Literal and metaphorical meaning: in search of a lost distinction

ABSTRACT

The distinction between literal and figurative use is well-known and embedded in ‘folk linguistics’. According to folk linguistics, figurative uses deviate from literal ones. But recent work on lexical modulation and polysemy shows that meaning deviation is ubiquitous, even in cases of literal use. Hence, it has been argued, the literal/figurative distinction has no value for theorising about communication. In this paper, we focus on metaphor and argue that here the literal–figurative distinction has theoretical importance. The distinction between literal and metaphorical needs to be captured by our account of communication because literal uses transmit information in a way that metaphorical ones do not. We argue that there is a way to explain the literal/metaphorical distinction that preserves the core of the folk-linguistic idea and gives the distinction theoretical relevance. We propose that literal uses of a word are made with the intention to conform to an established practice of use, while metaphorical uses do not so conform, but depend on this pre-existing practice. Our account can deal with data that are problematic for other theories. A further advantage is that it extends naturally to other non-literal uses of words, including metonymy.

KEYWORDS:

• Metaphor
• non-literal use
• lexical pragmatics

1. Introduction: the literal/non-literal distinction in the firing line

We are generally good at distinguishing between literal and non-literal meaning. In ‘The brightest object visible from earth is the sun’, ‘the sun’ is used literally; in ‘Juliet is the sun’ it is not. Recanati (2004, 68) submits that we have a folk-theoretic notion of the literal/non-literal distinction according to which non-literal meaning

involves a form of deviance or departure from the norm; a form of deviance or departure which must be transparent to the language users. (Recanati 2004, 81)

The non-literal is a deviation from the literal and the latter is in accordance with the norms of correct use.

The deviation understanding of the distinction between the literal and the non-literal needs support because it seems to rely on a conception of meaning that has come under pressure in the last twenty years. This conception assumes that literal meaning pertains to the expression-type and is determined prior to and independently of contexts of use.¹ Non-literal meanings of a word w deviate from this standing meaning and are not fixed in the same way.² If we already have in place a notion of literal meaning for a sentence type, it is tempting to say that an utterance of a sentence s is literally true if the utterance expresses the literal meaning of the sentence and things are as the literal meaning of the sentence represents them. When does an assertoric utterance of a declarative sentence express its literal meaning? If a speaker utters the sentence intending to express its literal meaning.³ In this case, the question of literal truth/falsity arises for the utterance. Otherwise, the utterance is not literally true. It may be metaphorically or approximately true etc.

Now, many philosophers of language and linguists have given good arguments for the conclusion that deviation from the meaning of an expression-type, if there is such a thing, happens more often than not. To put it somewhat paradoxically: deviation from the norm is the normal case, not the exception. The word ‘silent’ means ‘not making or accompanied by any sound’ (Oxford Dictionary of English), but a speaker who says ‘It’s silent here’ may mean to exclude only sounds that are clearly audible, not barely audible sounds like the humming of a refrigerator (Bezuidenhout 2001, 168). When you say ‘John cut the cake’, the meaning of ‘cut’ in your utterance is not the meaning, if any, of ‘cut’ independently of a particular utterance. The meaning of ‘cut’, if there is one, is so to say ‘fine-tuned’ to mean cutting appropriate to cake and the particular circumstances (Searle 1980). However, while the meaning ‘cut’ has in this particular utterance is distinct from the meaning of the word type ‘cut’ it is certainly not non-literal.

On the basis of such examples, relevance theorists claim that the ‘interpretation of almost every word’ is ‘fine-tuned’ (Wilson and Carston 2007, 231). Recanati (2004, 75) goes further: ‘processes like enrichment or loosening are universal: there is no utterance, however explicit, whose interpretation does not involve adjusting the conventional meanings of words to the particulars of the situation talked about’.

Such views are supported by work on recent polysemy. Some may argue that there is no occasion-specific meaning in Searle’s example, that ‘cut’ simply has its literal standing meaning, even though a hearer of the utterance typically forms a more specific view of the manner of cutting.⁴ Such scepticism is harder to maintain in the face of examples like ‘John’s new book is heavy’ and ‘John’s new book is available on Kindle’, where ‘book’ has a concrete sense (book object, as it were) in the former, but in the latter means something like book content.⁵ These different senses expressed in uses of polysemous words are components of the content of the speaker’s assertions (for evidence see Collins 2017, 679).

There is evidence from psycholinguistics that the stable, linguistically encoded meanings of polysemous words are neutral between word token meanings and not equal to any of them.⁶ Experiments show that homonyms (e.g. ‘bat’ [rodent] and ‘bat’ [sports]) and polysemes (e.g. different senses of the noun ‘book’) are processed differently. A key contrast is that in processing polysemes, there’s no bias towards the most frequent or otherwise dominant sense (Frisson 2009), while it has long been known that the dominant meaning of a homonym is retrieved faster and more easily (Frisson 2009, 113–114). Further, uses of polysemous words with different senses prime each other, while uses of the different meanings of homonyms inhibit each other. These findings suggest that unlike homonyms, each polysemous word has a single stored mental representation that is neutral between its senses and not equal to any of them (Frisson and Pickering 1999; Frisson 2009; 2015): the meaning of ‘cut’ in a competent speaker’s mental lexicon is neutral between the senses that can be expressed when a speaker says ‘cut grass’, ‘cut the cake’, cut hair’ etc., and similarly for the abstract and concrete senses of ‘book’. On this view, all uses of polysemous words – the ‘vast majority’ (Frisson 2009, 113) of content words⁷ – are ‘deviant’ in the sense discussed above: the word token meaning is never identical to the word type meaning.⁸

If meaning deviation is ubiquitous, it seems that one can either (a) hold on to the distinction between literal/non-literal meaning, but disentangle it from the deviation explanation of it or (b) hold on to the deviation model of literalness and take the notion of literal meaning to be useless in a theory of language use. On this second view, what a speaker explicitly communicates by means of an utterance rarely coincides with the standing meaning of the sentence-type even if the sentence-type has such a meaning, which is doubtful in the first place, and thus,

the distinction between literal and metaphorical language is rarely, if ever, reflected in a qualitative change in the psychological processes involved in the processing of that language. (Rumelhart 1993, 72)⁹

Similarly, Wilson and Sperber (2002, 622f) argue that the notions of literal meaning and literal use do no work in a theory that explains communication. Literal meaning and literal use are common sense notions or part of folk linguistics, but they are not explanatory notions of a theory that explains the mechanism of communication (ibid: 626). They and other relevance theorists have developed an account of lexical pragmatics which treats literal and loose use, hyperbole and metaphor as ‘arrived at in exactly the same way[: … ] there is no mechanism specific to metaphors, and no interesting generalisation that applies only to them’ (Sperber and Wilson 2008, 84).

