/cog, the leg of a table/chair, the foot of the stairs/page/mountain, etc.,all examples from CCED). Note that the specific manifestations of the object-as-organism mapping are all partial and unsystematic (cf. Lakoff & Johnson1980:52ff.), i.e. there are generally just one or two body-part terms that can beapplied to a specific object (for example, stairs have a head and a foot, but nota mouth, teeth, shoulders, etc.; mountains have a foot and a shoulder, but nohead, leg, etc.), and it is not possible to predict which body-part terms may beapplied to which object (e.g., why is the top of a mountain not called head, whyis the blade of a saw not called gums, etc.). The object-as-organism mappingis thus not the type of systematic and very generally applicable metaphor thatproponents of the cognitive hypothesis typically point to as support of theirclaims; according to Lakoff and Johnson, it is one of a set of idiosyncratic map-pings that “do not interact with other metaphors, play no particularly inter-esting role in our conceptual system, and hence are not metaphors we live by”(1980:55).11 Thus, the pair of patterns in (4a, b) is a perfect candidate to beginmy investigation without stacking the odds in favor of the cognitive hypothe-sis (I will discuss examples of more systematic and general metaphors furtherbelow). Table 2 lists the distinctive collexemes of in the center of NP and in theheart of NP.

The significantly distinctive collexemes of in the heart of NP consist exclu-sively of nouns referring to geographical locations, in the form of both com-mon and proper nouns. This semantic class also dominates the marginally sig-nificant and non-significant collexemes, with a single exception (season). Thedistinctive collexemes (significant and non-significant) of in the center of NPalso consist mainly of nouns referring to locations, but the nature of these lo-cations is more heterogeneous than in the case of the metaphorical pattern,in that not all of them are geographical locations. Instead, there are a num-ber of nouns that refer to parts of buildings (room, floor, wall, chamber, stage).In addition, there are several nouns that do not refer to locations, but to furni-ture (table), household items (oven, plate, machine), a group of people (defence,used here in the sports sense), and a body part (forehead).

Let us relate these findings to the expectations formulated in Section 2.1above. The main difference between the two patterns seems to be that the literalpattern occurs with all kinds of location nouns, while the metaphorical pat-tern is restricted to geographical location nouns. If we assume the cognitive

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.13 (801-801)


Table 2. Distinctive collexemes of in the {center/heart} of NP (BNC-1)

in the centre of NP (n = 1,102) in the heart of NP (n = 444)Collexeme Distinctiveness Collexeme Distinctivenessroom (78/1) 3.70E-11 city (27/40) 6.24E-08town (74/9) 5.58E-05 countryside (1/12) 2.72E-06table (25/0) 1.95E-04 country (5/17) 2.81E-06village (41/3) 3.04E-04 area (3/14) 6.13E-06floor (16/0) 4.30E-03 forest (1/10) 2.90E-05circle (15/0) 6.06E-03 London (11/17) 3.69E-04square (12/0) 1.69E-02 West End (0/6) 5.48E-04wall (10/0) 3.35E-02 City of London (0/5) 1.92E-03defence (9/0) 4.71E-02 estate (0/5) 1.92E-03oven (9/0) 4.71E-02 mountain (0/5) 1.92E-03which (9/0) 4.71E-02 region (0/5) 1.92E-03

capital (2/7) 3.17E-03kingdom (0/4) 6.74E-03England (1/5) 8.80E-03France (2/5) 2.35E-02business quarter (0/3) 2.36E-02centre (0/3) 2.36E-02city centre (0/3) 2.36E-02community (0/3) 2.36E-02Cotswolds (0/3) 2.36E-02Surrey (0/3) 2.36E-02

0.05 < p < 0.1 0.05 < p < 0.1

Darlington (8/0), field (8/0), courtyard (7/0),plate (7/0), road (7/0), Oxford (11/1)

valley (1/3), wood (1/3), Ayrshire (0/2), bed-sitland (0/2), Cheshire (0/2), constituency(0/2), docklands (0/2), East Oxford (0/2),land (0/2), Manhattan (0/2), Mayfair (0/2),nation (0/2), rain forest (0/2), scenery (0/2),season (0/2), shopping area (0/2), Snowdonia(0/2), Somerset (0/2), Student Village (0/2),The Hague (0/2), Whitehall (0/2), YorkshireDales (0/2), Europe (9/8)

0.1 < p < 0.25 0.1 < p < 0.25

Funchal (6/0), hall (6/0), machine (6/0), bay(5/0), chamber (5/0), forehead (5/0), island(5/0), line (5/0), site (5/0), space (5/0), stage(5/0)

empire (1/2), Newark (1/2), territory (1/2),Rome (3/3)

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.14 (801-840)


hypothesis, this can reasonably directly be related to our expectation concern-ing complexity. The non-geographical locations are much smaller than the ge-ographical ones; assuming that larger entities are more complex than smallerones (because they are more difficult to conceptualize in their entirety, cf. As-sumption 1), the cognitive hypothesis accounts for the fact that the latter donot require metaphors to make them understandable to us, and hence do notoccur in the metaphorical pattern. The same is true of the other words limitedto the literal pattern. A room, a table, an oven, etc. can easily be conceptual-ized in their entirety and thus are easier to process than a city, a countryside,or a forest. This line of reasoning can be extended to a comparison of the geo-graphical location nouns occurring in the two patterns: those occurring inthe literal pattern all refer to rather small built-up areas (village, town, Darling-ton, Oxford), while those occurring in the metaphorical pattern refer to largerbuilt-up areas (city, London, West End), large natural areas (countryside, area,forest), and countries (e.g. England, France).

A second, related aspect concerns the geometrical properties of the dis-tinctive collexemes’ referents (cf. Assumption 2). The entities referred to bythe collexemes of the literal pattern all have comparatively simple, clearly rec-ognizable shapes with easily determinable centers: most of them are eithersquare (like rooms, tables, floors, squares, walls, ovens, fields, courtyards, etc.),or round (like circles, plates, and our idealized representations of towns, vil-lages, islands, etc.). In contrast, the entities referred to by the collexemes of themetaphorical pattern mostly have complex shapes that are difficult to concep-tualize or describe, even if we know them (like London, England, Surrey, etc.),or they do not have clearly delineable shapes at all (like city, countryside, coun-try, area, etc.). Here, in the heart of NP allows us to understand the notioncenter in a functional rather than a strictly geometrical way.

In sum, this case study has yielded results that are straightforwardly inter-pretable in terms of the cognitive hypothesis, even though the metaphoricalmapping underlying in the heart of NP is not one of the general, systematicmappings that proponents of the cognitive hypothesis typically point to as ev-idence for their claims. The metaphorical pattern is associated with distinctivecollexemes referring to concepts that are plausibly regarded as more complexthan those referred to by the literal pattern’s distinctive collexemes, and hencethe notion of conceptual support is confirmed.

This does not mean that stylistic factors can be completely ignored in thechoice between the two alternatives; clearly, the expression in the center of NPis stylistically neutral and likely to be found in all genres and registers, while

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.15 (840-900)


in the heart of NP is associated with written genres. Thus, at a very generallevel, stylistic considerations clearly play a role. However, this role cannot beseen as support for the stylistic hypothesis, as stylistic considerations merelyinfluence the availability of the choice between the literal and the metaphoricalexpression – they do not appear to influence the choice itself.

. The pattern at the center/heart of NP

The preceding section compared two patterns that typically have a concretenoun in the NP slot. In this section, I will investigate a superficially similar setof patterns that occur almost exclusively with abstract nouns. The patterns areshown in (5), the examples in (6a, b) show that the alternatives may occur inthe same contexts and can thus be regarded as rough paraphrases:

(5) at the {center/heart} of NP

(6) a. Curricular considerations are therefore at the centre of the decision-making process.

b. All this may seem mere words, but words and their meaning are at theheart of the planning process.

