Lexico-grammatical alignment in metaphor construal

1.1 Theory

The overall approach is built on one strand of Conceptual Metaphor Theory, which assumes the metaphors that occur in natural language are based on systematic bindings between set source and target domains. These links arise, at an unconscious level, from real world co-occurrences between realms of experience and the abstract concepts associated with them (Gallese and Lakoff 2005; Lakoff and Johnson 1980, Ch. 12; Lakoff 1993). For example, the conflation of physical, translational motion and the ‘passage’ of time results in a set of metaphors used to understand the abstract notion of time (Núñez and Sweetser 2006; Moore 2006, 2014), demonstrated in examples like Christmas is around the corner. In another illustration, the concurrence of quantity increase with vertical growth (as when water is poured into a glass) gives rise to the primary conceptual metaphor quantity is verticality (Lakoff and Johnson 1980; Grady 1997, 1999) and is manifested in language in examples like rent prices have risen again. Primary metaphors like states are locations and change is motion combine with others to structure more complex metaphors and build detailed role and inference information. For instance, in the common mapping, life is a journey, consistent experience with travel structures common elements associated with the journey frame: traveler, vehicle, origin, destination, and path (Fillmore 1982, 1985; Fillmore and Baker 2010). These frame elements systematically map over to construct the concept of life and life stages: birth as origin, death as destination, choices as paths, impediments as obstacles. The notion that metaphor is primarily conceptual is not uncontroversial (Gibbs 2017), and although the conceptual nature of metaphorical mappings has been supported experimentally in the psychological literature (cf. Boroditsky 2000; Casasanto and Boroditsky 2008; Gibbs 2017, Ch. 5; Thibodeau and Boroditsky 2011, 2013, among others), many in the linguistic community dispute metaphor as a structural and grammatical primitive (Pinker 2006, 2007) or take a hedged view contending that some metaphors are conceptual and unconscious, while others conscious and deliberate (Steen 2007, 2011, 2015).

Conceptual metaphor adversaries, led by vocal critics like Steven Pinker (2007), argue against the cognitive nature of metaphor mapping. “People don’t dig down to a conceptual metaphor every time they understand a conventional one” and “Though metaphors are omnipresent in language, many of them are effectively dead in the minds’ of today’s speakers and the living ones could never be learned, understood, or used as reasoning tools unless they were built out of abstract concepts that capture the similarities and differences between the symbol and symbolized” (p. 248, 276 respectively). Through the lens of lexicalization, old debates about conscious metaphor activation can be abandoned and replaced with systematic measures of how lexical items come together to produce metaphorical phraseology, whether conventional (i.e. frequent in the speech community) or creative and novel (infrequent expressions that fit in line with established conceptual metaphors). Taking the view that conceptual metaphors serve as cognitive input used in lexicalization processes (cf. Talmy 2000a, 2000b), this study introduces new methodology in the investigation of lexicalization. By preselecting established metaphors and then searching for lexical patterns, the methodology inverts the common approach to metaphor study, in which existing language is used to find conceptual mappings. This methodological inversion allows for a systematic examination of several important research questions pertaining to the interface between metaphor, grammar, and the lexicon.

1.2 Structure

The paper is structured as follows. After providing theoretical background (§2) and outlining the methodology (§3), the first study (§4) examines the syntactic realization of source and target pairings. Acknowledging metaphorical mappings are more constrained by autonomous target domains, Sullivan (2013) hypothesizes a preference for target-domain language to populate semantically more autonomous syntactic positions in modifier-noun constructions, predicting that source and target triggers should align with predicate-specifying and argument-specifying syntactic positions, in that order. Given a lexeme’s origin – source or target – we should be able to predict syntactic alignment. To test this hypothesis, seed words like financial and storm (identified as frequent metaphor triggers in The Economist magazine [Lederer 2016]) are used to probe fixed and semi-fixed economic phraseology. Using the Corpus of Contemporary American English (Davies, 2008–present), the large-scale examination includes 12 frequent metaphorical target triggers (bank, invest, market, firm, debt, finance, rate, price, capital, growth, economy, money) combined with 84 source triggers (including lemmas like sail, storm, channel, conditions, freeze, cloudy, turbulent, ailing, and health) to produce 2016 ordered pairings, i.e. investment freeze and turbulent market. Each lexical combination is evaluated for its rate of occurrence in five syntactic constructions: NP-of-NP (captain of finance; wave of capital), A-NP (economic headwinds; ailing economy), N-N compound (debt bubble; bubble market) V-NP (steer the economy), and X is Y (the economy is a sinking ship).

