18.1 INTRODUCTION

YOU might have fallen on this page by chance, hopped from chapter to chapter, hunted down our title or stumbled on it while taking a gentle stroll through the volume or even rushing through; however you made it to these lines, you probably understood the past sentence—metaphors and all—effortlessly. Metaphor has been studied since antiquity as a rhetorical and poetic tool, yet it is pervasive in everyday communication. Indeed, it is estimated that a minute of speech will contain approximately six non-literal expressions, including idioms, novel and conventional metaphors (Glucksberg, 1989). In the case of metaphors, such as ‘Fashion is a religion’ or ‘Politics is a game’, terms are not understood literally, but figuratively: the topic of the metaphor (e.g. fashion, politics) borrows properties from the vehicle (religion, game), which generally belongs to a different conceptual category. Bearing with the Greek etymology of metaphor—metapherei, to transfer across—this process has been described by theorists from various backgrounds as a mapping across distinct cognitive domains (e.g. Lakoff, 1993; Fauconnier & Turner, 1998; Kövecses, 2010).

Metaphors differ from similes, which make an explicit comparison between the topic and the vehicle (‘Fashion is like a religion’) and can be understood as literal statements. They should also be distinguished from idioms. In idioms the meaning of a string of words is not deducible from the sense of the individual words; it may additionally be opaque, in which case the idiom is only interpretable if previously learned as such (e.g. to be ‘off one’s rocker’). As a result, the cognitive processes involved in understanding idioms might be closer to what Swinney (1979) uncovered for ambiguous words (which have two encoded meanings) than to the kind of pragmatic inference involved in metaphor comprehension. Finally, metaphor is generally separated from less-studied tropes such as hyperbole, metonymy, and understatement. Like metaphor, all these figures of speech attribute to a being, object, or action an expression that is not literally applicable. However, only in metaphor does this expression belong to another conceptual domain altogether. Nonetheless, recently some theoreticians have put forward a unified account of these phenomena, arguing that they all involve the same kind of pragmatic processes (Wilson & Carston, 2006, 2007; Sperber & Wilson, 2008).

The pervasiveness of metaphor in discourse and our ability to understand it quickly and effortlessly gives rise to a host of questions and has resulted in several programmes of research in the past decades. In this chapter we will focus on metaphor comprehension and the psychological processes that may account for it. The relation between metaphor and literal meaning is the first issue begging to be tackled. First, we must attempt to clarify how (if at all) the comprehension of metaphorical expressions differs from those interpreted literally, as well as the role played by the linguistically encoded sense of the term in the process of retrieving the intended metaphorical interpretation of an expression. The second pressing question is how the metaphorical meaning is arrived at and which factors are at play in the interpretative process. Here, the domain of research is dominated by the debate on whether metaphors are understood through a property-matching comparison process between the topic and the vehicle (Wolff & Gentner, 2000; Gentner et al., 2001), or whether metaphorical meanings are directly grasped as categorical assertions (Glucksberg, 2001, 2003). A further line of enquiry, which has received renewed interest recently, probes the differences and similarities in processing between different types of metaphors and between metaphors and other tropes. For instance, while novel metaphors are understood as readily as conventional ones, it is unclear whether they are processed in the same manner (Giora, 1997; McElree & Nordlie, 1999; Glucksberg, 2001). Finally, we will look at how the ability to interpret metaphors develops through childhood. This is a topic of interest in its own right, which additionally bears on the cognitive processes behind our proficiency in grasping metaphors.

18.2 METAPHOR AND LITERAL MEANING

18.2.1 Theoretical approaches to metaphor

According to classical rhetoric, in tropes such as metaphor, hyperbole, metonymy, and irony, the encoded literal meaning is replaced by a related figurative meaning. The standard pragmatic account of metaphor provided by Grice (1975/89) follows this stance in many ways. Metaphors, as well as other tropes, are blatant violations (floutings) of the rule of conversation that enjoins us to be truthful (the maxim of quality: ‘Do not say what you believe to be false’). For all these tropes, these violations are designed to convey a related true implicature. What changes between tropes is the type of related implicature they communicate. Since politics is clearly not a game, the sentence ‘Politics is a game’ is literally false and will trigger the search for an alternative interpretation that makes sense in context. This theory posits that metaphor involves an element of indirectness, something which may not always accord with our intuitions. Furthermore, as we will see section 18.2.2, while Grice’s theory is not meant to be a processing account (Noveck & Spotorno, 2013; Geurts & Rubio-Fernández, 2015; Spotorno & Noveck, Chapter 17 in this volume), its most renowned adaptation in processing terms—the three-stage model—is not born out by the empirical evidence.

Some recent theories of metaphor suggest, instead, that metaphorical expressions are directly understood with a relevant occasion-specific sense, different from, but related to, their literal meaning. This interpretation results from context-driven meaning adjustments, which involve broadening (and occasionally narrowing) of the word’s encoded meaning, leading to the construction of a context-specific ad hoc concept (Recanati, 2004; Wilson & Carston, 2006, 2007; Sperber & Wilson, 2008; Carston, 2010). The kind of pragmatic inference put forward by this account is different from the one defended by Grice and other post-Griceans (like John Searle), even though they all crucially require taking into account the context of utterance, the common ground between speaker and hearer, and the recognition by the hearer of the speaker’s communicative intention. François Récanati’s views on metaphor share some central features with those of relevance theorists (such as Robyn Carston, Dan Sperber, and Deirdre Wilson), which depart from Gricean accounts. According to them, the metaphorical interpretation is not part of an implicature conveyed by the utterance; rather it is a lexical adjustment or modulation of the linguistically encoded concept. As such, it is a phenomenon which takes place locally—unlike implicatures which operate on the whole proposition expressed by the utterance. The metaphorical meaning is therefore viewed as contributing to the explicit meaning of the utterance—to the truth-conditional content of what is asserted.

Récanati and relevance-theorists thus agree—against Grice and other post-Griceans—that metaphorical interpretations are the result of a meaning modulation, rather than an implicature, and that they contribute to what is explicitly—rather than implicitly—communicated by the utterance. Yet their accounts differ when it comes to the cognitive processes involved in grasping the metaphorical interpretation. Relevance theorists defend the claim that metaphors, despite contributing to the explicit meaning of an utterance, are understood through a pragmatic inferential process (Sperber & Wilson, 2008). By contrast, while Récanati also considers that implicatures are retrieved through inference, according to him what gives rise to the metaphorical interpretation is a more associative pragmatic process (Recanati, 1995, 2004). The possibility that some pragmatic processes, including metaphors, are arrived at through associative processes rather than inferential ones is the topic of active debate (Mazzone, 2011, 2014a,b; Mazzarella, 2014).