In this paper we will focus on a representative kind of non-literal use or meaning, namely metaphor, and argue that the literal–metaphorical distinction has explanatory weight in a theory of communication. Why? Because communication is often intended to transmit knowledge about a subject matter. Metaphorical utterances cannot contribute to this process since they are not about their vehicle (i.e. the term used metaphorically) as we explain below. We, therefore, need to respect and clarify the literal/metaphorical distinction to understand how communication as sharing of knowledge is effected.

2. Lexical pragmatics and the literal/non-literal distinction

Let us start by looking in more detail at an influential view of lexical pragmatics that is in tension with the literal/non-literal distinction. Sperber and Wilson assume that monomorphemic words encode atomic concepts that in turn provide access to encyclopaedic information about the objects in their extension (1986, 87, 91–92). According to the account of lexical pragmatics they and Robyn Carston have developed, in interpretation hearers construct senses by selecting encyclopaedic properties (Carston 1997; Sperber and Wilson 1998; Wilson and Carston 2007).¹⁰ These are selected in order of accessibility (which is affected by context) and the candidate sense (‘ad hoc concept’) depends on which are selected. The account claims that ad hoc concepts have extensions that differ from the lexicalised concept: they are broader or narrower. This process occurs in mutual adjustment with other pragmatic processes including disambiguation, assignment of referents to indexicals and generation of implicatures. The result is a candidate interpretation of the utterance, which is accepted by the hearer’s pragmatic system as the speaker-intended one if it satisfies expectations of relevance in the context. (For helpful examples, see Sperber and Wilson 2002, 19–20; Carston and Powell 2006, 345–346.)

According to this account, the generation of ad hoc concepts is operative in almost all utterances (Carston and Powell 2006, 345; Wilson and Carston 2007, 231). Consider typical utterances of the following three sentences:

1. The surgeon opened John’s mouth. (His jaw was broken and had been wired together) (Carston and Powell 2006, 344)

2. Paul Pogba was born in Paris. (He was in fact born in a town well outside of the municipal boundary of Paris which is nowadays regarded as a suburb of Paris)

3. Paul Pogba is no saint.

If there is a standard context-independent meaning of the underlined words, it is not the meaning put forth in the utterances we are concerned with. But, according to common sense, it is only (3) that is a clear case of figurative speech. Similarly, as we have seen ‘cut’ seems to have different – but equally literal – senses in typical utterances of ‘cut hair’, ‘cut cake’, ‘cut grass’ and so on (Searle 1980, 222–223; cf. Searle 1983, 145–146 on ‘open’).

One conclusion one might draw from these observations is that, given the omnipresence of deviant occasion-sensitive meaning, distinguishing between literal and figurative meaning is pointless. If everything is a deviation, nothing is: the distinction between deviation and norm loses its point and purpose and can no longer ground the literal/non-literal distinction. Hence, we should give up the latter distinction entirely. Or so a thorough-going revisionist might argue.

The claim that we oppose in this paper is, however, the more conservative one that a theory of communication should make no distinction between literal and metaphorical uses.¹¹ More specifically, the relevance-theoretic view is that the same method of arriving at the occasion sensitive meaning is operative in arriving at the occasion sensitive meaning of utterances of (1) to (3):

[F]ully unified accounts reject the traditional distinction between literal and figurative meaning and claim that approximation, hyperbole and metaphor are not distinct natural kinds, requiring different interpretive mechanisms, but involve exactly the same interpretive processes as are used for ordinary, literal utterances. (Wilson and Carston 2007, 231)

Similar conclusions have been widely reached in experimental pragmatics. The key finding here is that, ‘with enough background information, a novel metaphoric phrase can be read as routinely as a non-metaphoric one’ (Noveck 2018, 159). This is evidence against the literal-first hypothesis: that in processing a metaphorical utterance the addressee must first entertain a literal interpretation, then reject it and construct the correct metaphorical interpretation. It also suggests a more general conclusion: metaphorical utterances are processed the same way as literal ones. As Noveck says, reviewing the experimental literature:

It would soon become commonplace to find articles whose introductions resembled the following: ‘Revered scholars have thought that there is something special, unique, or exceptional about metaphor and now we know that this is not the case.’ (Noveck 2018, 160)¹²

Likewise in lexical semantics, as witness the scare quotes in the title of a recent paper, ‘Figurative’ uses of verbs and grammar (McNally and Spalek 2019).¹³ This paper (which we return to below) treats the literal/figurative distinction as intuitively obvious in some cases, contrasting e.g. ‘cut the cake’ and ‘cut the deficit’, but does not attempt to provide any theoretical basis for this distinction. Rather, it argues that these ‘literal’ and ‘metaphorical’ uses are similar, as we shall see in section 4.

3. Why the literal–metaphorical distinction is needed in the theory of communication

There are many areas of life where the distinction between literal and figurative is important. Let us consider the law as a representative example. Drafters of statutes and contracts aim to avoid non-literal expressions, except highly conventionalised ones, with the aim of avoiding unclarity (Tiersma 2001, 454; Lewison 2015, para 15).

If you want to encode my rights, you should avoid figurative expressions because the formulation ought to be reliably and easily interpreted as the law maker intended it to be interpreted. Here we have a subject-matter-specific norm of communication. Wilson and Sperber (2002, 626) acknowledge other less specific ‘folk-theoretic norms’ that they take to be the source of the notions of literal meaning and what is said. They don’t deny the importance of the literal/metaphorical distinction in relation to norms of communication; what Wilson and Sperber (and Rumelhart) deny is that literal meaning and literalness play an explanatory role in the theory of communication.

