CHAPTER 19

METONYMY

PETRA B. SCHUMACHER

19.1 TYPES AND FUNCTIONS

IMAGINE that the contributors to this volume meet and I introduce myself by saying ‘I am metonymy’. What I mean is something along the lines of I am the person who has been invited to write the chapter on metonymy. So here an expression is used that does not explicitly name the intended person/concept but is merely related to it. This is an example of what rhetoric and linguistics call metonymy: a figure of speech or a cognitive operation that is used to refer to an entity by means of an expression that has some meaning relation to that entity. This relation between an expression and its intended meaning can be more closely defined as one of association or contiguity, in an attempt to distinguish metonymy from metaphor, which represents a relation of similarity. Cognitive linguistics further considers metonymy as an operation within a particular cognitive domain, while metaphor is subject to cross-domain mappings (Lakoff & Johnson, 1980b; Croft, 1993). The notion of domain has, however, not been delineated substantially.

In line with the idea of association or contiguity, metonymy is not arbitrary but relies on semantic and conceptual relations. Typical examples of metonymy are illustrated in (1) to (6), reflecting correspondence relations of the sort producer-for-product (1), place-for-event (2), container-for-content (3), institution-for-individual (4), or animal-for-statue (5) (for conceptual typologies, see Lakoff & Johnson, 1980b; or Norrick, 1981). Example (6) represents the classic case of Nunberg (1979) to illustrate the association between ham sandwich and ham sandwich orderer, which could be construed as a property-for-person relation.

(1) My grandmother read Dickens.

(2) My parents protested during Vietnam.

(3) Tim drank the goblet hastily.

(4) The newspaper called.

(5) Lisa dusted the wooden turtle.

(6) The ham sandwich is sitting at table 20.

The most widely studied type of metonymy is referential metonymy and the current chapter also focuses on this type of metonymy. Other types of meaning adjustment include predicate transfer, which refers to cases like I am parked out back in which the property of being parked out back is adjusted to applying to somebody whose car is parked out back (Stallard, 1993; Nunberg, 1995). Ward (2004) discusses the special function of equatives (as in my introductory example I am metonymy) and argues that the meaning adjustment is made explicit by the copula, yielding a more restricted pragmatic operation.

Whether these relations are derived from lexical semantics or depend on pragmatic reasoning has been a matter of debate, and we return to this in section 19.3 when we review the experimental findings. Whereas examples like (1)-(5) may more easily rely on correspondence rules or information in the lexical entries of the metonymically used expressions, (6) represents a looser use of the property-for-person metonymy that could evoke inferential reasoning about the speaker’s beliefs and intentions—and hence rely on Theory of Mind—as well as the situation of utterance (restaurant setting).

Turning to the pragmatic characterization, the central communicative function of referential metonymy has been related to economy considerations; that is, a speaker chooses a particular expression for reasons of brevity. This expression has to be in a particular relation to the intended referent (i.e. evoke a cognitive correspondence such as producer-for-product) or refer to a noteworthy and salient feature of the referent (as in the case of the property-for-person metonymy) (cf. Copestake & Briscoe, 1995; Nunberg, 1995). Metonymy has further been argued to reflect a tradeoff between brevity and clarity (Egg, 2004). In this regard it is also considered to be a referential shorthand; that is, it represents the easiest way to refer to someone—for instance, because the name of the person is not known or not relevant for the communicative goals of the speaker. Moreover, an additional effect may be achieved by associating the person with the property used. For instance, when referring to somebody with the expression the polka-dotted shirt, this can invite the hearer to draw inferences about the speaker’s attitude and might evoke derogatory connotations. In fact, effects of this sort arise from the use of metonymy in prose and poetry. In Fitzgerald’s The Great Gatsby, the narrator uses a place-for-person metonymy in (7) not only to refer to individuals but to allude to differences in social status (people born of wealthy families live in East Egg, newly rich people in West Egg) and attitude (snobby and corrupt East Egg vs. modest West Egg). Similarly, Gottfried Benn’s poem Night Café provides the reader with an impression of a bar scene and exploits property-for-person metonymy (8) to characterize the customers in the bar in a charmless manner and to reduce the musicians to their function in the bar context.

(7) Instead of rambling this party had preserved a dignified homogeneity, and assumed to itself the function of representing the staid nobility of the country-side—East Egg condescending to West Egg, and carefully on guard against its spectroscopic gayety. (Fitzgerald, The Great Gatsby, 1991: 37)

(8) The cello has a quick drink. The flute

belches throughout three beats: his tasty evening snack.

The drum reads on to the end of the thriller.