The notion that metaphorical mappings feed lexicalization patterns is by no means an endorsement for a ‘slot and filler’ (Sinclair 1991: 109) approach to language representation. Instead, through a series of measurements, the second series of studies probe the ‘single choice’ or ‘idiom principle’ (Sinclair 1996, 1998, 2004). Under this ‘co-selection’ view (Sinclair 1996, 1998, 2004), words or words and grammatical constructions are often chosen simultaneously to express a particular meaning (Herbst et al. 2011: 27). The co-selection of source and target triggers is explored through three measures of lexicalization (§5). In the first measure, metaphorical phrases are catalogued along a scale of conventionality and entrenchment (Gibbs 1984) (§5.1). Frequency statistics are used as a proxy for conventionalization (Cheng 2012). On one end, highly frequent collocations of set source and target triggers occur as common fixed or semi-fixed phrases; on the other end, more creative and novel pairings will be less frequent in corpus data, and, thus, not considered conventionalized, metaphorical jargon. Economic phraseology can be further dissected by looking at patterns of source and target trigger behavior. A second line of investigation (§5.2), focuses on distributional coverage and ‘metaphorical combinability’ (Hausmann 2007): Which source triggers are the most permissive in their ability to co-occur with target triggers and vice-versa? The factors that might contribute to combinability scores include a lemma’s rate of lexical variation, base rate frequency in economic discourse, and keyness score – relative frequency in economic discourse compared to overall use in English (Cheng 2012; Ahmad 2005).

In a final section of analysis (§5.3), the study turns to asymmetric ordering patterns of specific source and target pairings to examine features that might contribute to token frequency differences between sequences of the same triggers when both sequences can be used metaphorically. For example, in COCA, financial health occurs much more regularly than healthy finances. Both possibilities come from the same metaphor, the economy is an ailing body; however, the source and target triggers do not fill the A-N slots at the same rate. Based on the examination of base-rate scores and keyness statistics, these asymmetries seem to be motivated by discourse context (public discussion of economic topics focuses more on macro- versus micro-economics) rather than overall word frequency differentials.

2.1 The economy as a metaphorically based cognitive model

Individual source-target mappings work in tandem to organize the cognitive models that structure complex abstract concepts (Lakoff 1987). The economy, as a metaphorical model, fits several types outlined by Steen (Steen 2011: 57): it is “implicit” in nature – reflective of cultural models; “official”, in the sense that it is instilled through education and the academy, while at the same time ‘emergent’ – economic discourse arises through social interaction, such as within the business and academic economic communities.

Economic concepts are inextricably linked to a metaphorical understanding of money. The target domain, money (currency), is structured through an understanding of the movement of liquid or water. Money circulation is the flow of capital from banks to consumers to business and back again as illustrated in examples taken from The Economist magazine (2)–(5):

The source domain liquid is instantiated and extended through rich concepts related to water including bubbles (soapy water) and wells (containers for water) along with properties important in the description of water like hot and cold (activity is heat; inactivity is cold) and numerous manners of movement like rush or trickle.

Because liquid serves as the source domain for the understanding of money, a metaphorical understanding of the economy also relies on several source domains that have to do primarily with water. These are weather events, i.e. rain, storms, clouds, bodies of water like seas, oceans, channels, rivers, and the vehicles that navigate waterways: boats and ships. The concept of climate is used to discuss public sentiment on economic issues and economic performance is understood through perceptions pertaining to weather conditions: bad weather means negative performance; good weather is positive as in (6)–(9):

Weather language comes in a wide variety of lexis including specific conditions like stormy, sunny, wind, and drought, and concepts related to weather patterns like season, climate, and forecast.

The economy is assumed, like a vehicle, to move forward at varying rates of speed, mapping to slow and fast progress. Because water-based metaphors structure monetary theory, the vehicle is most commonly a ship or boat. Policymakers direct the economy, just like captains steer ships. A poorly performing economy is likened to a sinking ship and economic or financial failure is discussed as if it were a shipwreck as in (10)–(14):

In a different metaphor with an unrelated source domain, the economy is understood through the frame of a (usually sick) human body, a type of extended personification (Shenker-Osorio 2012: 43) as illustrated in (15)–(17):

In the analysis of a complex target like the economy, it is not surprising to find a concept, like temperature, playing a role in more than one source domain. For example, body temperature is one measurement used to diagnose health and disease in a medical setting, but air temperature is also a concept integral to our understanding of weather and machines. Thus, lexemes like hot and cold can activate alternative source domains in various contexts and can be compatible with multiple source domains simultaneously. The activation of temperature differentially applies to multiple metaphors. High temperature in a body is potentially grave, a sign of illness in the metaphor economy is an (ailing) human body as illustrated in (18):

But when used in reference to the economy is a weather event, heat, up to a certain point, is good and cold is bad (as prescribed by the underlying metaphor heat is activity). These inferences are shown in (19) and (20):

As the above examples demonstrate, figurative language often draws on more than one source domain. The intersecting nature of these related source domains helps explain the prevalence of lexis compatible with more than one metaphor. Metaphorical lexis activate source domains that that overlap in their conceptual structure as illustrated in Figure 1:

Figure 1:

Overlapping source domains give rise to metaphorical lexis that build the cognitive model.

The aforementioned conceptual metaphors work together to build a complex cognitive model of the economy as a ship moving through potentially treacherous waters due to inclement weather. Ruiz de Mendoza (2017: 147) discusses this type of overlap as epiphenomenal, a case of a frame complex enriching an image schema complex. In the vehicle model, the image schema involves forward motion along a path; the vehicle frame and obstacle frame elements are accordingly filled by conceptually coherent scenarios, ship and treacherous waters respectively. In a separate model, individual economies are understood as sick hospital patients, in need of medical attention, threatening to contaminate other economies. It should be noted, however, that the lexical overlap with terms like hot and circulate, is due to shared, but independent, concepts like temperature and liquid motion across unrelated source domains: the human body can run a fever (negative); whereas a vehicle engine needs heat for energy (positive).