Another very different account of metaphor prevalent in the past decades has been the conceptual metaphor theory first introduced by Lakoff & Johnson (1980b). According to this view metaphor is not merely a linguistic phenomenon. The constant presence of metaphoric expressions in our everyday speech is an epiphenomenon of the way our conceptual representations are shaped by cognitive metaphors (Lakoff, 1987, 1993). In fact, ‘[t]he locus of metaphor is not in language at all, but in the way we conceive one mental domain in terms of another’ (Lakoff, 1993: 203). This thesis diverges fundamentally from other accounts and the relation between literal and metaphorical meaning loses some of its importance there, with the shift of the debate from the linguistic to the conceptual domain. The theory has had numerous fruitful developments and elicited interesting experimental work (for a critical discussion of the thesis and related empirical evidence, see McGlone, 2001).

18.2.2 Literal first hypothesis

As noted before, Grice’s account is not a psychological one. Nevertheless, his treatment of metaphor has been adapted by psychologists into a processing hypothesis (Clark & Lucy, 1975; Janus & Bever, 1985). This theory has given rise to a three-stage model of metaphor comprehension according to which metaphors first receive a literal reading, which is then rejected and ultimately replaced by a figurative one. This adaptation predicts that the hearers will expect literal truthfulness (as required by Grice’s maxim of truthfulness) and should therefore try a literal interpretation first. If this reading is subsequently found to be nonsensical in context, this will, in turn, prompt the hearer to look for a contextually relevant non-literal alternative interpretation (e.g. Searle, 1979; Glucksberg & Keysar, 1993; Glucksberg, 2001). When taken as a processing account, this model has two important implications. First, the literal meaning should always be arrived at first, and only subsequently rejected, before being replaced by the figurative reading. This implies that reaching a metaphorical interpretation should take longer than processing a literal one. Second, unlike the processing of the literal meaning of an utterance, processing its metaphorical interpretation should be optional, in that it could be ignored and one could decide to stick with the literal interpretation and go no further. Experimental research indicates that these claims are both incorrect: the understanding of metaphorical expressions in context can be as easy and fast as that of a literal turn of phrase (Gibbs, 1994a; Glucksberg, 2001, 2003; Gibbs & Tendahl, 2006) and it is not optional, but mandatory and automatic (Glucksberg et al., 1982; Gildea & Glucksberg, 1983; Keysar, 1989; Wolff & Gentner, 2000). There is no priority of literal meaning (see Glucksberg, 2001, 2003, and the literature mentioned therein), although interestingly recent Event-Related (brain) Potential (ERP) studies suggest a possible prior role of the literal meaning at the neuronal level (Bambini & Resta, 2012; Weiland et al., 2014).

18.2.3 What place for the literal meaning?

As a result of the empirical evidence just mentioned, there is now a consensus to reject sequential processing models of metaphor, in which a default literal interpretation must be arrived at and rejected as contextually inappropriate before a more appropriate figurative alternative meaning can be derived. Instead, parallel processing models are favoured (Gentner et al., 2001; Glucksberg, 2001; Rubio-Fernández et al., 2014). These come in a variety of flavours. For instance, the Graded Salience hypothesis (Giora, 1997, 2008) holds that possible meanings of an expression (literal or metaphorical) are accessed in order of their salience. Salient meanings of an expression are those stored in the mental lexicon (i.e. which can be evoked in isolation, without contextual support); and meanings are also ordered by salience. In conventional metaphors (e.g. ‘Your computer is a dinosaur’), both the metaphorical (obsolete) and literal (extinct reptilian) meanings are stored in the lexicon; hence, both literal and metaphorical meanings of the expression should be activated regardless of context. This account further predicts that when the metaphorical meaning of an expression is particularly more salient than its literal meaning, then the literal meaning might not be accessed at all. Conversely, since the metaphorical meaning of novel metaphors is not stored in the lexicon, the salient literal meaning will be accessed first regardless of the context in which the expression appears.

An important corollary of Giora’s account is the suppression/retention hypothesis, according to which meanings that are incompatible with the context (e.g. literal meanings in metaphorically biasing contexts) may remain activated, unless they are a hindrance to constructing the final interpretation, in which case they are suppressed. Since suppression and retention, as described by this account, are not automatic processes but are sensitive to context and inference, they should not happen immediately but in later stages of comprehension. This hypothesis predicts that salient meanings that don’t fit the context will be active in the first stages of processing and will be suppressed in the later stages of processing if they are a hindrance to constructing the appropriate interpretation. Indeed, this aspect of the theory echoes several other accounts of metaphor processing. Most current accounts endorse the view that while processing novel nominal metaphors, features associated with the literal meaning of the metaphor vehicle lose activation. Some, like Giora, view this as evidence of active suppression taking place during the processing of novel metaphors (Gernsbacher, Keysar, & Robertson, 1995; Gernsbacher, Keysar, Robertson, & Werner, 2001; McGlone & Manfredi, 2001; Pierce et al., 2010; Rubio-Fernández, 2007).

In a cross-modal lexical decision experiment, for instance, Rubio-Fernández (2007) found that, after reading the metaphor ‘John is a cactus’ following a metaphor biasing context (e.g. John hates physical contact), literally related superordinate terms like ‘plant’ (which according to Giora are a hindrance to the interpretation) and distinctive features like ‘spike’ (which according to Giora are not a hindrance) were both activated during the first stages of interpretation (0 milliseconds (ms) and 400ms after target word onset). At a later stage (1,000ms after onset) the superordinate terms (plant) were suppressed whereas the distinctive features (spike) were not. This lends support to the general view that novel metaphor comprehension involves active suppression of irrelevant literal senses. These findings are also compatible with Giora’s (2008) view that, in later stages, non-automatic suppression processes will not target all the meanings that are incompatible with the context but only the ones that are a hindrance to constructing the final interpretation. However, whether these data are actually in support of the graded salience account hinges on the notion of a meaning being detrimental to the interpretation, which does not have a strict definition in Giora’s account.