Since the majority of nouns occurring in both patterns refer to abstract enti-ties, both patterns are actually metaphorical in this case. Take examples (6a, b):a process has neither a heart (since it is not an organism), nor a center (since it isnot even a physical entity). The metaphorical use of center is motivated by thevery general image-schematic metaphor central is important; other exam-ples of this mapping are The whole thing was . . . tangential to the real worldof business (CCED), The new category was peripheral to their main businessinterests, and of course the adjective corresponding to center, as in Arms controlwill become even more central to military preoccupations. According to Lakoff(1987), the existence of this mapping is based on our experience of living or-ganisms and the relative importance of their more central and more peripheralcomponents: “Injuries to central parts are more serious (i.e., not mendableand often life threatening) than injuries to the peripheral parts. Similarly, thecenter defines the identity of the individual in a way that the periphery doesnot.” (Lakoff 1987:274). The metaphorical use of heart is a further extensionof this image-schema, based on the personification metaphor discussed in thepreceding section.

Since both patterns are metaphorical, the question arises as to what differ-ences we might expect between them in light of the cognitive and the stylistic

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.16 (900-1042)


hypotheses. The stylistic hypothesis might lead us to expect differences in emo-tionality, since the noun heart is presumably more emotionally charged thanthe noun center. In contrast, the cognitive hypothesis would once again sug-gest that the main differences will be found in the complexity of the distinctivecollexemes’ referents. Although both patterns are metaphorical, the central-as-important mapping is rather schematic, and based on an abstraction overmany different types of experience; it thus adds less to our understanding of ab-stract entities than the additional, more specific heart-as-center mapping,which is much more directly experientially based. Thus, the expectations of thecognitive hypothesis are very similar to those formulated in the preceding sec-tion: at the center of NP should occur with nouns referring to concepts that areeasier to understand than those referred to by the collexemes of at the heart ofNP. Table 3 lists the distinctive collexemes for the two patterns.

Again, there is little evidence for the stylistic hypothesis. With the possibleexception of first-ranked Europe (which in the examples in question refers toEuropean politics, not to the actual geographical unit, as in The Prime Ministersaid ... that he wanted Britain to be at the heart of Europe), none of the distinctivecollexemes of at the heart of NP can plausibly be interpreted as being moreprone to occur in emotionally colored contexts than the collexemes of at thecenter of NP.

In contrast, an interpretation in terms of the cognitive hypothesis is onceagain possible and plausible: note that most of the significantly distinctivecollexemes of at the centre of NP refer to human activities and human-scaleevents (row, scandal, attention, controversy, storm – used exclusively in the senseof ‘controversy’/‘outrage’ in this pattern –, event, allegation, and case). In con-trast, most of the distinctive collexemes of at the heart of NP come from the do-main of abstract thoughts or ideas (Europe, defence – mostly used in the sense of‘defensive strategy in a sports game’ –, problem, matter, appeal, difficulty, policy,experience, and strategy). It is plausible to assume that human-scale events andactivities are more directly accessible to experience and easier to understandthan abstract thoughts and ideas, and consequently less in need of conceptualsupport by metaphor (cf. Assumptions 4 and 5). As in the previous section,then, there is some support for the cognitive hypothesis.

Next, note the semantic class formed by stage, web, circle, sphere, and thing,which are all distinctive for the more schematic at the center of NP. Note thatthese nouns are themselves rather schematic terms, which, in the corpus, areall used metaphorically (for example, the political stage, a web of relationships,a widening circle of freedom) or in their mathematical sense (in the case of circle

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.17 (1042-1042)


Table 3. Distinctive collexemes of at the {centre/heart} of NP (BNC-1)

at the centre of NP (n = 762) at the heart of NP (n = 896)Collexeme Distinctiveness Collexeme Distinctivenessrow (23/1) 1.95E-07 Europe (1/29) 2.09E-07scandal (15/2) 3.59E-04 defence (2/20) 2.28E-04attention (9/0) 8.92E-04 problem (0/13) 3.22E-04controversy (14/3) 2.29E-03 matter (0/9) 3.86E-03storm (10/1) 2.62E-03 appeal (0/8) 7.17E-03universe (20/7) 2.69E-03 difficulty (0/8) 7.17E-03stage (7/0) 4.27E-03 business (0/7) 1.33E-02galaxy (9/1) 5.25E-03 religion (0/7) 1.33E-02web (10/2) 9.24E-03 government (2/11) 2.25E-02event (6/0) 9.32E-03 policy (3/13) 2.31E-02group (6/0) 9.32E-03 experience (0/6) 2.47E-02circle (8/1) 1.04E-02 strategy (1/8) 3.37E-02allegation (5/0) 2.04E-02case (9/3) 4.07E-02sphere (4/0) 4.44E-02thing (15/8) 4.89E-02

0.05 < p < 0.1 0.05 < p < 0.1plan (5/1), bid (3/0), crossroads (3/0), deci-sion-making (3/0), earth (3/0), field (3/0),maelstrom (3/0), mosaic (3/0), ring (3/0),speculation (3/0), trade (3/0), wake (3/0)

process (3/11), education (1/7), book (0/4),economy (0/4), interpretation (0/4), opera-tion (0/4), policy-making (0/4), success (0/4),development (1/6), philosophy (1/6), project(1/6)

0.1 < p < 0.25 0.1 < p < 0.25power (4/1), argument (5/2), area (6/3),alert (2/0), all (2/0), black hole (2/0), brawl(2/0), bust-up (2/0), ceiling (2/0), chamber(2/0), crowd (2/0), drama (2/0), each (2/0),frame (2/0), hall (2/0), incident (2/0), junc-tion (2/0), line (2/0), mansion (2/0), me(2/0), monument/2/0), mound (2/0), orgy(2/0), page (2/0), picture (2/0), police (2/0),preoccupation (2/0), scare (2/0), screen (2/0),section (2/0), sensation (2/0), side (2/0), siege(2/0), sociology (2/0), table (2/0), Ulthuan(2/0), wrangle (2/0)

organisation (1/5), proposal (1/5),<software-title> (0/3), approach (0/3),change (0/3), charter (0/3), darkness (0/3),discourse (0/3), England (0/3), enterprise(0/3), exhibition (0/3), faith (0/3), law(0/3), reality (0/3), relation (0/3), report(0/3), school (0/3), thinking (0/3), tragedy(0/3), community (2/6), work (5 10), being(1/4), crisis (1/4), issue (1/4), research (1/4),dispute (3/7), what (3/7)

and sphere); their strong preference for the schematic pattern is probably bestinterpreted as evidence for a kind of ‘metaphorical cohesion’, i.e. a reluctanceto mix metaphors (or, in this case, levels of schematicity) unnecessarily (cf.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.18 (1042-1115)


Martin (to appear) for solid corpus evidence of such cohesion phenomena; theexistence of such phenomena may be the source of the prescriptive rule not tomix metaphors).

Finally, a methodological remark: note that the systematic differencesobserved for the distinctive collexemes also occur consistently among themarginally significant collexemes, and to some degree even among the non-significant distinctive collexemes; this is evidence for the suggestion made inStefanowitsch and Gries (2003) and Gries and Stefanowitsch (2004a) that whenusing Fisher-Yates exact p-values as association measures, the usual significancethreshold of p<0.05 should not be seen as an absolute cut-off point, but that itis the ranking of collexemes provided by the method that is most important.

. The pattern at the beginning/dawn of NP

The preceding sections have dealt with spatial metaphors; in this section, I willlook at an example of a literal-metaphorical difference from the domain oftime. The patterns in question are shown in (7), evidence that they are roughparaphrases in some contexts is shown in (8):

(7) at the {beginning/dawn} of NP

(8) a. Communal life survived until the beginning of the nineteenth cen-tury and traditions peculiar to that way of life had lingered intothe present.

b. It is ironic that it has taken until the dawn of the 21st century to re-alise that the best methods of utilising . . . our woodlands are thoseemployed a millennium ago.