2.2 Background – lexicalization of source domains

The use of corpus methodology to examine the lexicalization patterns of conceptual metaphors has been approached from several angles (Charteris-Black 2004; Musolff 2006; Oster 2010; David et al. 2014; Stickles et al. 2016; Petruck 2016; Koller et al. 2008; L’Hote 2014; Demmen et al. 2015). In critical metaphor studies (Charteris-Black 2004; Musolff 2006), researchers manually mine small corpora for metaphorical language, qualitatively assessing data for common metaphorical patterns. Other semi-automated techniques involve identifying lexical units attached to specific source domains, for example, automatically searching a corpus for lexical items associated with particular semantic domains, and then manually reviewing the data for instances of specific types of conceptual metaphors (Oster 2010; Koller et al. 2008; L’Hote 2014; Demmen et al. 2015).

In some cases, the combination of a metaphoric verb and a non-metaphoric noun is what determines the metaphoric interpretation of the sentence (Petruck 2016: 135), or collocated pairs of source and target triggers in general. Hence, fully automated approaches to metaphor identification include a focus on target and source word pairs (Shutova 2015 and citations therein), with select projects directed toward cataloguing the relationship between grammar, metaphors, and the system of conceptual metaphor (David et al. 2014; Stickles et al. 2016).

2.3 Metaphor and grammar

In cognitive theories of grammar, metaphor plays an important role in governing and explaining semantic patterns that contribute to the syntactic configurations structuring English and other languages (Langacker 1999, 2009; Talmy 1978; Lakoff 1990a; Goldberg 1995; Dancygier and Sweetser 2014). Dancygier and Sweetser (2014: 127) (also see Moder 2010) show how the Caused-Motion construction (Goldberg 1995), consisting of Subject Verb Theme-NP Path, encodes a starting point, path, and endpoint of motion as in (21):

When the verb drag, which encodes explicit caused-motion, is replaced with the non-motion, usually intransitive laugh, the sentence retains a sense of motion; however, in this case, the motion is metaphorical. Joe’s ‘movement’ is from a state of depression to a state beyond depression in (22):

Because states, like depression, are metaphorically understood as locations in space, and change in state is metaphorically understood as change in location, the verb laugh can be used transitively to describe the force that results in the change in state.

communication, as a target domain, draws on a related source domain, object transfer, in this case, from speaker to listener, sender to recipient (Reddy 1979). Thus, the same syntactic construction can be used to describe events of communication as in (23), in which traditional verbs of communication fill the verb slot:

But, so too can prototypical verbs of literal motion, liquid-motion in (24):

A liquid-motion verb like leak can be used in this construction because communication is metaphorically understood as physical transfer. In fact, quite a few verbs of liquid motion function similarly: spew, spill, drip, trickle, etc., demonstrating the productivity of the metaphor. Whether lexicalized with prototypical communication verbs, as in (23), or a verb of explicit motion as in (24), the construction activates the same conduit metaphor in which communication is understood as object transfer (cf. David 2016, Ch 4, for a discussion of the interaction between constructional and lexical metaphor activation). Target domains (whether they be states of emotion or communication) that are structured by the same source domain (object-transfer) are all equally able to participate in the Caused-Motion construction.

2.4 Deep dive into discrete constructions

Particular grammatical positions prompt particular roles in metaphorical construal (Croft 1993: 355, Dancygier and Sweetser 2014: 134). This is true in a basic, literal copular construction (cf. Mikkelsen 2005), where the predicational material occurs after the copula (Jane is a dancer, not A dancer is Jane). Metaphorical copular constructions are no different, as in life is a journey. As noted by Dancygier and Sweetser (2014: 134), the reverse ordering – a journey is life – is grammatical, but does not make sense as a metaphorical construction absent context because the intended source domain in this sequence, life, is an abstract concept, and cannot be used as a grounding mechanism to understand the more concrete frame of a journey, at least not in any conventionalized usage pattern. Because source domains are, for the most part, experiential in nature (Lakoff and Johnson 1980: Ch. 12, 1999), they tend to encode concrete information which serves as the predicational material used to understand the more abstract, non-experiential target domain, hence the conventionalized reporting of conceptual metaphors as target is source.

Sullivan (2009, 2013) argues that the same conceptual dependency – target domain dependent on source domain – explains broad regularities concerning which grammatical slots are filled by source and target domain vocabulary in isolated constructions when both domains are overtly expressed. In general, Sullivan (2013: 63) contends there are strong tendencies for attributive adjectives to be source domain triggers while the nouns they modify evoke the target domain. In her example, blood-stained wealth (money gained through criminal means, not bills soaked in actual blood), the modifying adjective blood-stained elaborates the source domain of the metaphor immorality is filth (converse of morality is cleanliness, i.e. clean vs. dirty money) (Sullivan 2013: 63; Kövecses 2010). Sullivan explains the connection between modifiers and source domains through dependency patterns noted in literal language. Her explanation rests on observations of trajector-landmark dependency relations within the Cognitive Grammar framework.