Green teeth, pimples on his face,

waves to conjunctivitis.

(excerpt from Benn/Sander, 1987, Night Café, 1912, translated from German by Michael Hamburger)

This chapter reviews experimental research that investigated how comprehenders process metonymically used expressions, what the underlying lexical representations may look like, how metonymy is processed in real time, and whether there is evidence for the application of correspondence rules or pragmatic reasoning. Research on the production of metonymy is relatively sparse. The chapter is organized as follows. Section 19.2 situates metonymy comprehension in the context of the processing of other entities that come with multiple meanings. Section 19.2.1 elaborates on core experimental findings from homonymy (section 19.2.1: one word, multiple meanings) and section 19.2.2 on polysemy (section 19.2.2: one word, multiple senses) before section 19.3 provides an overview of the experimental research on metonymy as a label for meaning extensions that apply to an expression with a core lexical meaning (multiple senses vs. extended senses). It will be shown that we need to distinguish between different types of meaning extensions that may represent discrete stages in meaning change and that these stages evoke distinct processing profiles (section 19.4). Research on metonymy in language development (section 19.5) and language disorder (section 19.6) will complement this discussion. Section 19.7 charts future directions for experimental research on metonymy.

19.2 MEANING ALTERNATIONS

19.2.1 One word, multiple meanings

Beginning in the 1970s, research on semantic processing has been concerned with words that express two or more distinct and unrelated meanings (so-called homonyms). An example is the expression bug, which can either refer to an instance of an insect (9), a covert listening device (10), or even a wild card in poker (11). It can also be used metaphorically, as illustrated by (12) in the context of computer defects.

(9) The bug had bitten hard.

(10) The bug was attached to the back of the TV set.

(11) The card used as the bug is a joker.

(12) The update incorporates a thousand bug fixes.

Psycholinguistic research has identified a number of factors that contribute to the processing of homonymy, such as frequency of occurrence of the individual meanings or contextual information, which may support and hence boost the retrieval of one meaning over another (Swinney, 1979; Rayner & Duffy, 1986; Duffy et al., 1988; Sereno et al., 2006). In the absence of a disambiguating context, the more frequent meaning is accessed faster. Context, however, is a strong disambiguation cue, which makes the contextually relevant meaning available faster. A critical research question has therefore been whether there is a dominant meaning that is automatically accessed first or whether it is the contextually relevant meaning that is accessed first. In a seminal study, Swinney (1979) investigated the processing of homonyms utilizing a cross-modal lexical decision priming paradigm. The task is cross-modal in that participants simultaneously engage in listening (auditory modality) and reading (visual modality). While listening to sentences for comprehension, participants performed a lexical decision to a visually presented letter string. Acoustically presented sentences included an ambiguous word (e.g. bug in (13)) and the words used for the lexical decision task (‘primes’) were related to one of the meanings (ANT, SPY) or entirely unrelated in the baseline condition (SEW). The spoken sentence was presented with each prime word separately and the visually presented prime word occurred once at a predetermined point in time, for instance, immediately after the ambiguous word (indicated by in the examples) or in a separate testing a few syllables downstream (indicated by ).

(13)a. Homonym, unbiased context:

Rumor had it that, for years, the government building had been plagued with problems. The man was not surprised when he found several bugs in the corner of his room.

b. Homonym, biasing context:

Rumor had it that, for years, the government building had been plagued with problems. The man was not surprised when he found several spiders, roaches, and other bugs in the corner of his room.

When the prime word was presented at the offset of the ambiguous word (indicated by above), lexical decisions for both related primes (SPY, ANT) were reliably faster than for unrelated primes in the unbiased context (13a). Interestingly, in the context supporting the insect reading (13b), both the contextually appropriate (ANT) and the inappropriate meaning (SPY) showed facilitation effects, yielding faster decision times relative to the unrelated prime. This indicates that lexical access of both meanings cannot be blocked by a strongly biasing context. Crucially, when the prime was presented three syllables after the homonym (indicated by above), only the contextually relevant meaning showed facilitation, implying that the unsupported meaning was no longer accessible. These important findings suggest that all meanings of a homonym are initially accessed during an early lexical processing stage, but quickly decay in activation as the parser selects the appropriate interpretation on the basis of contextual information. Similar results have been obtained for ambiguity based on distinct syntactic category membership (I bought the watch vs. I began to watch; Tanenhaus et al., 1979) and using other experimental measures (eye movements: Frazier & Rayner, 1990; Event-Related Potentials (ERPs): Hagoort & Brown, 1994).