The metaphorical use of dawn is motivated by the mapping a timespan isa day, which, as far as I know, has not been discussed in the literature. It isprobably related in some way to the mapping lifetime is a day posited byKövecses (2002:44). Kövecses does not discuss possible motivations for the ex-istence of this mapping, and, interestingly, the examples he mentions are allfrom poetic language, for example, Shakespeare’s Sonnet No. 73: In me thouseest the twilight of such day/As after sunset fadeth in the west,/Which byand by black night doth take away,/Death’s second self, that seals up all in rest.In this example, twilight corresponds to old age and night correspondsto death. While these metaphors are quite widespread in Modern Englishpoetry,12 it is difficult to find examples from everyday language. The same istrue of the more general a time period is a day; note that only the words eve

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.19 (1115-1239)


and dawn/dawning are used systematically (compare the eve of the civil war, thedawn of the 20th century with the non-occurring *the morning/noon/evening ofthe civil war, *the dusk of the 20th century). It seems, then, that we are dealingwith a metaphorical mapping here that is also partial, but in a different sensethan the mappings discussed in the preceding sections: it seems to be limited,for the most part, to poetic language. Thus, it provides an extreme test casefor the two hypotheses – if stylistic factors have any substantial influence, wewould expect to see it with a mapping that is restricted to literary forms oflanguage. In other words, the odds are stacked even more strongly against thecognitive hypothesis here. Of course, this does not mean that the cognitive hy-pothesis has nothing to say about the mapping in question. As for a possibleexperiential basis of the more general mapping, it is clearly provided by ourdirect repetitive experience of days with their natural beginning and end. Thisnatural, easily perceivable delimitation of a day should make it a useful conceptin understanding less clearly delimited time periods. Thus, our expectation ac-cording to the cognitive hypothesis is that the metaphorical the dawn of NPshould be associated with such non-delimited time periods, while the literalthe beginning of NP should be associated with time periods that have a clearlydefined beginning and end. Table 4 lists the distinctive collexemes of the twoexpressions.

As the semantics of the two patterns would lead us to expect, the lists ofdistinctive collexemes for both expressions are dominated by nouns referringto time spans. For the literal expression, the significant collexemes are bothfrom this semantic class (year and century), as are most of the marginally sig-nificant expressions (end, period, term, week, month, May, and September); theonly exceptions are an event (war), and chapter, which can be seen as a quasi-spatial use or as a metonymy whereby chapter also stands for an event (thereading of the chapter). In addition, although the individual names of monthsand the noun month itself are not significant collexemes of the literal pattern,the category ‘names of months’ (i.e. January, February, etc.) as a whole is sig-nificantly associated with this pattern (486/0, p = 7.79E-07); the same is truefor the category ‘names of years’ (i.e. 1840, 1970, etc.) (288/0, p = 2.95E-04).

For the metaphorical pattern, there are also many nouns referring to timespans (time, age, era, beginning, future), and a number of nouns referring toevents (dream, enlightenment, awakening, etc.); in addition, there are a numberof time spans metonymically referred to by situations or entities (such as civ-ilization, history, mankind, regime, or culture).13 Also note that the literal useof day has been included in the count; since it is the only word capable of oc-

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.20 (1239-1239)


Table 4. Distinctive collexemes of the{beginning/dawn} of NP (BNC-1)

at the beginning of NP (n = 4,232) at the dawn of NP (n = 117)Collexeme Distinctiveness Collexeme Distinctivenessyear (315/1) 1.35E-03 civilisation (1/19) 6.36E-30century (385/4) 1.65E-02 time (36/13) 2.30E-10

history (14/9) 3.23E-09age (10/7) 1.33E-07dream (1/4) 2.44E-06mankind (0/3) 1.90E-05era (37/8) 2.02E-05day (29/7) 3.73E-05consciousness (0/2) 7.18E-04enlightenment (0/2) 7.18E-04regime (0/2) 7.18E-04culture (1/2) 2.12E-03

0.024 < p < 0.05awakening (0/1), beginning (0/1), being(0/1), collectivism (0/1), Dino (0/1), easter(0/1), future (0/1), hope (0/1), knowledge(0/1), language (0/1), light (0/1), litera-ture (0/1), memory (0/1), mind (0/1), music(0/1), people (0/1), philosophy (0/1), relation(0/1), stardom (0/1), teenage (0/1), torment(0/1)

0.05 < p < 0.25 0.05 < p < 0.25chapter (101/0), war (93/0), end (88/0), pe-riod (61/0), term (61/0), week (58/0), month(57/0), May (55/0), September (53/0)

democracy (1/1), experience (1/1), man(1/1), Surrealism (1/1), art (2/1), creation(3/1), revival (5/1), 1990s (6/1)

0.25 < p < 0.50 0.25 < p < 0.50career (49/0), April (48/0), book (47/0), De-cember (45/0), June (44/0), October (44/0),session (43/0), February (40/0), March(40/0), life (38/0), August (36/0), 1990(35/0), July (33/0), reign (31/0), decade(30/0), section (29/0), 1991 (27/0), 1992(27/0), process (26/0)

summer (11/1), world (18/1)

Note: Due to the vast difference in frequency of the two expressions, it is very difficult forthe distinctive collexemes of the much more frequent literal pattern to reach the significancethreshold of five percent even if they do not occur at all in the metaphorical one, whileanything that occurs exclusively in the metaphorical pattern even once will automaticallyreach this threshold. Therefore, the latter have been listed separately.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.21 (1239-1289)


curring with the literal meaning of dawn in this pattern, omitting it would notmake much of a difference.

With respect to the cognitive and the stylistic hypotheses, an interpretationalong the same lines as that proposed for in the {center/heart} of NP suggests it-self (cf. Assumptions 1 and 2): The events and time spans referred to by thecollexemes of the literal pattern (year and century as well as ‘names of years’and ‘names of months’) are much shorter and much more clearly delineatedthan those referred to by the distinctive collexemes of the metaphorical expres-sion (civilisation, time, history, age, etc.), and this generalization holds even ifwe include the less significant and non-significant collexemes (which gives usweek, month, and many names of individual months and years for the literalpattern, compared to nouns like democracy, creation, etc. for the metaphoricalpattern).

Thus, the differences between the metaphorical and the literal expressionfor the beginning of something can once more be plausibly related to the cog-nitive hypothesis, even though the metaphorical mapping in question is plau-sibly regarded as more typical of literary registers. Again, this does not meanthat we can simply negate the influence of register on metaphorical language:as before, there are clearly registers where the metaphorical at the dawn of NPwould seem out of place and is thus not available as a choice (while at the be-ginning of NP is register-neutral and thus always available). However, it shouldagain be stressed that this is by no means evidence for the stylistic hypothesis(cf. the arguments in Section 3.1 above).

. The pattern (ADJ) increase/rise/growth in NP

In this section, I will extend the methodology introduced in the preceding sec-tions slightly in two directions. First, I will look not only at one position forpotential collexemes, but at two: in addition to the prepositional object slotinvestigated in the preceding sections, I will also take into consideration thepotential adjective slot in the pre-modifier position of the head noun. Sec-ond, and more importantly, I will explore not only differences between a literaland a metaphorical expression, but I will consider two alternative metaphoricalexpressions and also consider differences between these.

The pattern under investigation is shown in (9); the examples in (10a–c)show that the alternatives are indeed rough paraphrases at least in some cases(note that, unlike at the heart of NP and at the dawn of NP, the metaphoricalexpressions do not have any literary flavor):

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.22 (1289-1347)


(9) (Article) (ADJ) {increase/rise/growth} in NP

(10) a. The 1990 budget ... included a 40 per cent increase in welfare spend-ing.

b. The 1989 defence budget called for 1.7 per cent growth in spending.c. ... a ‘transitional’ budget, which forecast a 29.8 per cent rise in state

spending.

The pattern in (10b) is based on the very general mapping more is up. Otherexpressions instantiating this mapping are, for example The number of bookspublished each year keeps going up; My draft number is high, or the verb corre-sponding to rise, as in My income rose last year (examples from Lakoff & John-son 1980:15f.; cf. also Lakoff 1987:276f. for discussion). According to Lakoffand Johnson, the basis for the existence of this mapping is our experience that“[i]f you add more of a substance or of physical objects to a container or pile,the level goes up” (Lakoff & Johnson 1980: 16).

The pattern in (10c) is based on the more specific mapping abstractdevelopment is natural physical growth, which according to Kövecsesis based on the more general complex abstract systems are plants (cf.Kövecses 2002:98ff.; note that Kövecses uses the label complex abstract systemsas a cover term for a wide variety of target domains), but which may be bet-ter described more generally as (abstract) entities are organisms, sincethere are extremely few plant-specific expressions instantiating the mapping(Kövecses mentions the branch of a company), while there are many expres-sions related to the life cycle of organisms in general that are mapped ontoabstract entities (e.g., This criticism is born out of ignorance, The content of theNBC programmes has ceased to be lifeless, a government programme to rescuethe country’s ailing economy, The tradition died out shortly afterwards, etc.).Kövecses sees the basis for the existence of this mapping in our “rich knowl-edge” about plants; more generally, we might say that the mapping is based inour knowledge about organisms (including ourselves) and their life cycles.