During metaphor comprehension it seems that metaphor-relevant information must be attended to, while information irrelevant to, or inconsistent with, the metaphoric interpretation must not be allowed to interfere with processing. For this reason, inhibitory control may play an important role during metaphor processing. Inhibitory control allows individuals to inhibit responses to irrelevant stimuli while working towards a cognitively represented objective (Rothbart & Posner, 1985; Eslinger, 1996; Zelazo et al., 1997). The psycholinguistic research we just discussed indicates that novel metaphor interpretation involves the active suppression or inhibition of metaphor-irrelevant features (Gernsbacher et al., 2001; McGlone & Manfredi, 2001; Rubio-Fernández, 2007; Pierce et al., 2010). If this is the case, then inhibitory control could be recruited during metaphor processing to regulate the activation of relevant and irrelevant features. Indeed, studies with some neuroatypical populations suggest that their known difficulties with inhibitory control affect their ability to suppress the literal interpretation in metaphor processing—e.g. Alzheimer’s disease (Amanzio et al., 2007), Asperger syndrome (Gold et al., 2010), and Parkinson’s disease (Monetta & Pell, 2007).

There is now a consensus that metaphorical meanings are understood as quickly and automatically as literal expressions (but see Bambini & Resta, 2012, and Weiland et al., 2014, for a more nuanced view on the time-course of metaphor interpretation and the potential role of a literal stage). Yet most issues about their interpretation and their processing remain unsettled. For instance, are we to understand metaphors as the result of inferential or associative processes? Little experimental work has tackled the question directly (but see Rubio-Fernández, 2012; with critical comments in Mazzarella, 2013; Mazzone, 2016). It is also still unclear how the linguistically encoded meaning is silenced; is it a process of decay or one of active suppression? If the latter, then what is the involvement of inhibitory control? In section 18.3 we will turn our attention to the metaphor processing debate that has dominated the scene in the past decades, and still does: do metaphors involve a comparison process between the topic and the vehicle, or are they understood as categorical assertions, as the work of post-Griceans such as relevance theorists and Récanati suggests?

18.3 HOW ARE METAPHORS PROCESSED?

There is a long-standing debate in the psycholinguistic literature between two views of how we process metaphors: categorization and comparison. The categorization view of metaphor comprehension holds that we process metaphors of the form X is a Y as categorization statements in which we modulate the meaning of the vehicle term. For example, we understand ‘his computer is a dinosaur’ to mean that ‘his computer’ belongs to a metaphorical category of ‘dinosaurs’ which includes not only extinct reptiles but also old computers and other ‘obsolete’ things. The comparison view, on the other hand, holds that metaphors are implicit similes and so they are understood by comparing the literal categories of topic and vehicle terms. So ‘his computer is a dinosaur’ is understood as the simile ‘his computer is like a dinosaur’, which invites the hearer to infer which features of the literal category of dinosaurs are shared with the topic (e.g. old, extinct, etc.). The idea that all metaphors are understood by comparison has lost currency in metaphor research, while the idea that some metaphors are understood by comparison and others by categorization has gained popularity. Each of these two views is subscribed to by a number of theories—e.g. Relevance Theory, Récanati and the Graded Salience hypothesis subscribe to the categorization view. We will focus on two accounts which are representative of these two positions: the class-inclusion model and the Career of Metaphor hypothesis.

18.3.1 Class-inclusion model

In Glucksberg & Keysar’s (1990) class-inclusion model, metaphors are class-inclusion statements where the vehicle term acquires dual reference: it refers both to the category denoted by its conventional meaning and to a new category which extends to all the things that exhibit some relevant characteristics of the vehicle. In our example, ‘dinosaur’ can refer either to an extinct reptile or to the category of old/obsolete things. The class-inclusion account was then expanded into the interactive property-attribution model (Glucksberg et al., 1997), which describes how the properties of the vehicle which determine the new metaphoric category are determined by the topic’s relevant dimensions of attribution. For example, computers have the relevant dimension of attribution of how old they are. The interactive property attribution model makes more specific predictions of what makes a metaphor good and how topic and vehicle make different contributions to metaphor comprehension. A good metaphor is one where the vehicle clearly and unambiguously exemplifies candidate properties for attribution and the topic provides relevant dimensions of attribution.

18.3.2 Career of Metaphor hypothesis

The Career of Metaphor hypothesis (Bowdle & Gentner, 2005; Gentner & Bowdle, 2008) proposes that we understand a metaphor either as a comparison or as a categorization depending on our experience with the metaphorical use of the vehicle. If the vehicle is being used in a novel metaphorical use the whole metaphor is novel and it will be understood as a comparison. Novel vehicles do not have a stable (i.e. conventional) metaphorical meaning and they don’t evoke a metaphorical interpretation in isolation. For example, ‘this audition is a door’ is a novel metaphor as there is no conventional metaphorical meaning for door. According to the Career of Metaphor account, we understand this metaphor to mean ‘this audition is like a door’ and we interpret it by aligning the two conceptual structures (Gentner & Clement, 1988) of ‘audition’ and ‘door’ to match features of the two concepts that play parallel roles (e.g. both an audition and a door can be a gateway to something new). This comprehension procedure for novel metaphors makes use of Gentner’s (1983) structure mapping theory of analogy, which Gentner and colleagues extended to metaphor comprehension (Gentner & Clement, 1988; Gentner & Wolff, 1997). Conventional metaphors instead have vehicles carrying a stable metaphorical meaning which can also be evoked in isolation. For example, dinosaur has a conventional metaphorical use and therefore, according to the Career of Metaphor account, we understand my computer is a dinosaur as a categorization statement which puts ‘my computer’ in the category of metaphorical dinosaurs. Therefore, the comparison view and the Career of Metaphor both maintain that when a metaphor vehicle has a conventional vehicle that can be retrieved from the lexicon, the metaphor is processed as a categorization statement.

18.3.3 Aptness and conventionality

The empirical predictions of the two accounts we outlined earlier have often been formulated in terms of the effects of aptness and conventionality on metaphor comprehension. These two properties have both been found to facilitate access to the metaphorical sense of an expression in processing.