In literal language, Langacker (1987, 1991) posits that in modification constructions, there is an asymmetry of conceptual dependency. The head is more semantically autonomous whereas the modifier is more dependent. He uses the contrast between big elephant vs. big molecule – in each phrase it is the noun that determines the size of the scale (Langacker 1987: 285). Langacker shows the reverse is true for predicate-argument, V-N, constructions – the verb (head) is conceptually and semantically dependent on the noun: In open the book or open your eye, comprehension of the associated motion indexed by the verb depends on specific knowledge about the opened object (we can only simulate an opening event accurately through our understanding of detailed information about the argument-book-opening involves a completely different motor program than does eye-opening). In each case of predication, whether adjective or verb, the semantically predicated information (OPEN and BIG) is dependent on the modified argument. ^[1]

However, not all modifiers are alike. At a broad level, modifiers can be divided into two types: ‘domain’ and ‘attributive’ (sometimes called ‘relational’ and ‘qualitative’ (cf. Gunkel and Zifonun 2008: 283). Domain adjectives are predominantly formed with non-Germanic suffixes, often of Latin origin, i.e. -al, -an, -ary, -ic, etc. (Gunkel and Zifonun 2008: 284). However, these same suffixes and their allomorphs (Huddleston and Pullum 2002: 1707–1712) can also be used to form attributive adjectives. As Langacker notes (1991), attributive modifiers, as predicates, are semantically dependent on the nouns they modify –red does not mean the same thing in red pen and red army^[2]. But domain modifiers are semantically independent – the meaning of religious in religious leader is basically the same as the meaning of religious in religious book. The job of a domain adjective is to specify a (usually abstract) domain to which a concept or entity belongs, to reference the general category by which we can evaluate the noun. Sullivan (2013, Ch. 5), discussing domain adjectives, introduces the phrase moral wealth (a phrase that refers to high level of morality (from morality is money based on the moral accounting metaphor [Lakoff 1997]). In moral wealth, moral does not function as predicational material; the phrase does not mean wealth is moral. We understand moral wealth as a certain type of (metaphorical) wealth, in contrast to personal, material, national, mineral, etc. In phrases like this which include domain modifiers, the adjective position aligns with the target domain, and the modified noun, with the source. (In this case, wealth elaborates the money source domain.) So unlike attributive adjective constructions, which align with a source-target ordering, domain adjective constructions warrant a target-source alignment.

Based on these observations, we should expect particular alignment patterns between source and target domain triggers and the syntactic constructions they fill. These patterns extend beyond A-N constructions to others. Because source-target alignment is conceptual (and only cued by the syntax), in each case, the predicating/attributive modifier position, whether a noun, adjective, or verb, is expected to be filled by the source trigger and the modified noun position by the target trigger. Likewise, the alignment between domain-modifier and target trigger can be extended into other syntactic constructions. For example, domain specifiers often fill the first slot in N-N compounds and the possessor position in possessive constructions (Gunkel and Zifonun 2008: 284), i.e. education problems, educational problems, education’s problems, problems of education. The semantic and conceptual relations encoded in simple syntactic constructions allow a forecast of how source and target triggers will combine. Corpus data involving the same pairs of source and target triggers, thus, provide the input to test predictions about the lexicalization patterns of conceptual metaphors.

3.1 Lexicosyntactic alignment

Ordered searches ensured that source and target triggers were inputted into both positions of each construction, resulting in a total of 10 possible alignments: NP(Source)-of-NP(Target), NP(Target)-of-NP(Source), A(Target)-NP (Source), A(Source)-NP (Target), N(Target)- N(Source), N(Source)-N(Target), V(Source)-NP(Target), V(Target)-NP((Source), X is Y (Target is Source), X is Y (Source is Target). Triggers always filled head roles in the case of complex sub-constituents.

A total of 12,816 manual searches (McEnery and Wilson 2001) in COCA were carried out systematically by inputting (to COCA’s search interface) each lexical pair in both orders (i.e. financ* health*; health* financ*), allowing for up to five intervening wild cards (finance* ***** health*) in order to permit intermediate words to occur between the source and target triggers, i.e. steer (verb-source) the (wildcard) American (wildcard) economy (noun-target). Concordance line results from COCA were reviewed one by one and, when an example fit one of the ten syntactic configurations, the token was tagged into the appropriate syntactic construction grouping. This methodology allowed annotators to bypass COCA’s automated part of speech tagging algorithm.

3.2 Tagging metaphors

Coding for metaphorical use followed the MIP(VU) method, a commonly accepted standard for metaphor identification (Pragglejaz Group 2007; Steen et al. 2010). This method, applied to all searched collocations involves (1) reading each concordance line corresponding to collocation searches, (2) establishing the contextual meaning of the collocation in question and (3) marking the phrase as metaphorical when it is not used to signal a more basic, literal referent. For example, a floating market would not be collected as a metaphorical phrase when it references a farmers’ market in Southeast Asia built on floating platforms; however, a floating market, used to reference a numerical index of a particular economic market, would be marked as metaphorical. Only tokens that fit the syntactic construction and were used in metaphorical reference to the economy were logged as metaphorical hits. Two annotators agreed on all tokens tagged as metaphorical.