On the basis of these claims, we can formulate some general expectationsabout differences between the contexts in which each of these two mappings ismore likely to be found if they serve a cognitive function.

First, note that the more-as-up mapping is more schematic than thesystem-as-organism mapping; thus, we might expect the latter to providemore conceptual support than the former. If this is true, nouns associated withgrowth in NP should in general refer to more complex concepts than nouns as-sociated with rise in NP (recall that the distinction between a more schematic

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.23 (1347-1526)


and a more specific metaphor is similar to the one between a literal expressionand a metaphorical one, cf. Section 3.2 above).

Second, growth, as a natural process, is comparatively slow and unlikelyto change direction suddenly, making it relatively predictable. In contrast, al-though the adding of a substance or objects to a container or pile may bea natural process (as with rain falling on a lake, snow accumulating on theground, etc.), it does not have to be (as when humans manipulate objects orsubstances). As a consequence, the process of rising is less clearly associatedwith a particular speed than that of growing, and it is not inherently direc-tional, making it less predictable (we can remove objects from a pile just aseasily as adding them). Thus, we might expect the pattern growth in NP to beassociated with slower and more predictable processes than the pattern rise inNP. Note that states and slow, predictable processes may be regarded as lesscomplex than fast, unpredictable processes (based on the gestalt principles ofgood continuation and stability, cf. Wertheimer 1938).

Before we discuss the results in light of these expectations, however, let usbriefly compare the two metaphorical expressions to the literal one at a gen-eral level. Table 5 lists the significant distinctive collexemes of the literal in-crease in NP and the metaphorical {growth/rise} in NP (the figures for the twometaphorical expressions were combined and overall distinctive collexemeswere calculated; see below for a separate analysis of the two).

The differences between the literal pattern and the combined metaphoricalpatterns are much more subtle than in the case of the patterns investigated inthe preceding sections, but on closer inspection, at least two noteworthy pointsof divergence can be observed. First, note that many of the collexemes on bothlists come from the domain of economics (in the case of the literal patternabout a quarter, in the case of the metaphorical patterns, almost half of thesignificantly distinctive collexemes). If we focus on this semantic domain, wecan observe that many of the literal pattern’s collexemes refer to (aspects of)economic processes that are fairly directly influenced by human agents (third-ranked funding, and money supply, contribution, taxation, allowance, pay, childbenefit, competition, and borrowing). In contrast, most of the collexemes asso-ciated with the metaphorical patterns refer to (aspects of) economic processesthat are less easy to influence by human agents (the only exceptions being 24th-ranked export, and wage, sale, and income). It is not implausible to relate thisto the cognitive hypothesis (cf. Assumption 4): human activities are presum-ably easier to understand than processes like those determining prices, inter-est rates, unemployment, the gross national product, etc., which, although ulti-

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.24 (1526-1526)


Table 5. Distinctive noun collexemes of {increase vs. growth/rise} in NP (BNC-1)

increase in NP (n = 6,159) {growth/rise} in NP (n = 3,247)Collexeme Distinctiveness Collexeme Distinctivenesstime (38/1) 1.40E-06 economy (0/34) 1.76E-16amount (59/7) 7.70E-06 price (148/177) 5.54E-14funding (37/2) 1.49E-05 interest rate (37/76) 1.02E-12resource (25/1) 2.40E-04 unemployment (61/99) 1.04E-12activity (81/18) 2.42E-04 market (5/34) 1.34E-11money supply (80/18) 3.06E-04 GDP (6/34) 5.96E-11contribution (28/2) 4.12E-04 sector (4/24) 3.21E-08speed (18/0) 4.85E-04 employment (26/49) 4.69E-08term (18/0) 4.85E-04 temperature (26/39) 2.25E-05frequency (22/1) 7.65E-04 profit (55/59) 1.06E-04number (545/228) 1.11E-03 GNP (1/10) 1.80E-04risk (21/1) 1.12E-03 importance (5/15) 2.57E-04power (62/14) 1.49E-03 earning (21/30) 3.21E-04secretion (15/0) 1.73E-03 faith (0/7) 5.82E-04taxation (22/2) 3.46E-03 inflation (11/19) 1.18E-03synthesis (18/1) 3.51E-03 industry (1/8) 1.25E-03flow (17/1) 5.11E-03 output (70/62) 1.97E-03density (12/0) 6.19E-03 demand (113/90) 2.15E-03height (12/0) 6.19E-03 living standard (1/7) 3.25E-03size (76/22) 6.41E-03 business (18/23) 3.70E-03capacity (20/2) 6.86E-03 admission (0/5) 4.89E-03allowance (16/1) 7.41E-03 private sector (0/5) 4.89E-03capital (11/0) 9.46E-03 pH (4/10) 5.36E-03confidence (11/0) 9.46E-03 export (29/29) 1.06E-02national income (11/0) 9.46E-03 dioxide (0/4) 1.42E-02<unclear> (19/2) 9.62E-03 fraud (0/4) 1.42E-02volume (52/14) 1.33E-02 MV (0/4) 1.42E-02radiation (10/0) 1.45E-02 National Certificate (0/4) 1.42E-02pay (27/5) 1.53E-02 organisation (0/4) 1.42E-02staff (14/1) 1.55E-02 status (0/4) 1.42E-02child benefit (9/0) 2.21E-02 wage (51/43) 1.54E-02complexity (9/0) 2.21E-02 standard (12/15) 2.00E-02liquidity (9/0) 2.21E-02 creatinine (1/5) 2.09E-02cancer (13/1) 2.23E-02 government (1/5) 2.09E-02competition (13/1) 2.23E-02 illegitimacy (4/8) 2.33E-02benefit (16/2) 2.59E-02 sale (66/51) 2.53E-02percentage (16/2) 2.59E-02 income (69/52) 3.20E-02problem (12/1) 3.19E-02 index (15/16) 3.72E-02discharge (8/0) 3.37E-02 audience (0/3) 4.11E-02load (8/0) 3.37E-02 capitalisation (0/3) 4.11E-02

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.25 (1526-1742)


Table 5. (continued)

proliferation (8/0) 3.37E-02 NAIRU (0/3) 4.11E-02release (8/0) 3.37E-02 PYY (0/3) 4.11E-02space (8/0) 3.37E-02 service sector (0/3) 4.11E-02length (11/1) 4.54E-02 education (4/7) 4.65E-02borrowing (14/2) 4.87E-02 stock (4/7) 4.65E-02

mately based on individual human actions, are emergent of these actions andare outside of the influence of individuals or even large groups of people (theyare so-called ‘invisible hand’ processes, cf. Smith 1961 [1776]; for example,prices and interest rates are determined by supply and demand).

A second interesting point is the asymmetrical distribution of the wordsspeed, frequency, number, density, height, size, capacity, volume, complexity, liq-uidity, percentage, load, space and length, which may be said to have a commonsemantic feature along the line of ‘measurability’ or ‘quantifiability’, and arethus (quasi-)scientific terms. All of them are distinctive for the literal pattern;there are almost no such nouns among the collexemes for the metaphoricalpatterns, an exception being temperature, and possibly dioxide, creatinine, andindex. The preference of such nouns for the literal pattern could be due toa reluctance concerning the use of metaphor in scientific contexts, and thuspoints to additional influencing factors beyond the cognitive and the stylistichypothesis; these are outside the scope of this study.

In sum, while there is reason to assume that some of the differences be-tween increase in NP and rise/growth in NP can be accounted for by the cogni-tive hypothesis, such an interpretation is less immediately obvious than in thecase of the patterns discussed in the preceding sections. Part of the reason maybe that we have collapsed the values for two different metaphorical patternsthat may be quite different from each other. A detailed separate comparisonof each of the metaphorical patterns with the literal pattern may shed furtherlight on this issue, but let us instead turn to the main focus of this section, thedirect comparison of the two metaphorical patterns with each other. Table 6alists the distinctive noun collexemes for the two patterns.

A clear semantic difference emerges between the two metaphorical pat-terns. The top of the collexeme list for growth in NP is dominated by the ‘com-plex abstract systems’ from Kövecses’ characterization of the mapping involved(e.g. economy, market, sector, business, etc.), while the list for rise in NP is dom-inated by nouns defined relative to amounts of something (e.g. price, (interest)rate, cost, value, wage, charge, profit, etc.).