The effect of aptness has been interpreted as lending support to the class-inclusion model (e.g. Jones & Estes, 2006). In a very general sense, aptness is a measure of how good a metaphor is. However, aptness has sometimes been defined more strictly as ‘the extent to which a comparison captures important features of the topic’ (Chiappe & Kennedy, 1999). Aptness of metaphors is often measured through norming studies in which participants give a subjective judgement of how apt a metaphor is or how natural it sounds (e.g. Katz et al., 1988; Blasko & Connine, 1993). The aptness of a metaphor has repeatedly been found to correlate with how fast it can be processed (Blasko & Connine, 1993; Chiappe & Kennedy, 1999). Although this variable has been widely studied in the psycholinguistics of metaphor, it is a fairly vague construct. Aptness is likely to capture the effects of several psychological dimensions and it is debatable whether the information gathered from this variable can reliably inform theory (Blasko & Connine, 1993; Gentner & Bowdle, 2008). As Thibodeau & Durgin (2011) put it, aptness ratings can only measure the perceived quality of a metaphor, but they cannot explain it. Indeed, it may be productive to try to explain why some metaphors are more apt than others rather than use aptness as an explanatory tool. The interactive property attribution model would allow aptness to be broken down into more specific components which would be better suited to testing the predictions of that account. However, to the best of our knowledge, no method has yet been proposed which would quantify the two components of dimensions of attribution and candidate properties.

The effect of conventionality is bound to the Career of Metaphor hypothesis (Bowdle & Gentner, 2005; Gentner & Bowdle, 2008), which proposes a qualitative difference in processing between novel and conventional metaphors. At the time this account was first put forward there was already empirical evidence that the familiarity of a particular metaphor facilitates comprehension. For example, Blasko & Briihl (1997) found that familiar metaphors are processed more quickly than unfamiliar ones. Furthermore, in a cross-modal lexical decision task, Blasko & Connine (1993) found that familiar metaphors have an immediate priming effect on words related to the metaphorical and literal senses of the vehicle while unfamiliar metaphors only prime literally related words (this effect interacted with aptness as unfamiliar-apt metaphors showed some activation). However, we should distinguish conventionality from the operational definitions of conventionality used in these studies. Within this theoretical framework, conventionality is in fact construed as a more specific formulation of familiarity which depends only on the vehicle. Bowdle & Gentner (2005) measured conventionality by first asking a group of people to read a list of metaphors (e.g. my computer is a dinosaur) and generate a property which captured the meaning of each metaphor (e.g. old). They then took the most common interpretation given by the first group of participants and asked a second group to rate how conventional that property was as an alternative meaning of the metaphor vehicle (e.g. how conventional is old as an alternative meaning of dinosaur?). The average rating collected with this last measure for each metaphor is the metaphor’s conventionality score.

Since the Career of Metaphor hypothesis proposes that novel metaphors are comprehended as similes (i.e. comparisons) while conventional metaphors are comprehended as categorization statements, experiments investigating conventionality sometimes compare the categorical form of a figurative statement (My computer is a dinosaur) and the comparison form (My computer is like a dinosaur). Bowdle & Gentner (2005) found that a figurative statement is both preferred and easier to process in the form which reflects how it is processed according to the Career of Metaphor hypothesis, in that novel metaphors are processed as similes and so they are preferred in the comparison form, while conventional metaphors are processed as categorization statements and therefore they are preferred in the categorical form. Theorists who do not hold similes and novel metaphors to be equivalent assume that metaphors and similes are simply different statements (Chiappe et al., 2003; Glucksberg & Haught 2006). Although the form preference paradigm is not a crucial test of their theories, they will predict that aptness increases the preference for categorical form of a metaphor over the comparison form (Jones & Estes, 2006). This is because low aptness metaphors are harder to comprehend, but they become more comprehensible if transformed into a simile as anything may resemble any other thing to some extent. Another possible explanation for why apt metaphors may be preferred in the categorical form is linked to a type of implicature sometimes carried by simile, in that the equivalent categorical statement is not true (My computer is like a dinosaur may implicate that my computer is not a dinosaur). The comparison form might therefore be dispreferred because it carries this implicature (Rubio-Fernández, Geurts, & Cummins, 2016).

Finally, the effects of conventionality on processing have also been measured with ERPs. Arzouan et al. (2007) performed a lexical decision task with two-word expressions and found that there was a linear increase in the N400 effect from literally related words (problem resolution), to conventional metaphors (lucid mind), to novel metaphors (ripe dream), to semantically unrelated pairs (indirect blanket). Lai & Curran’s (2013) findings are more telling with regards to the Career of Metaphor predictions. They found that reading a metaphor (Ideas can sometimes be bumpy) after a related sentence (I can see the path of his ideas) decreased the difference in N400 between reading a metaphorical and a literal statement for conventional metaphors but not for novel ones. Conversely, they found that reading metaphors after a related simile (Ideas are like roads) decreased the metaphorical-literal N400 effect for novel metaphors, but not for conventional ones. They interpreted these results as showing that novel metaphors are interpreted by a process of comparison that can be facilitated by a comparison-statement prime.

Despite being tied to the predictions of competing accounts, aptness and conventionality have been found to correlate (Bowdle & Gentner, 2005; Jones & Estes, 2006). Jones & Estes (2006) criticized the practice of studying one of these variables without controlling for the other and they constructed a set of items where aptness and conventionality were not correlated. They found no evidence for an effect of conventionality on the preference or reading speed of metaphors in the comparison (vs categorical) form. These kinds of results have been used by supporters of the categorization account to argue that aptness is mediating the effect of conventionality. However, the results of in vitro conventionalization experiments (e.g. Gentner & Bowdle, 2005; Thibodeau & Durgin, 2011) suggest that conventionality does have an effect on metaphor comprehension independently of aptness.

In summary, we have presented two views on metaphor processing, the categorization view, embodied by the class-inclusion model, and the Career of Metaphor hypothesis, and some of the experimental work associated with these two views, which is often tied to the variables of aptness and conventionality. One issue in metaphor processing research is that although the constructs used are defined differently (e.g. see Table 18.1), the variables that operationalize these constructs, which are mostly measured with subjective ratings, often correlate and possibly overlap in the phenomena they actually measure. For this reason, it is difficult to draw conclusions about their relative importance if they are studied separately (Jones & Estes, 2006). One way to address this issue is to move away from subjective ratings where possible, as Thibodeau & Durgin (2011) do for familiarity. Another way is to replace constructs that may be too vague when operationalized, which Gentner & Bowdle (2008) argue is the case for aptness, with more specific and more theoretically informative ones. The debate between the two views we presented in this section is still open and there is room for experimental research to inform this discussion; in particular, more research is needed on the processing of completely novel metaphors.