This procedure, totaling approximately 13,000 manual searches in COCA, and subsequent coding process, resulted in a database of 9,001 phrasal tokens representing 331 out of 1,008 possible pairing types.

4.1 A-NP prediction

On the surface, the A-NP construction seems to show no preference for source-target ordering alignment; results show 86 pairings in which the source trigger filled the adjective position and the target trigger filled the head noun (A(S)-N(T)) and 96 pairings in which the target trigger filled the adjective position and the source trigger filled the head noun (A(T)-N(S)). But, as outlined in (§2.5), adjectives, as modifiers, serve two separate functions -to specify domains (domain-specifying) and to encode attributive material (attributive-modifiers). Given the same pair of triggers, Sullivan predicts that the former type will be lexicalized via the target domain and the latter by the source. This tendency is overwhelmingly confirmed in the corpus data as illustrated in Table 2. Eighty-six pairings follow the Attributive-Adjective(S)-N(T) alignment, while there are no instances in which the attributive adjective is lexicalized with a source trigger (Attributive-Adjective(T)-N(S)). Ninety-one pairings follow the Domain-Adjective(T)-N(S) alignment, while there is only one instance in which the domain adjective is lexicalized with a source trigger (Domain-Adjective(S)-N(T)).

4.2 V-NP prediction

Since verbs, like attributive adjectives, provide predicational material about their arguments, according to Sullivan’s theory, we should expect the verb to be lexicalized with a source, not target trigger. Eighty-three pairings follow the V(S)-N(T) alignment, while there are only seven pair types in which the verb is lexicalized with a target trigger (V(T)-N(S)). Total counts are shown in Table 3.

4.3 N-N prediction

Sullivan’s predictions extend to N-N compounds. Noun compounds are complex in their internal semantic relations given the array of possible relationships expressed (cf. Ryder 1994). However, in metaphorical compounds that include one source and one target trigger, these are limited to two types. In one case, the first slot in N-N compounds is a domain modifier, as in job title (literal) or economy climate (metaphorical), while in the other cases, the first noun is filled by an argument and the second by a (deverbal) nominal, which encodes predicational material, as in car race (literal) or money flow (metaphorical). Thus, based on the notion that source domains provide predicational material for the corresponding target, we should predict a N(T)-N(S) alignment for both types of N-N compounds, but for different reasons. Compounds with domain-specifying nouns followed by their categorical constituent should mimic the Domain-Adjective(T)-N(S) alignment. Likewise, compounds of argument nouns followed by predicate-encoding (deverbal) nouns should also realize the source-domain trigger in the second slot, since source domains provide the predicational material. By evoking the source domain, the syntactic head, in both cases, carries the metaphorical load. The target trigger, is then relegated to a secondary, dependent syntactic role. The source domain must be syntactically prominent. The prediction is supported in the corpus data as illustrated in Table 4. Ninety pairings follow the N(T)-N(S) alignment, while there are only eight instances in which the first noun is lexicalized with a source trigger (N(S)-N(T)).

4.4 NP-of-NP prediction

The domain-specifier pattern of A-NP constructions and N-N compounds extends to relationships of possession, i.e. job title, title of job (literal) and economy climate, climate of the economy (metaphorical) In N-of-N sequences, the second noun typically fills the possessor role, semantically encoding the domain of belonging. As an illustration of the same semantic pattern, we should expect the target trigger to fill the possessor role, the second noun slot. The prediction is supported in the corpus data as illustrated in Table 5. One hundred twenty-six pairings follow the NP(S)-of-NP(T) alignment, while there are only five pairings (eight tokens) in which the first noun is lexicalized with a target trigger, the NP(T)-of-NP(S) configuration.

Instances of the minority pattern in which the target trigger fills the head position, the first noun slot, are explained through two subpatterns. In some cases, source domain language like (cash) flow in rate of (cash) flow, has been regularized to reference specific economic measures or institutionalized concepts, as is the case with bubble. In these tokens the source domain, not the target, serves as the domain specifier. In the second subpattern, exemplified through collocations like firms of health, the source trigger is attributional in nature. Health predicates information about the (positive) performance of a given financial institution.

4.5 X is Y

On a rare occasion, conceptual metaphors for the economy are lexicalized in copula construction. Predicates in copular clauses provide information about their subjects (Halliday 1967; Akmajian 1979: 162–165), mirroring the role source domains play in establishing the conceptual architecture to structure the target. In all save one of the 33 cases from the data, the source domain is realized in the predicate as would be expected. Specific source triggers appear mostly as predicate adjectives (market is healthy), occasionally as a predicate nominal (market is a bubble), and, in one case, within a prepositional phrase (on life support). Total counts are shown in Table 6.

The lone counterexample to the target is source ordering of the copula construction, the sole remaining cloud is corporate debt, is specificational in nature (cf. Mikkelsen 2005); the predicate (target) is classified though a source concept.