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.26 (1742-1742)


Table 6a. Distinctive collexemes of {growth/rise} in NP (BNC-1)

rise in NP (n = 1,989) growth in NP (n = 1,258)Collexeme Distinctiveness Collexeme Distinctivenessprice (175/2) 6.66E-36 economy (0/34) 7.56E-15interest rate (76/0) 3.78E-17 GDP (1/33) 4.25E-13rate (88/2) 5.69E-17 market (3/31) 2.05E-10unemployment (94/5) 3.20E-15 size (1/21) 2.83E-08temperature (39/0) 4.32E-09 sector (2/22) 8.62E-08cost (44/3) 3.90E-07 employment (12/37) 1.50E-07level (35/1) 4.65E-07 business (2/21) 2.07E-07pressure (21/0) 3.25E-05 power (0/14) 1.64E-06inflation (19/0) 8.74E-05 supply (3/20) 2.40E-06value (40/6) 1.14E-04 use (5/23) 2.99E-06money supply (18/0) 1.43E-04 provision (1/15) 6.47E-06tax (22/1) 1.90E-04 productivity (0/11) 2.87E-05wage (37/6) 3.34E-04 membership (0/10) 7.45E-05charge (15/0) 6.29E-04 demand (39/51) 3.62E-04profit (48/11) 6.89E-04 trade (4/15) 4.12E-04crime (32/5) 7.17E-04 activity (4/14) 8.54E-04duty (13/0) 1.68E-03 sale (20/31) 1.09E-03standard (14/1) 6.63E-03 education (0/7) 1.30E-03pH (10/0) 7.37E-03 faith (0/7) 1.30E-03concentration (16/2) 1.07E-02 number (118/110) 1.56E-03fee (9/0) 1.21E-02 traffic (3/11) 2.85E-03income (40/12) 1.21E-02 job (0/6) 3.36E-03illegitimacy (8/0) 1.97E-02 debt (1/7) 6.87E-03premium (8/0) 1.97E-02 industry (1/7) 6.87E-03ratio (8/0) 1.97E-02 spending (18/25) 7.40E-03index (14/2) 2.30E-02 budget (5/12) 7.76E-03living standard (7/0) 3.22E-02 government (0/5) 8.69E-03rent (7/0) 3.22E-02 knowledge (0/5) 8.69E-03salary (10/1) 3.60E-02 labour force (0/5) 8.69E-03

private sector (0/5) 8.69E-03service (0/5) 8.69E-03GNP (2/8) 9.76E-03revenue (9/15) 1.55E-02population (22/26) 2.08E-02money (0/4) 2.25E-02National Certificate (0/4) 2.25E-02organisation (0/4) 2.25E-02volume (5/9) 4.72E-02

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.27 (1742-1760)


Table 6b. Distinctive collexemes of ADJ {rise/growth} in NP (BNC-1)

ADJ rise in (n = 1,989) ADJ growth in (n = 1,258)Collexeme Distinctiveness Collexeme Distinctivenesssharp (74/2) 4.84E-14 real (0/16) 2.43E-07steep (25/0) 4.50E-06 rapid (38/50) 3.78E-04sudden (14/1) 6.63E-03 considerable (4/15) 4.12E-04general (15/2) 1.57E-02 enormous (4/15) 4.12E-04<unmodified> (1271/757) 1.80E-02 slow (4/15) 4.12E-04immediate (7/0) 3.22E-02 strong (4/15) 4.12E-04multiplied (7/0) 3.22E-02 steady (20/25) 1.57E-02late (6/0) 5.27E-02 unprecedented (3/8) 2.37E-02meteoric (6/0) 5.27E-02 annual (7/12) 2.68E-02unexpected (5/0) 8.61E-02 continuous (0/3) 5.81E-02

incremental (0/3) 5.81E-02underlying (0/3) 5.81E-02further (11/14) 5.96E-02long-term (1/4) 7.76E-02predicted (1/4) 7.76E-02projected (1/4) 7.76E-02tremendous (1/4) 7.76E-02continued (4/7) 8.44E-02phenomenal (4/7) 8.44E-02sustained (4/7) 8.44E-02recent (11/13) 9.04E-02

Again, this difference is easily interpretable in terms of the cognitivehypothesis. Generally, the very systematicity of the distinction is evidencefor influencing factors beyond stylistic ones. More specifically, the differ-ence bears out the first expectation outlined above: complex systems can beassumed to be more difficult to understand in their entirety than the in-ternally non-differentiated amount nouns, and thus they receive strongerconceptual support.

The expectation concerning the speed and predictability with which thereferents of the nouns typically develop is also confirmed by the distinctive-collexeme analysis. The referents of the distinctive collexemes of growth in NPare abstract entities that typically change comparatively slowly, and hence rel-atively predictably, with respect to size or quantity. The complex abstract sys-tems just mentioned are a case in point, but the same is true of the amountnouns on the list (e.g. GDP, supply, membership, debt, GNP, etc.). In contrast,the distinctive collexemes of rise in NP refer to abstract entities that changecomparatively quickly, and hence much more unpredictably, with respect to

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.28 (1760-1829)


size or quantity. Again, the collexemes mentioned above are representative inthis respect (e.g. price, rate, cost, etc.), as are nouns like temperature, pressure,concentration [of chemicals], and index.

The factor of speed/predictability is also useful in interpreting the differ-ences between the adjective collexemes of the two patterns. Table 6b lists thedistinctive adjectives.

The factor of predictability manifests itself in adjectives like long-term, pre-dicted, and projected for growth in NP vs. sudden and unexpected for increasein NP, and the factor of speed manifests itself in adjectives like slow, steady,annual, continuous, incremental, continued, and sustained for growth in NP vs.sharp, steep, and meteoric for rise in NP. Note that even the one adjective forfast change that does occur with growth in NP, namely rapid, is construed asdeveloping through time, while sharp and steep are construed as instantaneous.

In sum, the results once again support the cognitive hypothesis. This doesnot come as a surprise at this point, since in this section we have actually lookedat the sort of very general mappings, pervasive in everyday language, that theproponents of the cognitive hypothesis actually cite in support of their claims.Since we already saw the principle of conceptual support at work with less sys-tematic and in some cases even register-specific mappings, it was of course en-tirely to be expected that the same should be true for fully systematic, registerindependent mappings. In addition, the analysis has provided some insightinto the motivation for the existence of alternative metaphors for the samegeneral target concept.

. The pattern understanding/grasp of NP

The patterns discussed in the preceding four subsections have all dealt withhighly general target domains (space, time, and quantity). In this final sub-section, I will extend the methodology to a slightly more specific target domainspecifically related to humans: understanding. The pattern I will investigateis shown in (11); the examples in (12a, b) show that the alternatives can berough paraphrases in some contexts:

(11) {understanding/grasp} of NP

(12) a. ... the kind of causal factors that will provide us with an understandingof the social world.

b. ... the group of disciplines that aim to provide us with a grasp of thesocial world.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.29 (1829-2023)


The metaphorical use of grasp in (12b) is motivated by the metaphors ideasare objects (cf. Lakoff & Johnson 1980:10f.) and understanding is seiz-ing idea objects firmly (cf. Jäkel 1995:202), which is a special case of themore general mapping mental activity as physical manipulation (cf. Jäkel1995:198). Further instances of this mapping are expressions like We have notyet come to grips with it, It’s not that easy to seize upon an idea, and, of course,verbal uses of grasp, as in The concepts were difficult to grasp (examples fromJäkel 1995:202). As far as the motivation for this mapping is concerned, Jäkel(1995:220) first notes that the human hand is “the most prominent organ inour intentional interaction with the physical environment”, a fact which makesit a likely source domain from the point of view of the cognitive hypothesis. Hethen claims that

there is an experiential correlation . . . between mental activity and manipu-lation. At both levels, there is an intentional gathering of knowledge aboutthe world: through tactile contact with objects as well as through learningand realization . . . abstract thinking is an internalized version of sensorimotoractivity. (Jäkel 1995: 221)

Note in this context that the metaphors postulated by Jäkel are sub-cases ofa much more general mapping, the mind is the body (cf. Johnson 1987,passim, Lakoff 1987:437–439). From this perspective, the expectations of thecognitive hypothesis are, again, that the literal understanding of NP should beassociated with concepts that are more easily understood than those associatedwith the metaphorical grasp of NP. Table 7 lists the distinctive collexemes of thetwo expressions.