These findings provide a first indication of the time-course of conceptual processing. They suggest that homonym processing involves two stages, lexical access and suppression or selection. The initial access of lexical meaning is widely considered to be automatic in nature and modulated by frequency of occurrence. The automaticity claim is supported by studies that utilize masked priming, a technique where the prime is ‘sandwiched’ between a forward and a backward pattern mask and is presented with a very short latency (~50 milliseconds (ms)), making it virtually undetectable for most participants, but nevertheless elicits robust priming effects (Forster & Davis, 1984; Holcomb & Grainger, 2006). However, there is still considerable disagreement as to whether the subsequent selection or suppression of meaning reflects conscious selection and attention that depends on task-specific strategies and expectations (cf. e.g. Neely, 1977) or automatic selection that is based on meaning competition (e.g. the interactive activation model in McClelland & Rumelhart, 1981; Rumelhart & McClelland, 1982).

19.2.2 One word, multiple senses

In contrast to expressions with unrelated meanings, the mental lexicon includes many expressions that can have multiple senses (i.e. different but semantically related interpretations; see also Rabagliati & Srinivasan, Chapter 22 in this volume). These words are called polysemes. An example of a polysemous expression is newspaper, which can for instance refer to the physical entity (14), the content (both the abstract idea and the physical text) (15), the building (16), the institution (17), or a person within the institution (18). There is a certain degree of systematicity between these different senses that—depending on the theoretical framework one adopts—can be captured by semantic rules like container-for-content, place-for-institution, or part-for-whole or can be derived from the qualia structure of a lexical entry (Pustejovsky, 1995a; Jackendoff, 1997).

(14) The newspaper was in shreds after the kitten played with it.

(15) The newspaper was boring and full of typographical errors.

(16) The newspaper burned down in 1892 and never rebuilt.

(17) The newspaper fired the editor.

(18) The newspaper called at noon and asked for an interview.

From a processing perspective, there are interesting differences between homonyms and polysemes that point towards different underlying representations. Polysemous words show processing advantages compared to homonyms and do not exhibit a processing disadvantage relative to words with fewer or just one sense and they also do not elicit frequency of occurrence effects for the different senses (Frazier & Rayner, 1990; Frisson & Pickering, 1999; Klepousniotou, 2002; Rodd et al., 2002; McElree et al., 2006a). These findings first and foremost indicate that polysemous words and unrelated ambiguous words are represented differently. The observations were taken to suggest that there is no immediate commitment to one of the senses of a polysemous word when it is first encountered. Instead, an underspecified representation is accessed that provides sufficient information for interpretation to proceed.¹

What do these findings tell us about how the different meanings and senses are represented in the mental lexicon? Two major approaches can be identified here. The most extreme scenario is the fully specified account, requiring the representation of multiple meanings and senses as separate lexical entries, which are connected with each other in the case of polysemy (Clark & Gerrig, 1983, 1986; Deane, 1988; Lehrer, 1990). Yet the experimental findings counter such a view and point towards an analysis that posits a less specified lexical representation. Such underspecification accounts describe a (polysemous) lexical entry as being not fully specified for all its interpretations (see, e.g., Frazier & Rayner, 1990; Copestake & Briscoe, 1995; Frisson & Pickering, 1999). Accordingly, lexical representations contain core properties of a concept, and the relevant sense is derived on the basis of context information and certain correspondence rules (Caramazza & Grober, 1976; Nunberg, 1979). The underspecified representation may be ‘good enough’ (Ferreira et al., 2002) unless context renders it necessary to select a specific sense (see also the idea of minimal commitment put forth by Frazier & Rayner, 1990). Alternatively, the lexical representation may be more elaborate, as for instance suggested in Pustejovsky’s (1995a) framework where lexical representations include rich information that guides interpretive processes including formal features (such as shape, constitution, and other distinguishing properties) as well as functional features (specifying the purpose and origin of an entity).

19.3 MULTIPLE SENSES VS. EXTENDED SENSES

A potential challenge for underspecification accounts is to determine the core sense of a polysemous word. It has therefore been suggested that it is possible to further divide what has so far been discussed as polysemy into metonymy, where a core or basic sense can be identified and another sense can be derived by a productive rule (e.g. count/mass extension for count nouns denoting animals to mass nouns denoting food, illustrated in (20)), and polysemy (in the narrow sense) where no basic sense can be determined (cf. sense extension vs. constructional polysemy in Copestake & Briscoe, 1995). An example of polysemy in the narrow sense has been provided earlier with newspaper with the competing readings for object vs. content (and possibly others). One diagnostic for this distinction is that polysemy allows co-predication and coordination (Cruse, 1986; Copestake & Briscoe, 1995)—the combination of two senses in one utterance as in (22)—but for metonymy this typically yields unacceptable utterances such as (21).² Other diagnostics for metonymy are blocking of pronominal coreference to the basic representation (23) or clashes in subject–verb agreement (24) (cf. Nunberg, 1995).