Table 18.1. Definitions of variables in metaphor processing research

images

18.4 VARIOUS METAPHORS, VARIOUS TROPES

So far, we have discussed the relation between metaphor and literal meaning and how we arrive at a metaphorical interpretation during processing; throughout the chapter, though, we talked about metaphor as a single phenomenon. However, there are many different kinds of metaphors and it is an open question whether they are all processed in the same way. We will address this issue in the first part of this section, while turning to a comparison between metaphor and other tropes in the second.

18.4.1 Are all metaphors processed in the same way?

Metaphor is not an entirely unified phenomenon; in fact, discussion in previous sections highlights that the distinction between novel and conventional metaphors plays an important role in theoretical and experimental research. While it is established that apt novel metaphors are no more difficult to understand than familiar conventional ones (Giora, 1997; McElree & Nordlie, 1999; Glucksberg, 2001), they might be interpreted through different processes. One prominent feature of novel metaphors when compared to conventional ones seems to be a lingering of the literal sense in interpretation. The distinction between novel and conventional metaphors assumes special importance in some accounts of metaphor comprehension, such as the Career of Metaphor hypothesis discussed in section 18.2.2.

Giora (1997, 2008) offers another approach to the distinction in her Graded Salience hypothesis—discussed in section 18.1.3. Giora & Fein (1999b), for instance, asked participants to complete word fragments (e.g. h_r_e) with the first word they could think of after reading familiar and less-familiar metaphors in either a metaphorically biasing context or in a literally biasing context. For familiar metaphors, they found that participants produced both literally related and metaphorically related words, with an advantage for the word that matched the context—more metaphorically related words were produced after a metaphorically biasing context; more literally related words after a literally biasing context. For less-familiar metaphors, however, they found that participants produced more literally related words regardless of context and that they produced almost no metaphorically related words following a literally biasing context. They interpret these data as indicating that the metaphorical sense is more salient and more accessible in familiar metaphors than in unfamiliar metaphors. In familiar metaphors both literal and metaphorical senses are salient, therefore they are both activated and the meaning that fits the context ‘wins’. In unfamiliar metaphors there is no salient metaphorical meaning so the literal meaning will be highly active regardless of context and the metaphorical meaning will not be activated at all without contextual support (i.e. in literally biased contexts).

Rubio-Fernández (2007) provides still another viewpoint on the distinction between novel and conventional metaphors. Although she takes the view that all metaphors are understood as categorizations (see section 18.2), she presents the continuum from novel to conventional metaphors as a line connecting two different ways of processing metaphors: from meaning construction to meaning retrieval. Novel metaphorical uses are constructed from scratch in context since the word has no stored metaphorical sense. For words with a stable conventional figurative meaning, the metaphorical sense can be retrieved from the lexicon, as it would be for literal meanings. This is, in our view, a very useful perspective on metaphor processing and we hope future research will take it more into consideration. This perspective also paves the way for a stronger link between studies on ambiguity and metaphor research. Many constructs from ambiguity research can be fruitfully applied to metaphor processing research. For instance, the study of the relative dominance of the different senses of a word can be applied to metaphor in terms of the relative dominance of the metaphorical and literal meanings (Dulcinati et al., 2014).

While different approaches account for the distinction between novel and conventional or familiar metaphors in different ways, there is a general recognition that such a distinction should be made and that it corresponds to meaningful differences in how we interpret and process them. In fact, findings in neurolinguistics seem to corroborate as much. Extensive empirical evidence points towards a special involvement of the right hemisphere in metaphor processing—language processing being by and large a left hemisphere affair. Rather than being linked to metaphoricity, it seems right hemisphere processing is dedicated to novel and unusual meanings in general, whether figurative or literal (e.g. Chiarello, 1991; St. George et al., 1999). Nevertheless, the neural correlates of interpretations of literal vs. metaphorical expressions and conventional vs. novel metaphors are highly debated. Indeed, recent evidence—more attentive to unrelated factors affecting right hemisphere processing—suggests that a distinction in terms of left vs. right hemisphere processing or even type of neuronal processing is misleading, and that the differences are ultimately much subtler (Forgács et al., 2012; Forgács, 2014).

The difference between novel and conventional metaphors is central to several theories and has been in the spotlight for some time, but it isn’t the only distinction to be made among metaphors. Another example of how there may be different paths to understanding metaphors is the distinction between single metaphors and extended metaphors. Compare the kinds of metaphors we have considered so far with the following literary example from Montale (1971, our translation):

I went down, arm in arm with you, at least a million stairs
And now that you’re not here, there’s the void at every step

Carston (2010) proposed that while we understand single metaphors similarly to loose uses of language, literary and extended metaphors involve a different processing path. The effort of modulating the lexical meanings and constructing ad hoc concepts for each of the words used metaphorically in an extended metaphor is too great and the accessibility of the literal meanings is very high and primed by the rest of the extended metaphor. We therefore switch to a different mode of processing in which we access and maintain the literal meaning of the words used metaphorically without constructing related ad hoc concepts; we recover an overall metaphorical meaning only at the end of the extended metaphor. In the example above, for instance, we do not construct ad hoc concepts for ‘arm in arm’, ‘stairs’, and ‘void’, we only access their literal meaning. A consequence of this processing mode is that, while in single metaphors the metaphorical interpretation is the explicit content, in extended metaphors the explicit content is literal and the metaphorical meaning is communicated via a cluster of weak implicatures. The metaphorical meaning is arrived at more slowly than single metaphors and through reflective inferences that are more under conscious control. In the example above, some of the weak implicatures that compose the metaphor meaning might be that the speaker feels his life to be empty and grief causes him to struggle even with small tasks. Rubio-Fernández, Cummins, & Tian (2016) found initial empirical support for Carston’s view. In a series of experiments they asked participants to read passages that ended in a metaphorical target sentence that was either a single metaphor (i.e. the passage contained no other metaphors) or part of an extended metaphor that continued from the previous text. Rubio-Fernández and colleagues found that the target sentence took longer to read and was recalled better when it was a single metaphor compared to when it was part of an extended metaphor (or compared to a control literal sentence), while extended metaphors and literal controls did not seem to differ in this respect. They interpreted their results as indicating that single metaphors involve additional effort since they require the reader to recover the metaphorical sense immediately, which leads to longer reading times. They also involve a special process (i.e. local modulation) which makes them stand out from the rest of the passage and leads to better recall. This contrasts with control literal targets and extended metaphor targets which are not processed differently from the rest of the passage and at a first pass only involve recovering the literal meaning. Rubio-Fernández and colleagues themselves point out that extended figures of speech—such as parables and allegories—are a relatively unexplored topic and we echo their wish for further theoretical and experimental work in the area. Similarly, there is as yet almost no systematic empirical work on the interpretation of literary metaphor (but see Bambini, Resta, & Grimaldi, 2014, and Bambini, Canal, Resta, & Grimaldi, submitted, for a first attempt at it).