As before, there are several systematic differences between the lists of dis-tinctive collexemes for the two expressions. To begin with, there is a differencewith respect to the type of knowledge referred to by the distinctive collexemes:nouns referring to procedural knowledge (‘knowing how’) occur only withthe metaphorical expression – language, English, French, idiom and grammarare unambiguous examples, but other collexemes of the metaphorical alterna-tive also suggest this type of knowledge, in that they refer to knowledge sys-tems which are applied in investigating and explaining phenomena (field, sub-ject, principle, technique, speech, tactics, strategy, philosophy, psychology, style,and rule). In contrast, the collexemes of understanding of NP typically refer tothe phenomena themselves (process, nature, need, way, and the non-significantrelationship, society, mechanism, etc.), i.e. to declarative knowledge (‘know-ing what’). Another semantic class of nouns that occurs exclusively with the

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.30 (2023-2023)


Table 7. Distinctive collexemes of {understanding/grasp} of NP (BNC-1)

understanding of NP (n = 4,167) grasp of NP (n = 356)Collexeme Distinctiveness Collexeme Distinctivenesshow (165/3) 5.32E-04 detail (2/7) 5.55E-07process (87/0) 7.44E-04 language (29/15) 5.72E-07nature (96/2) 1.36E-02 reality (23/13) 1.56E-06

English (7/7) 3.73E-05essentials (1/4) 1.77E-04French (0/3) 4.84E-04field (2/4) 4.98E-04subject (22/8) 1.70E-03principle (29/9) 2.19E-03technique (8/5) 2.23E-03concept (48/11) 5.52E-03architecture (0/2) 6.18E-03speech (0/2) 6.18E-03tactics (0/2) 6.18E-03feature (3/3) 8.08E-03issue (47/10) 1.23E-02direction (1/2) 1.76E-02idiom (1/2) 1.76E-02strategy (1/2) 1.76E-02fact (11/4) 2.57E-02grammar (6/3) 2.84E-02mode (2/2) 3.33E-02philosophy (2/2) 3.33E-02psychology (2/2) 3.33E-02style (2/2) 3.33E-02theology (2/2) 3.33E-02form (7/3) 3.83E-02rule (7/3) 3.83E-02complexity (8/3) 4.96E-02

0.05 < p < 0.1 0.05 < p < 0.1need (57/1), science (31/0), way (50/1) kind (4/2), abstraction (0/1), advantage

(0/1), affair (0/1), appeal (0/1), basics (0/1),bibliography (0/1), communicating (0/1),conditional (0/1), counterpoint (0/1), depen-dence (0/1), domain (0/1), electronic (0/1),eugenics (0/1), evil (0/1), exigency (0/1),film (0/1), Hebrew (0/1), holism (0/1), im-mensity (0/1), melody (0/1), melos (0/1),metaphor (0/1), mood (0/1), myself (0/1)

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.31 (2023-2048)


Table 7. (continued)

Portuguese (0/1), real (0/1), realpolitik (0/1),sequence (0/1), set (0/1), software (0/1),strength (0/1), theatre (0/1), tongue (0/1),trivia (0/1), turn-taking (0/1), virtue (0/1),whole (0/1), matter (11/3), material (5/2),music (5/2)

0.1 < p < 0.25 0.1 < p < 0.25relationship (43/1), society (25/0), mecha-nism (24/0), role (40/1), religion (22/0),term (22/0), faith (21/0), why (21/0), what(199/13), aspect (18/0), change (18/0), ac-tion (17/0), disease (17/0), experience (17/0),value (17/0)

difference (12/3), system (30/5), idea (14/3),analysis (1/1), audience (1/1), banking (1/1),brief (1/1), capitalism (1/1), challenge (1/1),chronology (1/1), conversation (1/1), craft(1/1), criteria (1/1), death (1/1), decimal(1/1), ergonomics (1/1), generation (1/1),grace (1/1), intricacy (1/1), medicine (1/1),practicality (1/1), proportion (1/1), regular-ity (1/1), report (1/1), teaching (1/1), those(1/1), vocabulary (1/1), content (7/2), topic(8/2), meaning (16/3), theory (16/3), knowl-edge (9/2), activity (2/1), causality (2/1),convention (2/1), democracy (2/1), geogra-phy (2/1), information (2/1), link (2/1), pic-ture (2/1), scripture (2/1), terminology (2/1),structure (27/4)

metaphorical pattern consists of nouns referring to component parts or indi-vidual aspects of a greater whole (detail, essentials, feature, form, complexity,and the marginally significant basics, matter, and material). In contrast, theconcepts referred to by the literal pattern refer to the greater wholes themselves(notably, nature).

The latter difference can reasonably straightforwardly be interpreted interms of the cognitive hypothesis: since the concepts referred to by the nounsjust mentioned construe the world at a high level of resolution, they are pre-sumably cognitively more complex, and hence in greater need of conceptualsupport, than the ‘greater wholes’ referred to by the distinctive collexemes ofthe literal expression (cf. Assumption 3). The difference relating to proceduraland declarative knowledge is less straightforward. Although it does not soundimplausible to say that procedural knowledge is more complex than declara-tive knowledge (it is difficult to make explicit, it takes longer to acquire, etc.),one could also make the opposite case (declarative knowledge is less intuitive,

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.32 (2048-2094)


etc.). In other words, the issue of complexity is less clear here than in the caseof the other expressions. However, the primary motivation for the differencein question is probably a different one anyway. Note that it makes sense thatprocedural knowledge, which is applied knowledge of how to do something,is understood in terms of manual manipulation; it is this kind of knowledgefor which Jäkel’s account of ‘internalized sensorimotor activity’ is most plausi-ble. Declarative knowledge, in contrast, is knowledge about things, and is notusually acquired through physical activity. Thus, as in the case of the differ-ence between growth and rise, the function of metaphor here is not simply toprovide conceptual support, but to enable us to understand something in aparticular way.

. Conclusions

The application of distinctive-collexeme analysis to near-synonymous pairs ofliteral and metaphorical expressions has yielded clear evidence for the cogni-tive hypothesis. First, for all of the pairs investigated, the distinctive collex-emes showed systematic differences; this in itself is already evidence againstthe idea that metaphor is merely ornamental in nature. Second, the differ-ences are clearly semantic in nature, and in the overwhelming number of casesthey can plausibly be related to the factor of complexity. In all these cases, themetaphorical pattern (or, where both patterns are metaphorical, the patternbased on the more specific metaphor) is associated with nouns referring tomore complex concepts; this can reasonably be interpreted as evidence for theidea that metaphors provide conceptual support. In addition, two other factorswere tentatively identified: metaphorical coherence, more precisely, a possibleaversion towards mixing metaphors of different specificity; and the avoidanceof metaphor in particular contexts. Neither of these factors was investigated indetail however, and future research must show the existence and extent of theirinfluence. Finally, the comparison of two near-synonymous metaphorical pat-terns yielded evidence that points to the idea that metaphor not only helps usto understand abstractions, but that it can also help us to understand them indifferent ways.

Although the results of the case studies presented here all point towardsthe cognitive hypothesis, they must be considered preliminary in two respects.First, the case studies deal with only a handful of literal/non-literal pairs of ex-pressions, all of which instantiate a single grammatical context. This restriction

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.33 (2094-2135)


was necessary in the context of this study to make sure that the methodologyproposed would yield interpretable results. Since this has turned out to be thecase, it is now necessary to extend distinctive collexeme analysis to many morepairs of words belonging to a wide range of word classes and occurring in awide range of grammatical contexts; only a broad application of the methodcan show how useful it will ultimately turn out to be. Second, since this ex-ploratory study has yielded such promising results, it would be desirable to for-mulate more specific predictions based on these, so that more rigorous researchprocedures can be constructed. In this context, a less intuitive operationaliza-tion of the notion ‘complexity’ is required (for example, nouns (or other partsof speech) constituting potential collexemes could be coded by linguisticallynaïve informants for the semantic dimensions assumed here to correlate withconceptual complexity (size, delineability, etc)).