(19) Tim fed the chicken.

(20) There was chicken in the soup.

(21) #Tim fed and carved the chicken.

(22) The newspaper was in shreds and full of typographical errors.

(23) The French Fries left without paying. This happens regularly with her/*them.

(24) That/*Those French Fries is/*are getting impatient.

This distinction between metonymy and polysemy predicts that the former is more computationally demanding, because an interpretation must be derived from a core meaning, rendering the original meaning no longer accessible (as suggested by the linguistic diagnostics), while polysemy only requires access to an underspecified representation.

The first piece of evidence for such a processing distinction comes from count-mass alternations. Frisson & Frazier (2005) investigated metonymic processing involving ‘grinding’ (turning count nouns into mass nouns as in (20)) and ‘packaging’/‘portioning’ (turning mass nouns into count nouns). To examine the grinding operation, they compared underived and derived mass nouns (Melissa tasted inexpensive cola/clam) and found a processing disadvantage for the derived forms (inexpensive clam) reflected in later eye-tracking measures. Compared to narrow polysemy that engendered no costs (see references in section 19.2.2), this supported the idea that metonymic expressions come with an underlying representation that has to be adjusted from the count to the mass reading during grinding. For packaging they also observed enhanced processing demands (longer fixations on the critical word) for a derived than for an underived count reading (John bought imported pears/beers), also indicating that the count reading (imported beers) is arrived at from an underlying mass reading. The findings thus converged in revealing processing costs for both kinds of derived meanings and the time-course of the observed effects further suggests that grinding (with effects in later eye-tracking measures) is more demanding than packaging. This may be due to the extra conceptual demands required for arriving at a mass reading.

These data also inform theoretical claims about the directionality of the extension and the nature of the representation. It has been proposed on the basis of cross-linguistic observations that all languages have mass terms but that some languages do not have count terms (in particular classifier languages like Japanese or Chinese) (Quine 1968; Krifka, 1995b; Chierchia, 1998). Yet the findings of Frisson & Frazier (2005) indicate that the mass/count distinction does not operate on the basis of a core mass meaning. They reveal that some concepts have an underlying mass representation (e.g. liquids like beer), while others come with an underlying count representation (e.g. clam), with derived interpretations engendering costs in both directions.

Is this pattern found for other kinds of metonymic uses as well? Most of the research on the processing of extended senses has targeted relatively common instantiations like producer-for-product (My grandmother read Dickens), place-for-institution (That blasphemous woman had to answer to the convent), or place-for-event (A lot of Americans protested during Vietnam). In contrast to the mass-count alternations, eye movement measures revealed no processing differences between the core and the extended use of the word, supporting an underspecification account and accordingly a possible classification of these words as polysemes (cf. e.g. Frisson & Pickering, 1999, 2007). Electrophysiological patterns showed processing demands arising from the meaning extension of certain derivations (container-for-content: Tim drank the goblet hastily; animal-for-statue: the wooden turtle) but not for others (content-for-container: Ann put the orange juice in the fridge; Schumacher, 2013) (see section 19.4 for more details on these studies).

How can we account for these divergent patterns? One explanation may rest on the degree of typicality or conventionality of these alternations, where more typical alternations map onto a cognitive routine, while the others must be created ad hoc. Related to this is the question of how productive the alternations are. What happens, for instance, when a less common usage is encountered such as in (25)–(27) (cf. Nunberg, 1979, for detailed discussion)? The ham sandwich in (25) refers to a person contextually associated with the ham sandwich, for instance the sandwich orderer, uniforms stands for the wearers in (26), and scalpel for the surgeon (27). This sort of metonymy or reference transfer seems to require the creation of a new meaning to a much greater extent than in the examples earlier. The use of such expressions has therefore often been compared with metaphoric language use because the intended meaning goes beyond the core meaning, despite the fact that metonymic expressions typically show a closer relation between core and extended meanings (Lakoff, 1987; Gibbs, 1990).

(25) The ham sandwich is sitting at Table 20.

(26) There are 20,000 uniforms in this city.

(27) The scalpel was sued for malpractice.