However, even an account of the mechanics of communication needs to take into account literal use and the literal/metaphorical distinction. Consider an example. You are interested in finding out what kind of thing the mind is. A cognitive scientist who has gathered evidence and has formed a theory tells you:

(4) The mind is a computer.

You are surprised. Minds are not artefacts and don’t seem to be programmable, have a central processing unit etc. It’s reasonable to ask: Is the mind literally a computer? If the answer is ‘No’, the cognitive scientist has no theory of the mind yet. Why not? If it is only metaphorically true that the mind is a computer, we have only (at best) started saying what the mind is. Roughly speaking, the cognitive scientist draws our attention to something that is in certain respects analogous to the mind. But drawing our attention to the right kind of model for the development of a theory is not having developed and proposed one. It is a starting point, not the end point. Whether an utterance is put forth as literally or metaphorically (more broadly: figuratively) true matters for assessing the status of a proposal: is it a heuristic for finding an explanation or is it the explanation we seek? Here we find ourselves in agreement with Colyvan (2010, 299): a metaphor is only a proxy for an explanation that can be given without it. This thought about metaphor nicely explains some debates in science and philosophy. For example, Chalmers and Clark argue that ‘My notebook is an extension/part of the machinery of my mind’ is ‘quite literally’ true (Chalmers 2008, xxvi). Fodor (2007) denies that it is. If my notebook is only figuratively part of the machinery of my mind, the statement has less value for cognitive science:

If minds don’t literally have parts, how can cognitive science literally endorse the claim that they do? That Juliet is the sun is, perhaps, figuratively true; but since it is only figuratively true, it’s of no astronomical interest.

Fodor’s example is unfair: that Juliet is metaphorically the sun is indeed of little astronomical interest. But that minds metaphorically speaking have parts, among them, notebooks, could be of interest to cognitive science. Still, the interest is only an interest in a ‘line of thought’ or heuristic, not an explanation that can be confirmed or refuted.¹⁴

The main point here is that only a literal use of ‘sun’ can be a link in a chain of transmission of knowledge about the subject matter, namely the body at centre of the solar system: a metaphorical use can’t.¹⁵ This is in line with Searle’s (1993, 92) observation that a metaphorical assertion such as ‘Richard is a gorilla’ is only about Richard, an unpleasant guy. It is not about gorillas: how gorillas actually are – shy and friendly – does not matter for the truth or falsity of the metaphorical assertion. Similarly, while ‘The brightest object visible from earth is the sun’ is about the sun, ‘Juliet is the sun’ is not. If ‘computer’ is used literally in (4), we have something more concrete than a mere pointer in the direction of a theoretical identification, namely a statement that the mind is just that: a computer. The theoretical identification has been stated and now needs to be elaborated.

Prima facie, the observation that figurative uses block the transmission of information does not concern folk-theoretic norms of discourse. Whatever norms the speaker and hearer seek to observe in their interaction, accepting the words the speaker used metaphorically can’t extend the hearer’s knowledge about the sun. This seems to be a non-normative fact about communication which needs to be explained by a theory of communication. What is it about the mechanics of communication that limits knowledge transfer in this way? In section 6, we will offer an account that respects the folk-theoretic intuition that figurative uses are derivative and shows how this difference is theoretically relevant.

4. Recanati’s response: felt discrepancy

In his Literal Meaning Recanati sets out to show that ‘the ordinary, folk-theoretic notion of non-literal use can be rescued’ (2004, 68). Recanati (2004, 74–75) agrees with Sperber and Wilson that lexical modulation is universal and essential for communication. Yet, the folk-theoretic notion of literalness is supposed to help to explain communication. Let’s see how this is supposed to work.

Recanati claims that non-literalness is ‘transparent’ to language users (2004, 75). In a non-literal use of an expression there is a felt discrepancy between conventional meaning and the utterance meaning. Consider someone saying:

(5) The ATM swallowed my credit card, chewed it up and spat it out. (Recanati 2004, 77 fn 11)

Here someone who understands the utterance fully is supposed to feel a ‘tension’ between the standard meaning of ‘ATM’ and the predicates applied to its referent: ‘In the scene the ATM is both a machine and a human being – hence the feeling of discrepancy. The discrepancy […] is a tension between two aspects or components of the primary interpretation’ (2004, 79–80).

In contrast, when I say ‘Paul Pogba is from Paris’ there is supposed to be no felt tension. Hence, felt tension is supposed to make for figurative use.

The feeling of discrepancy marks out cases of non-literalness: the speaker feels the discrepancy and the feeling initiates a process of accommodation (ibid.). Hence, a theory of communication ought to incorporate the literal/non-literal distinction. It needs something that corresponds to or explains the felt discrepancy in the non-literal case, Recanati suggests.

So far, so good, perhaps.¹⁶ But felt discrepancy between meanings is a broader notion than non-literal use. Take a conceptual falsehood – an utterance of (6):

(6) The square Meinong thought of is round.

Squares are not round and if you possess the concepts of square and roundness you cannot fail to know this. Hence, you will feel that just as ATMs cannot swallow credit cards, squares cannot be round. There is a felt discrepancy. But there is no word in the utterance of (6) which is used non-literally. I literally and strictly speaking say something which is false (Meinongians will of course disagree). Similarly, every word in an utterance of ‘Prime numbers are green’ is used literally, but the utterance comes with the feeling of discrepancy. A feeling of discrepancy is not sufficient for non-literal use, then.