In any case, this study has shown that quantitative corpus methods in gen-eral, and distinctive-collexeme analysis in particular can be fruitfully appliedto theoretical issues that have not previously been considered amenable tothis type of research. Hopefully, this paper will inspire more corpus-based re-search in the area of figurative language in particular and more applications ofquantitative corpus-linguistic methods to theoretical issues in general.

Notes

* I would like to thank Kerstin Fischer, Stefan Th. Gries, Stefanie Wulff and two anony-mous reviewers for discussion and insightful comments on earlier drafts of this paper. Anyremaining errors and inconsistencies are, naturally, mine alone.

. The phenomenon of metaphor has been studied literally for thousands of years in disci-plines as diverse as theology, philosophy, rhetoric, literary criticism, political science, psy-chology, and linguistics. Thus, there are of course many more than just two hypothesesconcerning the aspects of metaphor discussed here and many other aspects. The two hy-potheses discussed here are the ones currently most widely discussed in linguistics; this doesnot mean, of course, that other hypotheses are not worthy of discussion or amenable tocorpus-linguistic investigation.

. See e.g. Deignan (1995, 1999); Partington (1998, 2003); Semino et al. (1997); Semino &Steen (2001); Stefanowitsch (2004, to appear a). Cf. also the contributions in Stefanowitschand Gries, to appear a and the overview in Stefanowitsch, to appear b.

. Note that the term stylistic is used in a rather restricted sense here; clearly, there are otherdimensions of language that are sometimes discussed under the label style (and that othersmay discuss under labels such as genre, register, communicative style, etc.), and my use of the

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.34 (2135-2199)


term is not meant to imply that these dimensions are not potentially relevant to studyingthe functions of metaphor.

. Unless otherwise indicated, all linguistic examples in this paper (numbered examplesas well as those occurring in the running text) are taken from the British National Corpus(BNC 1.0); the only modifications that have sometimes been made for expository reasons aredeletions (for example, of relative clauses etc.). Since the data come from a British corpus,the original British English spelling has been retained in all examples and word lists.

. Strictly speaking, in our earlier studies we limited the methodology to the investigationof the association of words to the constructions they appear in; we defined a constructionas a non-compositional form-meaning pair at any level of grammatical structure, in accor-dance with the theory of Construction Grammar (cf. Fillmore 1988; Goldberg 1995; Kay &Fillmore 1999). It is not clear that the patterns investigated here meet the criterion of non-compositionality, but since they do present conventionalized forms for expressing a partic-ular meaning, our method applies to these patterns just as well (note in this context, thatother construction-based theories do not see non-compositionality as a crucial requirementfor construction-hood, e.g. Langacker (1987)).

. Clearly, the strategy of lemmatizing the collexemes is not uncontroversial, given thatcorpus-based studies have repeatedly shown that different word forms may differ withrespect to their collocates. An explorative study of grammatical patterns associated withparticular words, for example, would clearly have to take into account different wordforms. In the present case, and in collostructional analysis in general, however, the pat-terns/constructions under investigation are decided upon a priori, and the issue under in-vestigation is, put simply, what semantic fields are associated with these patterns. In thecontext of semantic fields, it does not seem to make sense to focus on word forms. Prelimi-nary support for this assumption comes from a wordform-based collostructional analysis ofcausative constructions by Gilquin (2004), who does not find systematic semantic differencesbetween the behaviors of different word forms.

. Since the format in which the p-values are shown here may not be familiar to all lin-guists, a brief clarification may be in order: in the so-called ‘exponential notation’ the fig-ure after the code E- indicates by how many decimal places the figure before the letter E-must be transposed, i.e. 6.24E-08 must be read as 0.0000000624, 1.00E-08 must be read as0.00000001, etc. The reason for using this format will become obvious below: in some cases,the p-values are so small that they show up at the 15th, 30th, or even 36th decimal place.

. Since this procedure determines the degree of association between a word and a particu-lar pattern in comparison to another pattern, the p-value can actually be read in two ways: ifa word is attracted to one member of a pair of patterns (i.e., occurs more frequently than ex-pected with it) at a particular level of significance, then it follows logically that it is repelledby the other member (i.e., occurs less frequently than expected with it) at the same levelof significance. For example, we may read the Fisher-Yates exact p-value of p = 6.24E-08 asthe degree of attraction between the noun city and the pattern in the heart of NP or as thedegree of repulsion between the noun city and the pattern in the center of NP (in both casesin relation to the other pattern).

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.35 (2199-2264)


. Alternatively, the analysis could be limited to the pattern(s) shared by the two synonyms(in the case of go up vs. increase, the intransitive use, in the case of grasp vs. understand,the contexts [V + OBJcontent] and [V + that-clausecontent]). This procedure is not entirelywithout problems either, however, as it may be precisely the data thus discarded that couldilluminate the semantic differences we are after.

. I formulated these assumptions after deciding on the patterns to be investigated, butbefore the actual data coding and analysis. This strategy was meant to ensure that the criteriareferred to in the assumptions are applicable to the kinds of referents likely to be found in theNP slot of the relevant patterns (i.e., that no criteria were included which would not makesense in the contexts in question); at the same time, it was meant to prevent a post-hocformulation of criteria that was influenced by what I might have expected to find. Clearly,this strategy is not completely fail-safe, but it seemed to me to have an adequate cost-benefitratio in the context of a corpus-linguistic study (as opposed to a psycholinguistic study,which would clearly have to give this matter more attention).

. The seemingly marginal status of the centre-as-heart mapping in the cognitive lin-guistics literature is confirmed by the fact that Niemeyer (2000) does not even mention it inher survey of metaphorical expressions containing heart. It should be pointed out, however,that this marginal status in the literature does not reflect a low frequency of such expres-sions: they are frequent both in English and cross-linguistically, and they play an importantpart in the grammaticization of, for example, prepositions. What is at issue here is simplytheir very limited productivity in comparison to other metaphorical mappings.

. A relatively recent example is found in Dylan Thomas’ poem Do not go gentle into thatgood night, with its famous first verse Do not go gentle into that good night,/Old age shouldburn and rave at close of day;/Rage, rage against the dying of the light.

. These nouns are analyzed here as metonymically standing for time spans because a fullyexplicit paraphrase of the corresponding examples needs to make reference to a time span:for example, at the dawn of civilization would have to be paraphrased as at the beginningof the time span during which civilization exists; at the dawn of history would have to beparaphrased as at the beginning of the time span during which the events that constitute ourhistory occur; etc. The metonymy here is a state/event for the time span taken up by thatstate/event. Examples like at the dawn of mankind would have to be paraphrased as at thebeginning of the time span during which mankind exists; the metonymy is an entity for thetime span taken up by the existence of that entity.

References

Aristotle (1959). The Art of Rhetoric. (Transl. J. H. Freese). London: Dutton.Barnbrook, G. (1996). Language and computers. A practical introduction to the computer

analysis of language. Edinburgh: Edinburg University Press.Biebuyck, B., Dirven, R., & Ries, J. (Eds.). (1998). Faith and Fiction. Frankfurt am Main:

Peter Lang.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.36 (2264-2365)


Cameron, L. J. (2003). Metaphor in educational discourse. London and New York:Continuum.

Church, K. W., & Hanks, P. (1990). Word association norms, mutual information, andlexicography. Computational Linguistics, 16 (1), 22–9.

Church, K. W., Gale, W., Hanks, P., & Hindle, D. (1991). Using statistics in lexical analysis.In U. Zernik (Ed.), Lexical acquisition: Exploiting on-line resources to build up a lexicon(pp. 115–64). Hillsdale, NJ: Lawrence Erlbaum.

Collins Cobuild English Dictionary (1995). London: Harper Collins.Deignan, A. (1995). Cobuild Guides to English 7: Metaphor. London: Harper Collins.Deignan, A. (1999). Corpus-based research into metaphor. In L. J. Cameron & G. Low

(Eds.), Researching and applying metaphor (pp. 177–199). Cambridge: CambridgeUniversity Press.

Dunning, T. (1993). Accurate methods for the statistics of surprise and coincidence.Computational Linguistics, 19 (1), 61–74.

Fillmore, C. J. (1988). The mechanisms of ‘Construction Grammar’. In S. Axmaker, A.Jaisser, & H. Singmaster (Eds.), Proceedings of the Fourteenth Annual Meeting of theBerkeley Linguistics Society (pp. 35–55). University of California, Berkeley: BerkeleyLinguistics Society.