These ad hoc concepts also pattern with more standard metonymy when considering the tests mentioned earlier. The following examples (28)–(30) illustrate that the less conventional examples of meaning extension pass the co-predication diagnostic for metonymy. This suggests that a correspondence rule may be available for these alternations as well, such as salient property-for-person.

(28) The ham sandwich is sitting at Table 20 (# and has been burnt).

(29) There are 20,000 uniforms in this city (# that are made of various fabrics).

(30) The scalpel was sued for malpractice (# and was made of high carbon steel).

Evidence for treating these property-for-person cases like other metonymies comes from electrophysiological data that revealed processing demands for this meaning adjustment compared to an object-denoting use of ham sandwich, which were similar to the processing demands observed for animal-for-statue and container-for-content alternations (Schumacher, 2011, 2013). We return to a more detailed characterization of the observed costs in Section 19.4.

First, however, let’s take another look at the argument that the conventionality of meaning alternations may account for different processing patterns. What happens to metonymies that utilize a conventional rule but nevertheless create a novel meaning? Schematic derivational rules may be distinguished according to the familiarity of a certain metonymic use. For instance, not every place name can be used to refer to an institution; convent is a more familiar place-for-institution metonymy than stadium, rendering the use of stadium in (31b) less felicitous (as evidenced by plausibility ratings; Frisson & Pickering, 1999). Eye-tracking measures revealed that typical uses (convent) were processed effortlessly, while unfamiliar metonymies (stadium) registered longer first-pass reading times and more regressions to the critical word (Frisson & Pickering, 1999).

(31)a. Core interpretation (place reading):

These two businessmen tried to purchase the convent/stadium at the end of last April.

b. Extended interpretation (institution reading):

That blasphemous woman had to answer to the convent/#stadium at the end of last March.

Another study investigated producer-for-product metonymies and also revealed that familiar alternations were processed at no further expense relative to control sentences (My grandmother often read/met Dickens), but that novel metonymies (My grandmother often read Needham) engendered processing difficulties unless prior context identified Needham as an author (e.g. My grandmother has all the novels written by Needham in her library; cf. Frisson & Pickering, 2007). This effect of context suggests that comprehenders are able to access a transfer operation like producer-for-product even when a previously unknown name is encountered, if that name is associated with properties activating the correspondence rule.

19.4 TOWARDS AN ACCOUNT OF MEANING CONSTITUTION

Looking at three different types of ambiguous words in the lexicon—homonyms, polysemes, metonyms—it is apparent that the different senses and meanings are represented and accessed in distinct ways. At the same time, lexical composition is very creative and driven by a productive machinery, allowing transfer of meaning in new and innovative ways (as suggested by the ham sandwich and Needham cases earlier; see also Davies & Arnold, Chapter 28 in this volume, on conceptual pacts between interlocutors and speaker goals). The findings indicate that the different meanings of homonyms map onto distinct lexical representations, from which the parser selects one at the expense of computational demands, while polysemes appear to hold an underspecified representation that allows switching between senses for free. Metonymies—representing words for which a core sense can be identified—show an interesting variation in their processing profiles. Some metonymies pattern with polysemes in terms of the underlying processes and engender no particular costs, while others consume processing resources indicating that they rely on derivational rules and/or pragmatic operations. In the following, I want to propose an account for these different patterns on the basis of electrophysiological data and provide a historical motivation for the different operations.

What are electrophysiological measures? The electrical activity of neurons exerted during language processing can be recorded in a non-invasive manner via electrodes placed on the surface of the scalp. The electroencephalogram shows changes in the neural activity of the brain that are evoked by cognitive, sensory, or motor events. ERPs represent neural activity time-locked to an event (e.g. the onset of an ambiguous word) and are relative measures between a critical condition and a minimally differing baseline condition. ERPs are favoured for their high temporal resolution and they can further be characterized along different dimensions (latency, amplitudinal magnitude, polarity, scalp topography) that allow for a dissociation of the underlying processes.

In ERP research, a negative deflection peaking around 400 ms after a critical word (socalled N400) is a well-established signature of lexical-semantic processing. The N400 reflects the detection of semantic anomaly (He spread the warm bread with socks from Kutas & Hillyard, 1980b), as well as unexpected continuations (Van Berkum, Zwitserlood, Hagoort, & Brown, 2003; also in arithmetic and visual scenes: Niedeggen et al., 1999; Sitnikova et al., 2003). It also reflects priming effects, category membership, and semantic relations (e.g. Bentin et al., 1985; Rugg, 1985; Hagoort & Brown, 1994; Federmeier & Kutas, 1999;). The N400 shows repercussions of lexical access, mirroring differences in frequency, length, or lexical neighbourhood density (Van Petten & Kutas, 1990; Holcomb et al., 2002; Osterhout et al., 2002) and it can be understood as an implicit prediction error in semantic representation (cf. Rabovsky & McRae, 2014). In studies of lexical ambiguity, homonyms (compared to unambiguous words) registered a more pronounced negative deflection, and disambiguation towards the less frequent meaning also engendered an N400 (Hagoort & Brown, 1994). Comparisons of homonyms and polysemous words furthermore support the claim that homonymous meanings are stored separately in the mental lexicon, while polysemous senses are underspecified, as reflected by the absence of N400 effects for separate senses (Taler et al., 2009; Klepousniotou et al., 2012).