4.6 Interim summary

The asymmetrical distribution of source and target constructional alignment throughout the data is summarized in Figure 2; source domains lexicalize predicational material which corresponds to specific syntactic positions. These results are consistent with findings from other corpus studies showing that verbs and adjectives account for a large proportion of metaphorical expressions observed in natural language data (cf. Cameron 2003; Deignan 2005; Pragglejaz Group 2007; Shutova and Teufel 2010). The few exceptions to the very strong tendencies in the data come from lexemes that have been recruited into economic jargon such that they have shifted from simply encoding a nuanced concept to establishing a categorical domain., i.e. recovery, bubble, etc. The strong asymmetrical syntactic patterns in metaphorical lexicalization indicate that metaphoric construal is not generated by a simple pairing between a source and target trigger, nor a simple frame mismatch between two lexical items (as noted in Sullivan 2016: 151), metaphorical construal is based on a unidirectional mapping from source to target domain. Likewise, positional alignment between syntactic constituent and domain type indicates that the semantic function of syntactic positions plays a crucial role in the lexicalization of conceptual metaphors.

Figure 2:

Source-target type counts across five syntactic constructions.

Patterns of syntactic distribution reflect the predicational function of the source domain. These strong tendencies can inform predictions as to how source and target lexis will be encoded across the five construction and guide automated tagging systems (cf. Shutova 2015). Of the 530 pairing configurations, 95.8% (508/530) of type and 98.4% (8335/8471) of token counts follow the expected alignment. Table 7 illustrates the prediction for one source-target pair.

Reversals of the expected alignment result mostly in nonmetaphorical phraseology (marketed flood, flood market, market the flood, market of the flood, the flood is marketed, invested flood, flood investment, invest in the flood, flood investment, investment of the flood, etc.) These findings have important consequences for the meaningful identification of metaphorical language, both manual and automated. If systems can tag lemmas as belonging to pre-specified source and target domains in the taxonomy, they can predict their distribution when used in combination across these constructions. The corollary is also possible; when used in combination, the system can automatically categorize them as source or target lemmas.

5.1 Syntactic rigidity

Syntactic rigidity is one way to measure economic phraseology and the co-selection of source-target pairings. Another way to think about this concept is to gage which pairs are the most syntactically fluid – which pairings have the widest distribution across the syntactic constructions, and which seem to be frozen on one particular configuration. Lexical co-selection can be examined from several angles including consecutive sequences, n-grams (a lot of) (Cheng 2012: 72) and lexical bundles (you know what I mean) (Biber et al. 1999: 990), heads and dependents (play a role) (Cheng et al. 2009), colligation of content and function words (in some cases; in both cases; in most cases) (Stubbs 2002: 65), clause collocations (The reason … is because) (Hunston 2002: 75), genre and register specific patterns (Scott and Tribble 2006, Ooi 2008), and, finally, sequences of variable constituency, ‘concgrams’, (risk management; management of risk) (Cheng et al. 2006, 2009). Source-target pairings are best characterized through this last category, concgrams, since their co-selection can vary in syntactic configuration.

A large group of source-target/target-source collocations occur in only one syntactic pattern within the data probed: In COCA, 204 (20.2%) of all 1,008 possible pairings occur in one configuration only; 8.3% occur in two syntactic constructions. Only six pairings are found in all five configurations (the majority of possible source-target trigger pairings (785) do not occur at all, see Figure 3). This distribution suggests that metaphorical phraseology is variable in nature: Most source-target pairing that do occur in COCA are linked to one construction, while a small minority are syntactically fluid.

Figure 3:

Quantity of syntactic participation for source-target pairs.

5.2 Conventionality

A second measurement of lexicalization relies on frequency statistics as an indication of conventionality in a speech community (Cheng 2012). Conventionality can, in some instances, correlate with idiomaticity (Gibbs 1984; Gibbs and Nayak 1989). The identification of the most frequent source-target pairings for any given metaphorical model is an additional important metric for not only understanding metaphor lexicalization but for cataloguing metaphorical phraseology by domain. On one end, highly frequent collocations of set source and target triggers can be considered as common fixed or semi-fixed phrases (probably learned and stored as units [Sinclair 1991, 1994]). On the other end, more creative and novel collocations will be less frequent in corpus data, and, thus, not considered conventionalized, metaphorical jargon. Along the continuum, top pairings like economic recovery and market conditions occur in hundreds of examples, while other phrases are common, but not as prolific, i.e. forecast growth. This common jargon occurs in tandem with novel collocations found in just one or two corpus tokens.

These results confirm Gibbs’s (1984) observation that metaphorical expressions are spread along a continuum from, as Gibbs characterizes them, highly ‘conventional’, lexicalized metaphors to entirely ‘novel’ and creative ones, (sometimes referred to as ‘dead’ and ‘alive’ or ‘conventional’ and ‘innovative’ [cf. Traugott 1985: 33]). Conventionality (as gaged by frequency) allows for a more detailed picture of how conceptual metaphors are lexicalized and then adopted into speech communities as set phraseologies with different rates of usage.