Garner W. R. (1974). The processing of information and structure. Potomac MD: Erlbaum.Gibbs, R. W. (1994). The poetics of mind. Cambridge: Cambridge University Press.Gibbs, R. W., & Steen, G. J. (Eds.). (1999). Metaphor in cognitive linguistics. Amsterdam and

Philadelphia: John Benjamins.Gilquin, G. (2004). The V slot in causative constructions: A collostructional analysis. Paper

presented at the Third International Conference on Construction Grammar, July 7–10,2004. University of Provence, France.

Goldberg, A. E. (1995). Constructions. A construction grammar approach to argumentstructure. Chicago: University of Chicago Press.

Grady, J. (1999). A typology of motivation for conceptual metaphor: Correlation vs.resemblance. In R. W. Gibbs & G. J. Steen (Eds.), Metaphor in cognitive linguistics (pp.79–100). Amsterdam and Philadelphia: John Benjamins.

Grady, J. E. (1997). Theories are buildings revisited. Cognitive Linguistics, 8 (4), 267–290.Gries, S. T. (2003). Testing the sub-test: A collocational-overlap analysis of English -ic and

-ical adjectives. International Journal of Corpus Linguistics, 8 (1), 31–61.Gries, S. T., & Stefanowitsch, A. (2004a). Extending collostructional analysis: A corpus-

based perspective on ‘alternations’. International Journal of Corpus Linguistics, 9 (1),97–129.

Gries, S. T., & Stefanowitsch, A. (2004b). Co-varying collexemes in the into-causative. InM. Achard & S. Kemmer (Eds.), Language, culture, and mind (pp. 225–236). Stanford:CSLI.

Hopper, P., & Traugott, E. C. (2003). Grammaticalization (2nd edition.). Cambridge:Cambridge University Press.

Jäkel, O. (1995). The metaphorical conception of mind: ‘Mental activity is manipulation’. InJ. R. Taylor & R. E. MacLaury (Eds.), Language and the cognitive construal of the world(pp. 197–229). Berlin/New York: Mouton de Gruyter.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.37 (2365-2465)


Jäkel, O. (1997). Metaphern in abstrakten Diskursdomänen. Eine kognitiv-linguistischeUntersuchung anhand der Bereiche Geistestätigkeit, Wirtschaft und Wissenschaft.Frankfurt a.M.: Peter Lang.

Jäkel, O. (1999). Kant, Blumenberg, Weinrich: Some forgotten contributions to the cognitivetheory of metaphor. In R. W. Gibbs & G. J. Steen (Eds.), Metaphor in cognitive linguistics(pp. 9–27). Amsterdam and Philadelphia: Benjamins.

Johnson, M. (1987). The body in the mind. The bodily basis of meaning, imagination, andreason. Chicago and London: The University of Chicago Press.

Kay, P., & Fillmore, C. J. (1999). Grammatical constructions and linguistic generalizations:The What’s X Doing Y construction. Language, 75 (1), 1–33.

Kennedy, G. (1991). Between and through: The company they keep and the functions theyserve. In K. Aijmer & B. Altenberg (Eds.), English corpus linguistics (pp. 95–110).London: Longman.

Kövecses, Z. (1986). Metaphors of anger, pride, and love. A lexical approach to the structure ofconcepts. Amsterdam and Philadelphia: John Benjamins.

Kövecses, Z. (2002). Metaphor. A practical introduction. Oxford: Oxford University Press.Lakoff, G. (1987). Women, fire, and dangerous things. What categories reveal about the mind.

Chicago and London: The University of Chicago Press.Lakoff, G. (1993). The contemporary theory of metaphor. In A. Ortony (Ed.), Metaphor and

thought. Second edition (pp. 202–251). Cambridge: Cambridge University Press.Lakoff, G. (1996). Moral politics. What conservatives know that liberals don’t. Chicago and

London: University of Chicago Press.Lakoff, G., & Johnson, M. (1980). Metaphors we live by. Chicago and London: The University

of Chicago Press.Langacker, R. W. (1987). Foundations of cognitive grammar. Vol. 1. Theoretical prerequisites.

Stanford: Stanford University Press.Martin, J. (to appear). A corpus-based analysis of context effects on metaphor

comprehension. In A. Stefanowitsch & S.T. Gries (Eds.), Corpus-based approaches tometaphor. Berlin and New York: Mouton de Gruyter.

Niemeier, S. (2000). Straight from the heart – metonymic and metaphorical explorations.In A. Barcelona (Ed.), Metaphor and metonymy at the crossroads. A cognitive perspective(pp. 195–213). Berlin and New York: Mouton de Gruyter.

Partington, A. (1998). Patterns and meanings. Amsterdam and Philadelphia: Benjamins.Partington, A. (2003). The linguistics of political argument. The spin-doctor and the wolf-pack

at the White House. London: Routledge.Pedersen, T. (1996). Fishing for exactness. Proceedings of the SCSUG 96 in Austin, TX, 188–

200.Pedersen, T., Banerjee, S., & Purandare, A. (2003). Ngram statistics package 0.53.

http://www.d.umn.edu/∼tpederse/code.html (last access: 20 January 2003).Radden, G. (1997). Time is space. In B. Smieja & M. Tasch (Eds.), Human contact through

language and linguistics (pp. 147–166). Frankfurt am Main: Peter Lang.Richards, I. A. (1936). The philosophy of rhetoric. London: Oxford University Press.

JB[v.20020404] Prn:10/05/2005; 9:21 F: IJC10202.tex / p.38 (2465-2555)


Semino, E., & Steen, G. J. (2001). A method for annotating metaphors in corpora. InJ. Barnden, M. Lee, & K. Markert (Eds.), Proceedings of the workshop on corpus-basedand processing apporaches to figurative language (pp. 59–66). Lancaster: UniversityCentre for Computer Corpus Research on Language.

Semino, E., Short, M., & Culpeper, J. (1997). Using a computer corpus to test a model ofspeech and thought presentation. Poetics, 25 (1), 17–43.

Sinclair, J. (1991). Corpus, concordance, collocation. Oxford: Oxford University Press.Smith, A. (1776). The wealth of nations. Repr. 1961. London: Methuen.Stefanowitsch, A., & Gries, S. T. (2003). Collostructions: Investigating the interaction of

words and constructions. International Journal of Corpus Linguistics, 8 (2), 209–243.Stefanowitsch, A., & Gries, S. T. (2005). Covarying Collexemes. Corpus Linguistics and

Linguistic Theory 1 (1), 1–46.Stefanowitsch, A., & Gries, S. T. (Eds). (to appear). Corpus-based approaches to metaphor.

Berlin and New York: Mouton de Gruyter.Stefanowitsch, A. (2004). Happiness in English and German. In M. Achard & S. Kemmer

(Eds.), Language, Mind, and Culture (pp. 137–149). Stanford: CSLI.Stefanowitsch, A. (to appear a). Words and their metaphors: A corpus-based approach. In

A. Stefanowitsch & S. T. Gries (Eds), Corpus-based approaches to metaphor. Berlin andNew York: Mouton de Gruyter.

Stefanowitsch, A. (to appear b). Corpus-based approaches to metaphor and metonymy. InA. Stefanowitsch & S. T. Gries (Eds.), Corpus-based approaches to metaphor. Berlin andNew York: Mouton de Gruyter.

Sweetser, E. (1990). From etymology to pragmatics. Metaphorical and cultural aspects ofsemantic structure. Cambridge: Cambridge University Press.

The New Encyclopædia Britannica. Micropædia, Vol. 8. (1989). 15th Edition. Chicago: TheUniversity of Chicago Press.

Wertheimer, M. (1938). Laws of Organization in Perceptual Forms. In W. Ellis (Ed.), A sourcebook of Gestalt psychology (pp. 71–88). London: Routledge & Kegan Paul.

Winter, S. (1989). Transcendental nonsense, metaphoric reasoning and the cognitive stakesfor law, University of Pennsylvania Law Review, 137, 1105–1237.

Author’s address

Prof.dr. Anatol StefanowitschGruenebergstrasse 40D-22763 HAMBURGGermany

The function of metaphor: Developing a corpus-based perspective

Documents

Transcript of The function of metaphor: Developing a corpus-based perspective