ERP research can also shed light onto the different processing profiles observed for metonymy. Remember that certain derivations engender processing costs, while others do not. What is going on there? Let’s start with the property-for-person cases. An ERP study registered processing demands for the metonymy relative to a control condition that required the core meaning but this contrast was not reflected by an N400 but by a positive-going ERP signature with an onset latency around 600ms (Late Positivity) (Schumacher, 2011). When the context provided a proper scenario (e.g. medical practice), the metonymic expression the hepatitis (32a) registered a more pronounced late positive deflection relative to a literal control condition (32b).³

(32)a. Metonymy (with facilitating context):

The doctor asks his assistant who had called that early. The assistant responds that the hepatitis had called that early.

b. Core meaning (with facilitating context):

The doctor asks his assistant what it is that concerns so many people. The assistant responds that the hepatitis concerns so many people.

This Late Positivity was also reported for animal-for-statue (the wooden turtle) and container-for-content alternations (Tim drank the goblet hastily) relative to conditions with no meaning extension (Schumacher, 2013). By contrast, producer-for-product (The student read Dickens) and content-for-container alternations (Ann put the orange juice in the fridge) registered no positive deflection (Schumacher, 2013; Weiland-Breckle & Schumacher, 2017).

So why is there an ERP effect for some extended senses but not others? And why does the difference emerge as a Late Positivity and not an N400? I suggest that the Late Positivity effect reflects processing demands associated with reconceptualization in cases where the lexical representation does not provide immediate access to the intended interpretation. The comprehender shifts the meaning at the expense of the core representation being no longer accessible. This is evidenced by the coordination and co-predication tests mentioned in section 19.3—which fail in those cases where a Late Positivity is observable. The Late Positivity hence mirrors an operation that results in a new conceptual representation. The extra processing demands are thus an immediate reflection of the meaning extension and the updating of the representation during the referential shift (see Schumacher, 2013, for further details on discourse representational updating).

The difference between the two metonymic profiles is thus that one alternation type leads to reference shift (reflected in Late Positivity) while the other relies on sense selection within the lexical representation (no observable ERP effect). Sense selection can take place without recruiting additional processing resources but meaning selection during homonym comprehension requires extra processing costs (a pronounced N400). Hence ERPs can provide insights into discrete underlying processes and representational differences. Figure 19.1 provides a schematic illustration of the observed ERP patterns for words with multiple meanings (homonyms) (left panel), words with multiple senses (polysemes in the narrow sense) (middle panel), and words with extended senses (metonyms) (right panel).

A classification on the basis of neurocognitive profiles may be backed up by the linguistic diagnostics illustrated earlier. Co-predication, coordination, and pronominalization tests have been proposed as a first approach to distinguishing the different types of alternation (although see Cruse, 1986, Copestake & Briscoe, 1995, Nunberg, 1995, Frisson & Frazier, 2005, Schumacher, 2013, for discussion of potential caveats). In the reference shift cases these tests should fail as an indication that the original meaning is no longer accessible; in the other cases, lexical representation provides equal access to the two senses and hence co-predication, coordination, and pronominalization are possible.

FIGURE 19.1. Schematic illustration of expected ERP effects. Left panel: words with multiple meanings (homonyms, solid line) evoke an N400 vs. words with a single meaning (dotted line). Middle panel: words with multiple senses (polysemy in the narrow sense, solid line) vs. words with a single meaning (dotted line) show no ERP effect. Right panel: words with derived senses that require reconceptualization (metonymy, solid line) engender a Late Positivity vs. words with non-derived sense (dotted line)

In addition, the different types of meaning alternation may be rooted in historical processes. They may reflect discrete stages in the diachronic development of meaning where communicative needs result in the creative application of meaning alternation (cf. Papafragou, 1996b). At the earliest stage of this development, certain meanings are created ad hoc and in an innovative manner; at the next stage, a word is used in a particular discourse context and by a narrowly delineated group of interlocutors (e.g. the situation-specific use of hepatitis to refer to a patient or ham sandwich to refer to a restaurant customer); then a derivation undergoes generalization; and at the final stage, a word is used by the entire speech community and has a coded meaning in lexical representation (cf. Koch, 2001; Traugott & Dasher, 2002; Panther & Thornburg, 2003).