18.4.2 Metaphors vs. other tropes

The relation between metaphor and other tropes has been a regular matter of discussion, starting with simile. The three-stage model of metaphor comprehension inspired by Grice’s theory and discussed in section 18.1.2 often implied that metaphors are interpreted by transforming them into the corresponding similes. While this particular view is not empirically supported (e.g. Glucksberg, 2001; Glucksberg & Haught, 2006), the role of comparison in metaphor comprehension is complex and still very much debated, as we saw in section 18.2. Setting aside the role of comparison in the processing mechanism of metaphor itself, simile is generally seen as the ‘literal version of metaphor’, its poorer cousin, supporting less cognitive, literary, or rhetorical effects. Interestingly, it has recently been argued to give rise to a range of distinctive effects that can at times make it ‘more powerful’ than metaphor (O’Donoghue, 2009).

Metaphor is also often compared to irony, which also results from a blatant flouting of the maxim of quality (Grice, 1975/89). According to the Gricean model discussed in section 18.1.1 the non-literal meaning of both tropes is recovered in the same way; one should therefore expect similarities in the way they are processed. Unsurprisingly, then, some of the theories targeting the three-stage model for metaphor have also been applied to irony. This is the case for Giora’s (1997, 2008) Graded Salience hypothesis (section 18.1.3), but also Gibbs’ (1994a, 2002) Direct Access theory according to which contextually appropriate figurative meaning, including metaphor and irony, can be arrived at directly without an initial clash with the literal meaning during processing. Yet irony is fundamentally different from metaphor in several respects. First, it has long been recognized that the—generally negative—attitude of the speaker is a crucial element for irony, but not for metaphor (Grice, 1989; Wilson, 2006; Wilson & Sperber, 2012). Second, understanding irony seems to require a higher degree of metarepresentation than other tropes and its processing might therefore rely more heavily on theory of mind: for a thorough discussion of theoretical implications and empirical evidence on these and other aspects of irony, see Spotorno & Noveck, Chapter 17 in this volume. Additionally, not only do irony and metaphor processing differ in many ways, but they follow different developmental paths, too (Winner, 1988/97).

Comparing metaphor processing to that of literal meaning or to other pragmatic phenomena such as implicature or irony help us highlight the mechanisms at play behind pragmatic inferences in general. However, if we want to understand the specifics of metaphor, other meaning shifts may offer a more informative comparison set. The similarities between metaphor and other meaning shifts such as understatement, metonymy or hyperbole have long been noted (Recanati, 2004; Sperber & Wilson, 2008). Indeed, some consider that these phenomena fall together in a continuum of loose uses (Wilson & Carston, 2006, 2007; Sperber & Wilson, 2008). Studying metaphor through the prism of meaning shifts highlights the similarities between metaphor and other phenomena in this category, but it may also help identify the differences and hone in on the precise singular functioning of metaphor interpretation.

A couple of studies directly compared metaphor and hyperbole. One study using a self-paced reading paradigm found that the same expression is read slower when it is interpreted metaphorically (in (a)) rather than hyperbolically (in (b)) or literally (in (c)) (Deamer et al., 2010).

(a) Sam always got lost. The university was enormous. It was a forest.

(b) The back yard definitely needed pruning. It was a forest.

A second set of experiments using several methods by Rubio-Fernández and colleagues (2014) concluded that in hyperbole there is a greater degree of semantic closeness between the literal and figurative meaning than in the case of metaphor. In their view, while both tropes involve a meaning shift, it is different in nature: hyperbole involves a quantitative shift along a scalar dimension of the literal meaning (e.g. in (a), ‘forest’ exaggerates the volume and density of vegetation), while metaphor requires a qualitative shift pushing its figurative meaning much further from its literal meaning by dropping defining features (e.g. the metaphor the university was a forest focuses on the experiential aspects of being in a forest but demotes the physical aspects of literal forests). The degree of closeness between the figurative and literal meaning of the two tropes could also account for the difference in how 3-year-olds fare with them. In a picture selection task with novel metaphors as well as novel hyperboles, they found hyperbole slightly easier to understand than metaphor (Deamer & Pouscoulous, submitted).

It is worth noting that a similar pattern emerges from research comparing metaphor and metonymy both in adult processing (Bambini et al., 2013) and in children’s development (Rundblad & Annaz 2010b; Van Herwegen et al. 2013). Bambini and colleagues (2013) find that adult metaphor interpretation incurs more difficulty and a greater cost than literal meaning, but also approximation and metonymy; while Rundblad & Annaz (2010b) conclude from their developmental findings that metonymy is cognitively more basic than metaphor. Schumacher (Chapter 19 in this volume) discusses similar discrepancies between metaphor and metonymy comprehension in some populations with language deficits suggesting that metonymy is ‘grounded more strongly in semantic relations’. A possible explanation for these findings may be one of the core differences between metaphor and metonymy: metaphorical expressions involve two conceptual domains while metonymies involve only one (Gibbs, 1994a; Glucksberg, 2001; Schumacher, Chapter 19 in this volume). In metaphor, the domains of the topic and the vehicle can be wildly different—generating on occasion striking effects—while it is the very nature of metonymy that the metonymic referent should be associated in some way with the literal referent of the expression and thus belong to the same domain. The association between literal and metonymic meanings may be of many different kinds: subordinate-for-superordinate, or the reverse; part-for-whole; places-for-occupants. They might be culturally shared, such as producer-for-product (e.g. author’s names for their work), but also contextually specific, as in Nunberg’s (1979) classic example of orders for restaurant customers (e.g. ‘The ham sandwich left without paying’). In all cases, though, the mere fact that there is an association ensures that—unlike metaphor—meaning exploration remains within the same domain.