It is also not necessary. There are metaphorical utterances where there is – or need be – no felt discrepancy. Consider again an utterance of (3). It is a clear case of non-literal truth. But there is no felt discrepancy because it is denied that Pogba, the French footballer, is a saint. There are also positive metaphorical utterances (i.e. not denials) with no feeling of discrepancy.¹⁷ Recall the generalisation from experimental work on metaphor processing: where there is a good fit with context, metaphors can be understood as rapidly and easily as literal utterances. Carston (2010) notes that metaphors can differ in a number of respects, including how creative they are, whether they evoke imagery, whether they are spontaneous or highly wrought and literary, and whether they are ‘extended’. She suggests that there are two different modes of processing: ‘a quick, local, on-line meaning-adjustment process and a slower, more global appraisal of the literal meaning of the whole’ (297). Whether or not that proposal is right, it is highly plausible that in typical processing of a metaphor that is easy in context, not image-evoking, and not part of an sequence of related metaphors, there is no special phenomenology. Recanati could respond by saying that felt discrepancy may not always be present, but can be brought to mind whenever there is a metaphor, just as grammaticality intuitions often require some introspection.¹⁸ However, this claim lacks any empirical support, as far as we know, and of course would not help with the sufficiency problem, namely that there are literal utterances that have felt discrepancy.

If Recanati is right, there may be a distinction here that should be incorporated into a theory of communication, but it is not the distinction between literal and non-literal use that folk linguistics posits.

5. The literal/non-literal distinction and feature suppression

In linguistics it is generally assumed that words (or more technically, lexical items) have (or are composed of) linguistic features (e.g. Chomsky 1995 sees lexical items as bundles of features). Some of these features are phonetic or syntactic; others are ‘semantic’ in that they encode information directly relevant to the sense(s) of the word. Such semantic features are commonly appealed to in lexical semantics and psycholinguistics. A related but distinct assumption is that words give access to encyclopaedic information (e.g. in relevance theory: see section 2).

These assumptions set the stage for a view of non-literal uses which has been explored by Carston and Wearing (2011). The idea is that some semantic features or encyclopaedic information is/are privileged. If one of these is ‘suppressed’ i.e. not deployed in the word token meaning, the result is non-literal. For example, it might be that ‘cut’ has a lexical semantic feature [+AFFECTS MATERIAL OBJECT] and ‘cut taxes’ is a non-literal use because that feature is not deployed there. Similarly, ‘prison’ may have a semantic feature [+BUILDING] (or [+MATERIAL OBJECT]) and it is not deployed in constructing the occasion-specific sense expressed in an utterance of ‘John’s job is a prison’, so this is a metaphorical use.¹⁹

If this view captures the distinction between literal and non-literal uses or the core of this distinction, the distinction is not only important for folk-linguistics. For whether a privileged feature of a word is suppressed or not is a matter of the mechanics of communication. It concerns meaning features that figure as input for the processing whose output is the meaning of the utterance. Hence, this proposal promises to kill two birds with a stone: the literal/non-literal distinction is explained as well as its importance for the theory of communication ensured.

However, there are two problems for this view, neither of which is conclusive, but which together shed considerable doubt on this strategy. One is that semantic features or encyclopaedic information whose suppression one might take to be criterial for metaphorical use are also suppressed in literal uses. The second problem is that the semantic features for which there is most linguistic evidence are deployed in metaphorical uses as well as literal ones.

Turning to the first problem, noted by L.J. Cohen (1993), consider the metaphorical (7), and literal (8) and (9):

(7) That man is a rabbit.

(8) I saw a rabbit running.

(9) I am cooking rabbit.

The feature suppression hypothesis claims that what renders an utterance of (7) metaphorical is that a semantic feature of the word ‘rabbit’ (or encyclopaedic information that the word gives access to) is suppressed. Suppose ‘rabbit’ has the features [+ANIMATE, +FARM ANIMAL, +EDIBLE, +MEAT] (as proposed by Klepousniotou, Titone, and Romero 2008, 1535). Then (7) drops at least [+FARM ANIMAL] and [+MEAT]. But each of these features is also suppressed in one of the literal uses: [+MEAT] in (8) and [+FARM ANIMAL] in (9) (Foraker and Murphy 2012, 424; Falkum and Vicente 2015, 6–7), so their non-deployment can’t be criterial for metaphor.

L.J. Cohen (1993) considers several types of semantic feature whose suppression might be sufficient for metaphor, including ‘features […] intrinsic to the lexeme’s superordinate semantic category’ (64) and ones that are important (or ‘central’) to the word meaning (65–66). Examples (8) and (9) above illustrate that features of the first type can be suppressed in literal uses. As for importance, Cohen points out that [+ANIMATE] seems to be an important feature of ‘lion’, yet it is dropped in the non-metaphorical (10).

(10) A stone lion needs no feeding. (Cohen’s example (16), 62.)

In any case, importance of features seems ill-suited to explaining metaphor because ‘importance is a matter of degree, but though metaphors may be better or worse, a phrase’s or sentence’s meaning seems to be either literal or metaphorical’ (Cohen 1993, 66).

The second problem for the feature-suppression account of metaphor relates more narrowly to linguistic semantic features, not encyclopaedic information. The problem is that the features for which there is most evidence are preserved in metaphorical uses. It remains controversial that words have semantic features (see fn 17 above). The evidence is stronger for some types of words: it is much stronger for verbs than other content words where it is very thin (Chomsky 1992, 217). The best-evidenced semantic features of verbs appear to be carried over to metaphorical use (Glanzberg 2008). So there’s not much evidence for linguistic semantic features that are suppressed in figurative use.

A great deal of work in lexical semantics indicates that lexical meaning of verbs divides into two: ‘syntacto-semantic’ features that are visible to linguistic syntax, plus ‘idiosyncratic’ content. The features capture thematic role types (whether a verb requires an agent argument, for example) and event structures (e.g. whether a verb is telic – see below for explanation.) Several authors have proposed that this syntacto-semantic skeleton of a word persists in metaphorical uses (Glanzberg 2008; Ramchand 2014; Spalek 2015; McNally and Spalek 2019). We set out the argument using data from Glanzberg. The Italian word normally used to translate English ‘blush’ is ‘arrosire’. These verbs have very similar meanings, but differ in their lexical aspect. ‘Blush’ is an atelic verb: it means something like have red cheeks, while ‘arrosire’ is a telic verb: its meaning is closer to come to have red cheeks (cf. English ‘redden’).

That difference accounts for the following contrast in acceptability:

(11) John blushed all day long.

(12) *John arrossí tutto il giorno.

Metaphorical uses of both verbs are possible, as in the following examples:

(13) The sky blushed.

(14) Il cielo arrossí.