It is a matter of debate in historical pragmatics whether these stages represent steps in the life of a metonym and the genesis of polysemy (cf. Hansen & Waltereit, 2009, for a different position). For present purposes it suffices to treat these different levels as descriptive categories to characterize discrete stages of usage of an expression and the degree of its immersion within a language community. This allows us to view metonymy as a continuum. Along these lines, notions from historical pragmatics may be suitable for explaining patterns of language processing to the extent that they tap into different levels of representation. The distinct comprehension profiles may thus reflect the degree of dispersion of a word and its use, separating ad hoc meanings and uses within a particular speech community and a certain cultural or professional context (such as the property-for-person cases) on the one hand from routinized uses of the entire speech community (such as producer-for-product alternations) on the other hand.

19.5 METONYMY AND LANGUAGE ACQUISITION

There is evidence of the use of metonymy in children’s production from the age of 3. Falkum and colleagues (2017) carried out two production studies and showed that 3-year-old children already produce metonymic expressions in certain contexts. Nerlich et al. (1999) analysed instances of creative language use by a child, which they explain in terms of ‘creative metonymical shrinking’. They argue that children exploit conceptual associations and use metonymic expressions for novel thoughts while at the same time keeping articulatory effort to a minimum. They report, for instance, that the child uttered I love being a sandwich to express the idea that ‘he liked to be part of the children who were allowed to bring a lunch-box (i.e. a sandwich) to school and were not forced to have [school dinners]’ (from Nerlich et al., 1999: 370). The mini-corpus contains other creative uses of language, including the creation of new predicates, which suggest that the child is exploiting relational information for play and possibly due to economy considerations. At the same time, the same child (at the age of 4 years) shows a certain degree of meta-linguistic awareness in language comprehension as the following anecdotal story suggests:

(33) Matthew came home from school. I said: ‘Wow, you have eaten your whole lunch box!’ Matthew burst out laughing and said: ‘Oh Mummy, you got that wrong—it’s: you have eaten everything out of your lunch box!’ (from Nerlich et al., 1999: 375)

Further comprehension studies have indicated that 4- and 5-year-olds are beginning to understand expressions used metonymically. Nerlich et al. (1999) demonstrated this with a picture selection task in which children had to choose between a picture depicting the object or property denotation and one depicting the metonymic interpretation of a story in a picture selection task (Nerlich et al., 1999). Comprehension data from Falkum and colleagues (2017) even show that 3-year-olds outperform 4- and 5-year-old children, while meta-linguistic awareness and the ability to reflect upon the metonymic expression increases with age. In a different cross-sectional study involving children and adults, participants were presented with stories that ended with a metonymic expression (e.g. Kate is looking for a CD and utters I found Robbie Williams in the lounge) and were then asked for their interpretation of this utterance. This study showed increasing performance on metonymy with age and better comprehension of metonymy overall compared to metaphor (Rundblad & Annaz, 2010a). These developmental data suggest that metonymy comprehension improves with the acquisition of semantic relations and conceptual correspondences. Children further seem to be increasingly capable of grasping the communicative intention associated with metonymy and perceiving it as a way of referring or naming (cf. also children’s name-calling behaviour in Crozier & Dimmock, 1999).

19.6 METONYMY AND LANGUAGE DEFICITS

Data from impaired language competencies shed further light on the semantic and pragmatic underpinnings of metonymy processing. Theory of Mind capacity, that is, the ability to attribute mental states, intentions, and so on to others, has been claimed to be important for the development of pragmatic competences (cf. Rubio-Fernández, Chapter 31 in this volume). Accordingly, metonymy comprehension relies on awareness of the referential intentions of the speaker. Two groups with developmental disorders that have been studied in this regard are children with Williams syndrome and those with Autism Spectrum Disorder (ASD).

People with Williams syndrome, a rare genetic disorder, have relatively good language skills but show pragmatic difficulties, for instance in conversational interactions, narrative abilities, or detection of lies (Brock, 2007). (Conventional) metonymy comprehension in this group has been shown to be delayed relative to chronological age but appropriate for their mental age; in contrast, performance on metaphor has been poorer (Annaz et al., 2009; Van Herwegen et al., 2013). The child’s level of semantic knowledge emerged as a reliable predictor for good performance in metonymy (but not metaphor). This indicates again that metonymy is grounded more strongly in semantic relations.