Thus, taken together, these studies suggest that both metaphor processing and metaphor development differ in nature from that of other pragmatic inferences involving meaning shifts. The computation of metaphor seems to involve some difficulty over and above that observed in other types of meaning shifts, such as metonymy, approximation, and hyperbole. The special nature of metaphor might be due to several factors (for enlightening theoretical discussions, see Rubio-Fernández et al., 2014; and Carston & Wearing, 2015). One such potential factor is the involvement of inhibitory control, which as discussed earlier (section 18.1.3) might be crucial for metaphor, but possibly not for other meaning shifts.

We hope this section conveys the richness of recent research on the similarities and differences across types of metaphors and between metaphors and other tropes. These are also domains where many—perhaps most—questions remain open: how do lexical, sentential, or extended metaphors compare theoretically or in how they are processed? How do metaphors compare to fables or parables? And how do they compare to metonymy or hyperbole? In the past years the puzzle seems to have become more complex, bringing to light many new questions to be investigated.

18.5 HOW CHILDREN INTERPRET METAPHORS

Studies on development of metaphorical abilities follow, for the most part, two main lines of enquiry: identifying when children start understanding metaphors and the factors playing a role in the development of their metaphorical abilities. Developmental matters are, of course, interesting in their own right, yet they may also illuminate our understanding of metaphor as a phenomenon, as well as shedding light on some aspects of adult metaphor processing.

18.5.1 How early do children understand metaphors?

Since metaphor is pervasive in communication children must come to understand novel, as well as conventional, metaphors. When do they do so? Classic studies suggest that metaphor comprehension is a late developing skill (Nippold 1988/98; Winner 1988/97; Gibbs 1994a). For instance, Winner (1988/97: 41-4) argues based on a series of findings that genuine metaphorical interpretations do not emerge until 10 years of age. Yet some recent studies suggest a much earlier onset of metaphor comprehension (Waggoner and Palermo, 1989; Pearson, 1990; Özçaliskan, 2005, 2007; Stites & Özçaliskan, 2013; Deamer & Pouscoulous, submitted; Pouscoulous & Tomasello, submitted).

While different sets of findings seem conflicting, the gap between them may be bridged if we assume that the children’s observed difficulty in understanding metaphors might be due in part to impeding empirical factors, rather than reflecting children’s poor ability to derive pragmatic inferences. Several such factors may play a role, such as the use of age-appropriate world knowledge, task complexity, or the processing cost involved in the task, for instance (Vosniadou et al., 1984; Noveck et al., 2001). Two of these might be particularly relevant for the assessment of metaphor development. First, different types of figurative language are often conflated in experimental materials (Cacciari & Padovani, 2012). Conventional metaphors, or even idioms, do not rely upon the same interpretation process as novel metaphors. The former require learning culturally established links (e.g. the sexy or cunning fox), while the later are built on-line using the literal lexical meaning, world knowledge, and relevant contextual information. Novel metaphors, therefore, seem more appropriate for use in establishing children’s abilities to derive the pragmatic inferences involved in metaphor comprehension (Pouscoulous, 2011). Second, the metalinguistic tasks generally used to gauge metaphor competence require young communicators to reflect on the exact phrasing of the linguistic input (e.g. in a truth value judgement task or when asked to paraphrase or explain a sentence). As such, they place demands on children which may interfere with the accurate assessment of their pragmatic abilities. Indeed, Bernicot and colleagues (2007), for instance, find that children fail to show comprehension of a pragmatic phenomenon on a metapragmatic task despite understanding it several years earlier on a simple picture selection task (see Cacciari & Padovani, 2012 for a comparable argument concerning the acquisition of idioms). In fact, young children do poorly on metalinguistic tasks and seem unable to paraphrase metaphorical expressions correctly until maybe as late as early adolescence (e.g. Asch & Nerlove, 1960; Smith, 1976; Cometa & Eson, 1978). Asking children to explain metaphors verbally may lead us to underestimate their comprehension (Winner 1988/97; Gibbs, 1994a). It turns out that children fare much better with tasks that do not tap into their capacity to make metalinguistic judgements; based on these results they appear to understand metaphors as early as preschool (Waggoner & Palermo, 1989; Pearson, 1990; Özçaliskan, 2005, 2007; Stites & Özçaliskan, 2013) or even 3 years of age (Deamer & Pouscoulous, submitted; Pouscoulous & Tomasello, submitted).

An early ability to understand metaphor is in line with recent findings showing that 3- to 4-year-olds are proficient with other pragmatic inferences whose acquisition was believed to be delayed, such as scalar implicatures (Pouscoulous et al., 2007; Katsos & Bishop, 2011), relevance implicatures (Schulze et al., 2013) or presupposition (Berger & Höhle, 2012). It seems preschoolers can carry out complex linguistic pragmatic inferences when the experimental task is simple enough and when these inferences are relevant to them in context. Young children also master meaning shifts other than metaphor, which may be a better comparison group still than pragmatic inferences at large. They understand meaning shifts such as ‘The DVD was an hour long’ (Rabagliati et al., 2010), as well as metonymy (Falkum et al., 2017) as early as 3 years of age.

18.5.2 Which factors influence children’s understanding of metaphors?

As we noted in section 18.3.2, however, metaphor cannot be entirely conflated with other pragmatic phenomena or even other meaning shifts. In particular, it almost always involves category violation, which may imply that its processing and development require elements over and above an inferential procedure akin to other pragmatic phenomena. While some elements are crucial to enable young communicators to understand any pragmatic inference (e.g. semantic and syntactic knowledge, relevant context, and manageable processing cost), other abilities might be specific to metaphor development. What are the different elements involved in metaphor comprehension, whose development might influence it?