Crucially, they obey the same constraint as literal uses, as the following examples show:

(15) The sky blushed all day long.

(16) *Il cielo arrossí tutto il giorno.

This is not an isolated case. McNally and Spalek show a similar continuity of differences between a number of English and Spanish verbs including English ‘cut’ and its Spanish counterpart ‘cortar’ across literal and figurative uses. In general, they say:

[T]he respective conventional event structures and thematic role types that a given set of counterpart verbs impose on the events they (literally or figuratively) describe will remain consistent within each language for both literal and syntactically productive figurative uses, and potentially vary across languages. (McNally and Spalek 2019)

The arguments in this section do not disprove the claim that there are semantic features whose suppression or non-deployment is necessary for non-literal uses, but they put it under considerable strain. We, therefore, suggest a different direction.

6. Reviving the literal/figurative distinction

Our claim is that literal uses of a word²⁰ are made with the intention to continue a ‘tradition’ of using a word. The central case is that the speaker intends to conform and use the word as other people, in particular its originators, use or used it.²¹ A metaphorical use is one which does not so conform, but depends on the pre-existing practice which runs through the literal uses. So there is a distinction between a basis on the one hand and a use that depends on the basis on the other hand and deviates from it. In our view, this kind of distinction between basis and deviation is crucial to any understanding of the distinction between literal and metaphorical meaning and truth. The metaphorical is derived from and therefore dependent on the literal. Our view is thus a kind of ‘asymmetric dependency’ view. Both literal and metaphorical uses depend on the conventional meaning but the convention does not depend on metaphorical uses.²²

To see that this is plausible, consider this challenge: Coin a new monomorphemic word and simultaneously use it metaphorically!²³ Can you do it? No, you can’t! Why? Because the literal meaning and use has to be established before you can deviate from it. (One cannot start and subvert a new fashion at the same time).

According to our proposal, there are two ways in which the use of a word can catch on:

Conservative Chain: the word is originated and it is passed on from speaker to speaker where each of the speakers has the intention to conform to the people from which he acquired the word.

In a conservative chain the semantic value of a word may change because of mistakes speakers make. Gareth Evans made this point by means of his well-known ‘Madagascar’ example.

We learn from Isaac Taylor’s book: Names and their History, 1898: ‘In the case of Madagascar a hearsay report of Malay or Arab sailors misunderstood by Marco Polo … has had the effect of transferring a corrupt form of the name of a portion of the African mainland to the great African Island.’ (Evans 1973, 196)

According to Evans’ description, the same name ‘Madagascar’ changed its referent because of Marco Polo’s mistake. The important observation for our purposes is that the uses of the name before and after the change of reference are all made with the intention to conform. This thought motivates the idea of a conservative chain of uses. An originating use introduces a new word w via a baptism (‘This stuff is jade’), stipulation or a mistake that catches on, as in the ‘Madagscar’ example.²⁴ Continuation of the originating use requires the intention to use as word w with the semantic properties, whatever they are, the originators invested it with. A chain of uses of a word which starts with an originating use and contains only continuing uses is a conservative chain. Conservative uses of w are intended to preserve the semantic properties of the originating use. The relevant intention is of course not the intention to preserve these semantic properties or to continue a tradition or practice. This would demand too much from most users. Rather, the intention is to ‘go along’ with how this word was used by them. The right reference of ‘them’ is secured by past facts: the speaker learned the word from these people and not others.

This feature of conservative uses is, we submit, the core of making literal utterances. A name is, if its use is part of conservative chain, a ‘channel for the acquisition of knowledge’ (Williamson 2007, 264) about an object and this makes its use literal and scientifically important. Similar things hold for other kinds of expression.

Sainsbury (2005, 117) argues that a continuing use of a proper name also needs to be sensitive to information from the initiating use: some information associated with the name at the point of its introduction is ‘carried forward’. However, there is not a privileged piece of information that needs to be carried forward in every use that continues a practice. Sainsbury’s condition seems plausible for other kind of expressions. There is information that is more firmly associated with a word like the general term ‘cat’: the belief that all cats are animals is firmly held and important for our conception of cats, but we can envisage it being falsified. We can envisage finding out that what we call ‘cats’ are robots controlled from Mars.²⁵ In this situation, we could say ‘Hey, cats turned out not be animals at all!’. What ensured continuation of the ‘cat’ usage was the intention to conform with the originators and preservation of some information (say about shape and movement) associated with the originating use.

Conservative chains of use contrast with creative chains:

Creative Chain: the word is originated and it is passed on from speaker to speaker; but some uses are not made with the intention to conform to the people from which she acquired the word, yet these uses depend (and are intended to depend) on the originating uses and are sensitive to some information associated with such uses.

Take an example. The English noun ‘the sun’ originated as a name of, well, the sun. We can easily use it with the intention to conform to the people who originated it. If we do so, the proper name is part of a channel for the acquisition of knowledge about the sun. But when, for example, Shakespeare wrote:

(17) It is the east, and Juliet is the sun.

He neither intends to conform to the original use of ‘the sun’ − he does not want to use it as the originators used it − nor would it be charitable to impute such an intention to him. He would be massively mistaken if he thought that Juliet is the same thing as the star at the centre of the solar system.

But he is also not intending to create a new word by using ‘the sun’ in a new way as someone who dubs his child with the name ‘David’ does. Shakespeare used ‘the sun’ because that word already had an established use in which it refers to the sun. Without such an established use, non-literal utterances are not possible. This is the plausible core of the ‘folk concept’ of the non-literal to which Recanati drew our attention. So on the one hand Shakespeare’s use of ‘the sun’ depends on the original use of ‘the sun’ as a term for the sun, while, on the other hand, Shakespeare does not intend to conform to this use.