ASD is a neuro-developmental disorder that is characterized by deficits in social interaction and pragmatic skills. People with ASD typically have difficulties inferring the intentions of their interlocutors and exhibit impairments to their social cognition including non-verbal communication (for an overview, see e.g. Kim et al., 2014). Comprehension of figures of speech is typically impaired in this population. Using a story picture task as described earlier (Robbie Williams referring to a CD), Rundblad & Annaz (2010b) tested children with Autism (aged 5-11) and found better performance on metonymy than metaphor; metonymy comprehension further correlated with verbal mental age (i.e. receptive vocabulary scores) but not with first-order Theory of Mind ability (i.e. the Sally Ann task; Baron-Cohen et al., 1985). Another study that compared different types of figurative language, including metonymy, probed ASD children’s comprehension of informative and communicative intention and found that it was easier for these children to comprehend the meaning of a metonymic utterance than to understand the speaker’s intention (MacKay & Shaw, 2004). While some studies thus support the idea that Theory of Mind is central to ASD, others locate the processing difficulties with metonymy primarily at the semantic level.

In recent years, there have been attempts to correlate individual processing differences in typically developing participants with measures such as the Autism-spectrum Quotient (AQ; Baron-Cohen et al., 2001), which assesses differences in social and communicative skills. Testing metonymy as exemplified in (31) (purchase the convent / the stadium vs. answer to the convent / #the stadium) in a self-paced reading study, McKenna and colleagues (2015) showed that participants with low scores for social and communicative skills were slower to read metonymic expressions, and even slower for novel metonymic uses. Such findings of gradient differences in metonymy processing as a function of communicative skills call for more fine-grained investigations of pragmatic deficits in language processing, to find out whether they reflect a general lack of certain competencies or rather a resource limitation that affects the underlying processes.

Finally, research with people with neuro-degenerative diseases and acquired language deficits further strengthen the dissociation between (conventional) metonymy and metaphor and support the involvement of semantic networks in metonymy processing. For instance, semantic abilities are impaired in people with Alzheimer’s disease and this population does not show the typical real-time processing advantage for metonymy in priming studies relative to metaphor or homonymy (cf. Taler et al., 2009). People with right hemisphere lesions who typically show deficits in pragmatic reasoning also do not register processing deficits for metonymy (Klepousniotou & Baum, 2005). Together with neuroimaging studies (Rapp et al., 2011), these findings strengthen the semantic underpinnings of metonymy; but notice that patient studies have assessed the comprehension of conventionalized metonymies and have not yet made the distinction laid out in section 19.4.

19.7 FUTURE DIRECTIONS

Metonymy has been investigated with respect to its implications for both semantic and pragmatic processing. A clear semantics-pragmatics distinction appears to be challenging to support since metonymy is an instance of language use and reflects the speaker’s intentions; yet at the same time, it relies on well-defined semantic and conceptual relations. A fruitful endeavour in this regard would be to continue investigations into the time-course of metonymy processing in order to obtain a more fine-grained profile of the underlying mechanisms.

Future research should further aim at a better understanding of metonymy as a continuum both in terms of a finer-grained classification of types of metonymy (possibly as a reflection of different stages in semantic development as suggested in section 19.4) and from a comparative perspective with respect to other types of figurative language use. Experimental research has often contrasted metonymy with metaphor (and to a lesser extent other types of figurative language use) and found pronounced differences between these two types of language use (cf. e.g. Klepousniotou, 2002; Rundblad & Annaz, 2010a,b; Bambini et al., 2013; Van Herwegen et al., 2013). This approach should be extended to other phenomena in order to develop a typology of figurative language use and the underlying cognitive operations.

¹ Note however that this view has been challenged by findings from inductive reasoning and categorization tasks that suggested that senses are represented separately and have little semantic overlap (Klein & Murphy, 2001, 2002)—but see Frisson (2009) for the claim that these differences might be due to distinct tasks tapping different processing stages, lexical access vs. selection respectively.

² Some combinations of newspaper appear to thwart this diagnostic. However, examples like (i) may also indicate that certain expressions may come with a polysemous and a metonymic extension. In this case, the institution reading would serve as a metonymy that no longer allows combination with the content sense.

(i) The newspaper fired the editor (# and was full of typographical errors).

³ The Late Positivity was also observed when the context in (32) was deprived of a supporting scenario. An additional N400 in this situation indicates that, in the absence of supporting context, lexical access is encumbered (cf. Schumacher, 2014).