Semantic and syntactic knowledge are essential elements to feed into the process of understanding metaphor, as is the presence of context that makes the figurative reading relevant (this is particularly important for novel metaphors). Without these components, it is difficult to imagine how metaphor processing could even start. Furthermore, conceptual knowledge linked to both the topic and the vehicle of a metaphor is crucial to grasp the non-literal sense of a metaphorical expression (Gentner, 1977). Indeed, metaphor comprehension emerges contemporaneously across the same conceptual domain (Keil, 1986; Özçaliskan, 2005). Thus, the absence of sufficient or appropriate vocabulary or world knowledge might block children’s access to elements of metaphorical meaning.

Some abilities, nevertheless, seem specific to metaphor development and may not necessarily be involved in other pragmatic phenomena. Analogy perception, which emerges very early (Goswami, 1991, 2001), plays an essential part, even though, as was discussed in section 18.2, the exact nature of analogical reasoning involved in metaphor processing is subject to debate (Glucksberg, 2001, vs. Bowdle & Gentner, 2005, and Wolff & Gentner, 2011). Two mirroring skills must also be taken in consideration: the ability to refer to one and the same object with two different labels (‘alternative naming’ or ‘second labels’: Rubio-Fernández & Grassmann, 2016) and the ability to assign different senses to one and the same expression (polysemy). Preschoolers master the latter and understand ambiguous terms (Srinivasan & Snedeker, 2011), as well as meaning shifts (Rabagliati et al., 2010; Falkum et al., 2017). The picture is different for alternative naming, which 3- to-5-year-olds struggle with. However, they only do so when an entity is given two different names in the same context, creating two confronting perspectives within the same scenario (Perner, Stummer, Sprung, & Doherty, 2002; Perner, Brandl, & Garnham, 2003; Perner, Rendl, & Garnham, 2007; see also Matthews et al., 2010). This is therefore unlikely to be a direct issue for metaphor comprehension since, in experimental settings, the literal and metaphorical labels for an object are generally not used simultaneously. Finally, as discussed in section 18.1.3, inhibitory control may play a crucial role in metaphor comprehension. Importantly for metaphor development, children reach major milestones in their inhibitory control development between birth and 6 years of age, with a rapid improvement around the age of 4 (Reed et al., 1984; Livesey & Morgan, 1991; Thatcher, 1992; Gerstadt et al., 1994). In line with the work on neurotypical and neuroatypical adults, a recent study with typically developing 3-year-olds finds a link between performance on a picture selection task probing novel metaphor comprehension and two inhibitory control tasks (Deamer & Pouscoulous, submitted). These findings hint at a connection between difficulties with metaphor comprehension and inhibitory control development in early childhood. Interestingly, the same study also looked at 3-year-olds’ comprehension of novel hyperbole and found that they understood hyperbole even better than metaphor, but that inhibitory control did not play a role for hyperbole comprehension—unlike metaphor. This may be due to the fact that hyperbole, unlike metaphor, does not generally involve category violation, with meaning modulation operating within the same domain rather than across domains. Although some of the linguistically encoded features may not be relevant to the hyperbolic interpretation, they do not interfere either and need not be suppressed or inhibited. As suggested at the end of section 18.3.2, inhibitory control may thus be required for metaphor, but not for hyperbole.

While even preschoolers have been shown to understand metaphors, metaphor comprehension does not remain stationary through childhood. Three-year-olds are already proficient in deriving the intended metaphorical interpretation based on the literal sense and the context. Yet there is no doubt that their expertise with figurative language will continue to develop as their vocabulary, world knowledge, and processing resources grow (Vosniadou, 1987; Winner, 1988/97; Gibbs, 1994a; Pouscoulous, 2014). Future research should look carefully at the role of the different factors that are important in metaphor processing and its development. These include cognitive factors as well as linguistic ones, such as semantic knowledge, shown to be paramount in metaphor comprehension for children with Autism (Norbury, 2005; Gernsbacher & Pripas-Kapit, 2012). On the cognitive side, inhibitory control may play a fundamental role worth exploring further both in adults and children.

18.6 CONCLUSION

Metaphor has been studied relentlessly since antiquity and Aristotle. It has fascinated and puzzled scholars of all kinds: philosophers, theologians, rhetoricians, literary critics, linguists, etc. (see, e.g., Ashworth, 2007). While it has only become an object of scrutiny for psychologists in the past decades, this is but the most recent development of a very long history of intellectual endeavour trying to capture how metaphors work. The multiplicity of perspectives taken on metaphor—without yet coming close to exhausting its potential to spark research interest—is a tribute to the richness of the phenomenon. It is therefore no wonder that we have to close this chapter with more questions to be worked on than we can provide established answers for.

The psycholinguistic turn has yielded some important insights: metaphor is not an unusual type of speech. It is understood just as effortlessly and automatically as literal language is. This applies not only to conventional figures of speech, but also to apt novel metaphors—even though they might not be processed in the same way. Metaphors also feature in child speech much earlier and more naturally than previously believed.

Yet these findings seem to have done little more than open new avenues for research, both small and large. How we interpret metaphor is still very much debated—does it result from an associative or inferential process? Does it involve active suppression of the literal meaning? Which other cognitive and linguistic factors influence metaphor comprehension both in adults and children? What is the role of inhibitory control, for instance? How we process novel metaphors is also still an open question tied to an ongoing debate about whether a comparison or categorization process is involved in their comprehension. We hope to see more work on the way the theoretical constructs used in this line of research are operationalized and more experimental work on the comprehension of highly novel metaphors. Probing the similarities and differences between types of metaphors (lexical, sentential, extended) should also be an aim of research to come, as should a closer and more careful comparison of the cognitive demands imposed by metaphor and other figures of speech. Finally, several aspects of metaphor could not be discussed here; some of them are fascinating and still understudied. For instance, emergent properties (e.g. Wilson & Carston, 2006)—meaning features that are associated neither with the topic or the vehicle of the metaphor—have only given rise to some initial empirical work (Rubio-Fernández, 2012). Another area where research is still very much in its infancy is individual differences in metaphor comprehension, such as the role of literacy (Di Paola et al., 2016). This chapter aimed to summarize the research in a prolific field and we hope to have shown there is currently no shortage of theoretical and empirical issues to tackle, nor experimental methods with which to go about doing so.

ACKNOWLEDGEMENTS

We are grateful for Chris Cummins’ careful reading of the chapter and his helpful suggestions and comments.

CHAPTER 18

METAPHOR