In what way does Shakespeare’s non-conformist use of ‘the sun’ depend on the originating use? Here we agree with much of the relevance-theoretic view set out above. The original use and conforming uses of ‘the sun’ are associated with a body of beliefs, stereotypes and more about the referent of the name. We can think of them as collected into a file labelled ‘the sun’. Shakespeare’s use of ‘the sun’ is sensitive to some information in the original file.²⁶ For example, the original file will contain features like STRIKING, CENTRAL. If – in the context of use – Juliet is neither striking nor central, yet Shakespeare continues to assert that she is the sun, we might start to suspect that he is coining a new word. This suspicion turns out to be unwarranted, if his use of ‘the sun’, though non-conformist, is subject to some constraints that are associated with the originating uses. The tension between the non-conformist intention of the speaker on the one hand and the intention to respect some information from the originating uses needs to be resolved. The speaker wants to riff on an existing practice, not start a new one.²⁷ We need to ascribe to the use of the word a meaning that respects the non-conformist speaker intention and the fact that the speaker did not intend to coin a new word.

With this in mind, we can formulate the Non-Conformity View of Non-Literal Use (‘NCV’ for short) as follows. The basic idea is that a literal use of a word w is a use of it that is part of a conservative chain; a non-literal use, in turn, is a use of w that is neither an originating use of w nor made with the intention to conform to such a use, yet the use is sensitive to some information associated with the originating use of w.

Let us consider an example to illustrate NCV. Consider (4) again:

(4 repeated) The mind is a computer.

Suppose the speaker of (4) intends to conform to the original usage of ‘computer’. In this case the speaker ‘opened up an information channel’ to computers. Hence, accepting the utterance of (4) requires us to revise something we may have taken for granted, namely that only artefacts can be computers. Then the utterance may be literally true and if so may tell us something new about computers. Here we have a conservative modulation of ‘computer’. Some of the users may be mistaken about what it takes to conform to such a use. However, as long as all uses are made with the intention to conform to the original uses, the name is used literally and utterances containing it can be assessed for literal truth and falsity.

If, on the other hand, (4) is used without the intention to conform to the originating use of ‘computer’, the knowledge channel to computers is blocked. Someone who, for instance, fully understands Shakespeare’s utterance of ‘Juliet is the sun’ cannot thereby extend their knowledge about the sun. This gets at the feature that diminished the value of metaphorical assertions in science. Recall Fodor’s (2007) point: ‘That Juliet is the sun is, perhaps, figuratively true; but since it is only figuratively true, it’s of no astronomical interest.’ NCV helps to explain why ‘The brightest object visible from earth is the sun’ is about the sun, while ‘Juliet is the sun’ is indeed not about the sun and so not a contribution to astronomy.

While live metaphors do, dead metaphors don’t meet condition (iii) above. A dead metaphor came into being at some point because there was an originating use, but now it is no longer sensitive to the information associated with the originating use. Understanding a use of ‘eye of the needle’ does not depend on bringing information about eyes etc. to bear.²⁸ This is a good result, we think: dead metaphors are as little metaphors as toy ducks are ducks.²⁹

7. Advantages of the non-conformity view of non-literal use

Let’s first make clear why NCV is worth exploring and defending. We agree that the majority of uses of a word express a meaning that is distinct from its standing meaning (if it even has one). Yet we don’t take that to be a reason to give up the literal/non-literal distinction. Why? Many of the meanings that are generated ‘on the fly’ are expressed by uses of the word in which the speaker intends to conform to the originating use. We often need to stretch the use of a word – inter alia, to say something scientifically valuable about a topic. In these uses, we still want to say something about the original extension of the word, yet we want our audience to correct for contingent and distracting associations which might narrow or widen the extension. These uses are literal, even if they are distinct in meaning from the originating use.

To sum this up: The basic selling point of the Non-Conformity View of Non-Literal Use is that it makes room for lexical modulation, polysemy etc. in literal uses. That is, we have recast the traditional view of metaphor as usage that is in some sense deviant while making room for the (apparent) facts that literal meanings of a word are multiple, contextually sensitive etc.

Note that the NCV doesn’t make any substantial assumptions about the standing meaning of a word. For example, we don’t need to implausibly assume that there is such a thing in the first place or that it is determined by convention or by the semantic features of a word or the encyclopaedic information it gives access to. All we need is the less controversial thesis that there are originating uses and intentions to conform or lack thereof. The originating use may even be empty and yet subsequent uses may be made with the intention to conform to it.

The Non-Conformity View of Non-Literal Use has further attractive features. It applies neatly to cases that are problematic for the famous Gricean criterion. Grice saw metaphors as violations of a maxim of truthfulness. On this view, the blatant falsehood of the literal meaning of the sentence licences a metaphorical interpretation. One well-known problem for this view is the existence of ‘twice-apt’ cases like (18), which could be intended (and understood) as metaphorical, even though the sentence has a true literal meaning:

(18) Moscow is a cold city.

Our criterion fares better. If an utterance of (18) is intended metaphorically then it is not about coldness. As Fodor might have put it, it’s not a contribution to meteorology. Contrast with a literal use, where you could indeed learn that Moscow is cold, or, if you already know about temperatures in Moscow, about how the word ‘cold’ is used literally.

Many negated metaphors are similarly problematic for the Gricean view, but they are covered unproblematically by our criterion. Consider (19):

(19) No man is an island.

If this is intended metaphorically, it’s not about islands: it’s not a contribution to geography. Contrast with a similar sentence used literally, which is about islands and from which you could learn about geography:

(20) The Isle of Dogs is not an island.

A further advantage of NCV is that it applies not only to metaphors of the form Xs are Ys/An X is a Y where the word or phrase used metaphorically is nominal, but also to verbal metaphors like ‘The motor complained’ (King and Gentner 2022). Here the verb ‘complain’ is used in a way that depends on the literal tradition but which does not belong to it.³⁰ It is not about complaining: an anthropologist or sociopragmatician studying complaints should not and obviously would not take this kind of usage as suggesting that she extend her fieldwork to the utterances of engines.

Thus far we have focussed on metaphor, but another attractive feature of NCV is that it categorises metonymy as non-literal without any need for added stipulations. Consider some classic cases. The first use of ‘10 Downing Street’ to refer to a group of people who work at 10 Downing Street relied on the already extant tradition in which the name refers to a building, but wasn’t intended to belong to that tradition.³¹ Similarly with ‘The ham sandwich is parked out back’ (Nunberg 1995): you can’t learn about ham sandwiches from this use (except indirectly, by learning about people who consume them). It would be a mistake to infer from it that ham sandwiches drive cars. The information channel is blocked, just as it is in metaphorical uses.