5.3 Distributional coverage

‘Combinability’ measures the likelihood of a word or word class to appear in a recurrent word pattern (Eeg-Olofsson and Altenberg 1996). The grammatical status of a root inevitably affects its combinability score (Eeg-Olofsson and Altenberg 1996: 103). Verb roots like flow, which can manifest as a verb (flow), a noun (flow), and a modifier (flowing), are able to fill more syntactic positions than those roots which are morphologically restricted. Likewise, roots which combine with more derivational suffixes have an increased ability to appear in a larger variety of phrasal units. However, economic corpus data suggest that factors beyond just morphological variation contribute to a particular trigger’s rate and spread of usage. Corroborating past observations (cf. Winter 2016), frequency of metaphorical use in economic discourse overall correlates to a source trigger’s combinability score (r=0.59, p<0.05, n=48). On the other hand, neither a trigger’s metaphoricity rate (r=0.18, p<0.05, n=48) (given all instances, what percentage of token are metaphorical in economic discourse) nor keyness score (r=0.17, p<0.05, n=48) (rate of usage in economic versus basic contexts) correspond to a source trigger’s permissibility with multiple target triggers. (Keyness calculated based on rate of occurrence in The Economist corpus compared to the Academic section of COCA; metaphoricity is rate of metaphorical use in The Economist corpus.)

A given lexeme’s frequency in economic discourse and its increased combinability is a bidirectional correlative, not causal, relationship. Combinability and overall frequency of source triggers necessarily involve lexical semantics. Certain source domain concepts (and their associated triggers) play an outsized role in economic theory and discussion, and thus, pair with a wider range of target triggers. The top ten most widely distributed source triggers (flow, freeze, recover, sink, wave, climate, health, forecast, bubble, conditions) represent economic concepts central to the discussion of the economy. They are mostly nonspecific, abstract concepts, conventionalized in the economic literature because they evoke widely discussed theoretical notions. Flow, freeze, wave and channel are closely tied to the domain of water, central to the discussion of currency. Health and recovery are fundamental abstract concepts that index the (ailing) body frame. Forecast and conditions, similarly abstract, are concepts central to the weather source domain, but do not reference concrete elements like rain and snow. These highly distributed source triggers can be contrasted with source triggers very limited in their distribution.

The same reasoning explains why some triggers are limited to one or two pairings such as autopsy, keel, ballast, terrain and vagaries. These words participate in limited (semi-)fixed phrases in economic discourse because they reference very specific, in many cases, concrete, imageable aspects of the source frame. The precise and concrete nature of these lexemes prohibit their combinability and frequency. ^[3] For any given source domain, we can expect, within the constellation of triggers, concrete, specific, and imageable lexis to play a more limited role in metaphor lexicalization than abstract and semantically vague triggers, which will account for more instances of lexicalization.

The observed correlation between abstractness and combinability speaks to theories of consciousness and intentionality of metaphor usage. Steen (2011), among others, makes a distinction between deliberate and non-deliberate metaphor: ‘Deliberate’ metaphor, he argues, is a conscious analogical construction between domains, and ‘non-deliberate’ metaphors are frequent and conventionalized collocations or lexical senses. This is certainly a controversial notion in metaphor studies (cf. Gibbs 2017: 83–84), and not the focus of the present study. What seems clear, however, based on the results outlined above, is that any attempt to divide metaphor samples within a particular conceptual mapping should not be a categorical endeavor. That is, we should not take any one specific frequency or combinability score and posit it as a threshold for conventionalization. Some frequent source-target pairings like economic health index well established theoretical notions, while others like autopsy of the economy are based on the same set of metaphorical bindings but have not conventionalized. In the middle are hundreds of source-target pairs, used more than once, but not frequently.

5.4 Robust pairing asymmetries

Within some frequent source-target pairings, highly asymmetrical constructional participation can be found. For example, in COCA, there are 270 tokens of financial health and only two tokens of healthy finances. Financial health is used as a gage of the financial well-being of an institution: a bank or a local or national government as in (29)–(31):

Healthy finances is used in a similar context in COCA, but only twice, as in (32):

Part of the explanation for these asymmetries surely rests on the genre of discourse represented in COCA. COCA’s representation of economic discourse is skewed towards academic and journalistic genres focused on macroeconomic issues. Unlike financial health, mostly applied to institutions, healthy finances is often used to describe the financial state of an individual – a microeconomic discussion. For example, on the web, several self-help brochures include the phrase as in your guide to healthy finances and four elements to healthy finances. This latter discourse context may not be well represented in COCA. Thus, discourse genre can play a role too in the specific lexicalization pattern of a given source target pairing. Data source must be taken into account when looking at specific pairs. However, apart from discourse context, the health-finance disparity seems to mirror a larger pattern. Ordering asymmetries may also be indicative of a preference for target triggers to, when appropriate, lexicalize in domain adjective positions. Economy, as a target trigger exhibits the same asymmetry as does finance. Overwhelmingly, the two triggers are lexicalized through the A-Domain(T)-NP(S) pattern as in (33), even though the reverse ordering A-Attributive(S)-NP(T) is possible as in (34) (economic recovery=1,207, recovering economy=48; economic pain(s)=93, painful economy=1).

The adjectival use of economic in the A-Domain(T)-NP(Source) construction also eclipses the lemma’s use in the NP(S)-of-NP(T) construction: economic health=245, health of the economy=110; economic conditions=1184, conditions of the economy=9; economic recovery=1207, recovery of the economy=29). Other collocations with significant asymmetries appear to be idiosyncratic and thus, do not contribute to a generalized trend. When comparing the same source-target pairings across the NP(S)-of-NP(T) and N(T)-N(S) constructions, neither construction dominates: conditions of the market(s)=7, market conditions=349; health of the bank(s)=28, bank health=5; flow of investment=46, investment flow=73; flow of money=178, money flow=46.