On the other hand, prima facie our account is not a good fit with irony. It does not seem right to claim that an ironic utterance of (21) is not about lovely days, or lovely days for picnics: a key point of the utterance is that the speaker thinks today is not such a day. Equally, it’s not clear that the speaker aims to use any of the words here in a new way.

(21) It’s a lovely day for a picnic [spoken in pouring rain].

Given that irony is traditionally classified as a trope – that is, categorised with metaphor and metonymy – this might seem to be a problem for our account. We have two responses. First, it has become increasingly common to see irony as distinct from metaphor and other tropes on theoretical and empirical grounds. The empirical difference is most clear in the developmental trajectory of irony. In recent years, it has been shown that with suitable materials very young children (from around 3 years old) comprehend metaphor (Pouscoulous 2011; Pouscoulous and Tomasello 2020) and metonymy (Falkum, Recasens, and Clark 2017; Köder and Falkum 2020). In contrast, even with similar facilitating materials, irony comprehension is only found from around 6 years old (Pouscoulous 2013; Köder and Falkum 2021), evidence for dependence on a late-developing capacity, perhaps higher-order metarepresentation abilities (Happé 1993). This coheres with current theories of irony. Sperber and Wilson (1981, 1992) claim that irony involves attribution of a thought or utterance to another agent. On their account, in (21) the speaker aims to convey something along these lines: someone said or thought or might say or think that it’s a lovely day for a picnic, and that would be a ridiculous thing to say/think. Other post-Gricean theorists see irony as pretence (Clark and Gerrig 1984; Walton 1990; Recanati 2004; Camp 2012), and there are hybrid attributive-pretence accounts (see Wilson 2006). All of these views see irony as a matter of the speaker having a different attitude towards her utterance (or its content) from the attitude of endorsement characteristic of sincere assertions – including metaphorical ones.³² In contrast, metaphor and metonymy pertain to use of words. (In our view this contrast explains why metaphor can be used ironically – e.g. ‘You are the cream in my coffee’ expressing a bitterly negative evaluation – but not as far as we know the other way round.³³) To the extent that irony is an utterance-level phenomenon it is to be expected that it falls outside our NCV account given that the account is about non-traditional uses of words.

However, the second point we would like to make about irony is that there are cases where the dissociative attitude applies specifically to a word, and these seem to fall naturally within our account. Consider (22):

(22) I left my bag in the restaurant, and someone kindly walked off with it. (Wilson 2009, 192)

Here, just the word ‘kindly’ is ironic (as might be indicated in writing by putting quotation marks around it). One could see this as a use of ‘kindly’ that relies on the tradition of use of the word but intentionally deviates from it and is thus a non-literal use on the NCV account. A point in favour of this view is that ironic uses of words, like metaphorical and metonymic ones, can create new traditions. A great deal of lexical semantic change including the emergence of polysemic senses is driven by metaphors and metonymies that catch on (Traugott 2017; see footnote 25). But there are also cases which appear to show a new tradition founded on an ironic (or at least dissociative) use of a word. Here we have in mind some ‘meaning reversal’ cases such as the slang use of ‘bad’ to mean ‘very good’.³⁴

8. Denominal verbs

We conclude by sharpening our proposal in the light of a prima facie counter-example. So-called denominal verbs are derived from nouns. Consider an often used example:

(23) The boy porched the newspaper. (Clark and Clark 1979)

Is this a derivative use of the word ‘porch’? If it is, it is a use in a creative chain. The use of the word depends on an originating use, but it is clearly part of a creative chain: the user does not want to go on as before. So it may seem that NCV classifies such uses as non-literal. Yet is ‘The boy porched the newspaper’ a metaphor or metonymic? Our intuitions at least don’t pronounce clearly on this case.

The right response to this objection seems to us to deny the assumption that the noun ‘porch’ and the verb ‘to porch’ are the same word. After all, NCV is only concerned with literal and non-literal uses of one and the same word. The verb ‘to porch’ is a new word whose introduction was possible because ‘porch’ (noun) was already around and its introduction is inspired etc. by it. A subsequent utterance of ‘I porched the newspaper’ can be a normal literal use because ‘to porch’ is used with the intention to conform to the initiating uses of ‘to porch’.

Why is the verb a new word although it could only be introduced and used in virtue of the fact that the noun ‘porch’ was already established? The verb has a semantic value of a different kind and different syntactic and distributional properties. Intuitively, ‘to porch’ was also introduced with the aim of filling a gap in our vocabulary. There is a certain activity related to porches. The innovating speaker creates an expression that signifies this activity by making a new verb, but one that bears on its sleeve its relation to the noun. Hence, denominal verbs don’t threaten our proposal.³⁵

9. Conclusion

We have set out a deviation account of the literal/metaphorical distinction which is compatible with ubiquitous lexical modulation. In our view, the traditional deviation picture has two components which have not been sufficiently distinguished before:

1. metaphor (and other non-literal uses of words) contrasts with literal use in the following way: in literal use what is meant by a word is equal to the linguistically encoded or conventional meaning of the word, while in non-literal uses the contrary is the case: there is an occasion-specific meaning not equal to the standing meaning;

2. metaphorical uses (and other non-literal uses of words) depend asymmetrically on the literal.

We think that ubiquitous lexical modulation requires us to give up (i). We have tried to show that a deviation account based on (ii) without (i) is compatible with ubiquitous lexical modulation and attractive in other ways. It turns out that asymmetric dependency is the core of deviation views of the literal/non-literal distinction.

Is the rescued literal/non-literal distinction the one common sense makes? Maybe it is not exactly the same distinction, but a close successor. We aim to articulate the distinction which is tied to the preservation of subject-matter in literal talk, including much scientific discourse. This distinction is worth preserving and has theoretical import.³⁶

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

Nicholas Allott’s work was supported by the Research Council of Norway [grant number 302083] (FINNUT).

Notes

1 In our view, the linguistic semantics of a word is determined by a combination of conventions and cognitive abilities. For details see Allott and Textor (2017), and below.

2 For the deviation view see Ortony (1993, 2), White (2001, 24, 32–33), Recanati (2004, 81), Stern (2006, 250), Reimer and Camp (2006).

3 On a Gricean view of speaker meaning, this amounts to intending that its meaning is entertained or accepted by the addressee. We doubt that claim, but do not discuss alternatives here.

4 Fodor (1998, 53–54) pursues this strategy vs. the apparent polysemy of ‘keep’.

5 See also Searle (1983, 146), Vicente (2019, 921–922).

6 For useful summaries of this work see Falkum and Vicente (2015, 4–5), Liu (2022, §2.2), Quilty-Dunn (2020, 8–12).

7 80% on one estimate: Rodd (2020, 411).

8 This view has been contested. Klein and Murphy (2001, 2002) and Foraker and Murphy (2012) argue that polysemes have distinct stored representations for each sense. However Frisson’s (2015) results are hard for such ‘sense enumeration’ views to explain (Falkum and Vicente 2015, 4–5). It remains possible that sense enumeration is correct for some polysemous words, perhaps ones where the senses are less closely related (Frisson 2009, 120; Liu 2022, §2.2).

9 See also Bezuidenhout (2001, 179) and White (2001, 24).

10 For Wilson and Carston ‘encyclopaedic property’ is a shorthand for proposition that has an encyclopaedic property as a constituent (2007, 256, n. 22).

11 The view we oppose is endorsed by Rumelhart (1993), Bezuidenhout (2001), Carston (2002, 349), Wilson and Carston (2007), Sperber and Wilson (2008).

12 A prominent example is Giora (2008).

13 A revised version has been published as McNally & Spalek (2022).

14 Does this imply that only literally true statements can constitute a satisfactory theory? For a while philosophers believed so. Quine (1981, 189) suggests that at least ‘in the inner stretches of science’ tropes are cleared away. But Yablo (1998, 250–253) and others have argued that some metaphors are representationally essential.

15 See Evans (1982, 310) for the view that communication is essentially a mode of knowledge transmission. We think Evans has literal communication in mind.

16 Our reservation here is because a) the feeling might be epiphenomenal relative to processing utterances, and b) a theory of processing of Xs (here utterances) does not need to explain all the intuitions that agents have during processing of Xs. It’s a virtue of the theory if it does explain such intuitions, but that might be outweighed by other explanatory virtues possessed by rival theories that do not explain them.

17 Recanati agrees (2004, 77 fn 11) but classifies such examples as non-figurative!

18 Recanati glosses ‘available’ as ‘consciously accessible’ (2004, 42) suggesting that he has this distinction in mind. On introspection and linguistic intuitions see Gross (2021, 546): ‘it can take time and effort to get a reading of a sentence, and reflection and perhaps conscious inference to arrive at a truth-judgment’.

19 These proposed semantic features which are claimed to be necessary for literal use are similar to meaning postulates (Carnap 1947; Fodor et al. 1980; Fodor 1981). They may be vulnerable to familiar arguments against the analytic-synthetic distinction and lexical decomposition (e.g. Fodor 1998), but see Pietroski (2003) and Rey (2022b).

20 We regard homonyms as different words. We are not claiming that all literal uses of the string ‘bank’ contribute to one tradition.

21 We are influenced here by the view of concepts developed by Schroeter and Schroeter (2016), and the view of words set out in Allott and Textor (2017).

22 On asymmetric dependency theories of meaning, see Fodor (1987), Rey (2022a). Rey suggests that figurative uses asymmetrically depend on literal uses (§4.3).

23 The reason for the qualification ‘monomorphemic’ is that it may be possible for the first use of a morphologically complex form to be metaphorical. One might never have heard ‘hardball’ used other than metaphorically, and one could understand metaphorical uses on the basis of knowing how the words ‘hard’ and ‘ball’ are used literally, with some help from context. Indeed, for all we know, the first use of this compound may have been metaphorical. (Our thanks to Yael Friedman and Jack Wright for pushing us on this point.)

24 See Sainsbury (2005, ch. 3.5) for an instructive model of originating uses.

25 See Putnam (1962, 660f).

26 Note that our claim about sensitivity is compatible with metaphorical uses expressing ‘emergent’ features that are possessed by neither the metaphor vehicle nor the metaphor topic in isolation from each other (Wilson and Carston 2006). We also agree with Camp (2020, 315) that metaphors often traffic in ‘common higher-level features [which] are implemented in qualitatively different ways within the two domains’. Juliet is not literally radiant, but (as Camp says) her beauty is similar to the sun’s brightness in that they both outstrip potential competitors (other Veronese women / other astronomical bodies).

27 Of course, in some cases, a novel use catches on and starts a new tradition. Conventionalisation followed by lexicalisation of metaphorical uses is known to be among the key mechanisms of lexical semantic change (Traugott 2017; see also Bowdle and Gentner 2005).

28 The point and this example are from Rey (2022a, §4.3).

29 We agree on this point with Black (1977, 439).

30 This use of ‘complain’ is listed in some dictionaries, which suggests that it has become conventionalised. It is a distinct tradition that still maintains a relation to the original literal one.

31 This particular metonymic use has become conventionalised, but not dead. Like the use of ‘complain’ discussed above, it is a distinct tradition which maintains a relation to the literal one.

32 One theorist who might be thought to group irony and metaphor together is Kendall Walton, who sees both irony (1990) and metaphor (1993) as cases of pretence. But his view of irony, unlike metaphor, is that it also involves attribution (Walton 1990, 222–223).

33 Bezuidenhout (2001, 162–164) discusses and defends this generalisation.

34 https://www.macmillandictionaryblog.com/something-wicked-this-way-comes.

35 An alternative way to accommodate novel denominal verbs within our account would be to treat them as metonymic, as proposed on independent grounds by Ingrid Lossius Falkum and Deirdre Wilson in recent unpublished work.

36 The authors are very grateful for helpful comments and questions on earlier versions of this paper to audiences at UCL and the University of Oslo, Robyn Carston, Terje Lohndal, Ingrid Lossius Falkum, Georges Rey and Deirdre Wilson.