6.1 Metaphoricity

Dunn (2011: 53) proposes that grammatical encoding can affect judgments of metaphoricity. For example, he hypothesizes that a metaphor encoded as a nominal predicate in a copula construction: This unrequited love has become an empty bottle is more explicitly metaphorical than a similar expression in which the metaphor is encoded in a verbal predicate: This unrequited love has swallowed up my sanity. In fact, a quite extensive literature focuses on metaphorical constructions in which an established conceptual metaphor or a novel blend is lexicalized through a subject-predicate sequence (cf. Love is a journey) (see Moder 2010). But the results of the present study show, in fact, that the vast majority of lexicalization patterns are encoded within syntactic subjects or predicates (A-NP, NP-of-NP, N-N, V-N) not across syntactic subjects and predicates (X is Y), (see Figure 2). Source-target pairings are particularly interesting because they are not necessarily recognized as overtly metaphorical by speakers, especially given their constituent-internal nature. NP- (stormy economy) and VP- (forecast growth) internal co-selection is clearly more hidden than is cross subject-predicate co-occurrence (the economy is ailing). This obfuscational role of grammar stands in response to metaphor skeptics like Pinker (2006, 2007). Certain patterns of grammatical encoding hide or mask conscious access to metaphor, as has been argued for other grammatical phenomenon like the encoding of reflexive predicates (cf. Talmy 2000a). The notion that grammar can in fact hide or highlight conceptual mappings reinforces the importance of paying attention to the system of metaphor and grammar, rather than basing arguments on isolated metaphorical examples. Any given instance of the lexicalization of a conceptual metaphor should not be taken as evidence for or against the “consciousness” of the metaphor; rather the system should be evaluated as whole, including complex relationships between grade of metaphoricity and frequency of metaphorical tokens. The present study shows just how time-consuming this type of evaluation is on even a limited grouping of (related) metaphors. Nevertheless, this type of work must be scaled up if we are to understand the interaction between lexical form and conceptual representation.

Even among metaphor researchers, the role of grammar is often overlooked. Steen (2007, 2011), for example, makes a distinction between metaphor-in-thought and metaphor-in-language. The fact that certain source-target pairing occur across several different constructions raises questions about Steen’s rubric for the formal indicators of the deliberate nature of metaphor based on lexico-grammatical alignment. For example, if we believe that certain instances of metaphor are reflections of the conceptual system while other instances are not, we might expect a difference in the source domain lexicon used in the predicate of the copular construction (a more transparent syntactic vehicle) in comparison, to say, the attributive-adjective position in the A-NP configuration (a more disguised vehicle). However, without a priori criteria to evaluate what lexis should appear in which type of construction, we have no way to evaluate that prediction. If anything, we might expect less frequent, more novel triggers to appear exclusively in more overtly metaphorical constructions like the copular sentence. But, the results above do not support that analysis: choppy market (15 tokens) is much more frequent than the market is choppy (1 token).

6.2 The system

The inversion of investigative methodology – starting with conceptual mappings from the literature, identifying metaphorical lexis attached to each domain, and then searching for collocations in natural language data – is a crucial next step in metaphor research. Metaphor identification is different from metaphor analysis (Petruck 2016: 135). For the many reasons discussed in this paper, it is essential to probe corpus data systematically in the evaluation of metaphorical lexicalization process, but this process needs to be carried out in a way that is linked to the ontology of conceptual metaphors. Otherwise, we lack a conceptual explanation for linguistic behavior.

Gibbs (2017) questions the role and goals of counting metaphors. He argues that just one instance of metaphor can be very powerful in affecting reasoning, a counterargument to the idea that quantifying metaphorical lexicalization by counting instances of metaphor is a reflection of the power or salience of metaphor. Perhaps, rather than questioning the endeavor, it is the type of counting that should be scrutinized. Simply cataloguing random words in sentences as metaphorical or not does not help us understand the systemic nature of metaphor nor the lexicalization of cognitive models, but linking metaphorical phraseology to the ontology of conceptual metaphor does. By measuring lexicalization of specific source-target pairings across syntactic constructions, we can systematically catalogue conventionality and entrenchment-factors clearly related to speakers’ assignment and judgment of metaphoricity.

To this end, the study of metaphorical phraseology is one important subdomain in the investigation of the relationship between lexis and grammar. Patterns of metaphorical phraseology demonstrate the coordinated nature of the lexicon and grammar. By offering conceptual explanations for certain fixed and semi-fixed phrases, the connection between cognition and language structure becomes clearer. Metaphorical mappings feed co-selection of source and target lexis and co-selection feeds and reinforces these same metaphorical mappings, a process similar to the development of positive and negative semantic prosody associated with particular words. The results of this study yield a complex portrayal of metaphor, lexis, and grammar. Metaphorical lexicalization patterns, just like literal lexicalization patterns, cannot be explained by one and only one phenomenon (see Gibbs 2017). Discourse practice, culture, grammar, language use over time, etc. all play a role in shaping source-target pairings within topic specific genres.