Phraseological units, ranging from fixed collocations to idioms, are often defined as non-compositional strings, which reflects their formal characteristics but not their semantic traits. Semantic analysability – the ability to derive the meaning of the whole from the individual meanings of the components – is also a characteristic of phraseological units, but one that tends to be reserved for idioms. Drawing from extensive work into the relationship between metaphor, idiom and phraseology (Philip forthcoming), this contribution will demonstrate how semantic prosody interacts with semantic analysability. Semantic analysability operates along a continuum, and it will be seen that the more phraseological meaning is distanced from the meanings of its component parts, the more pronounced and important the semantic prosody becomes. Comprehensive discussion on the terminology used is included, as are detailed examples of the lexical items examined.
Semantic prosody. The term seems to be attracting a prosody of its very own. It is arguably one of the most problematic notions to have emerged from Sinclair’s post-Firthian approach to the analysis of language using electronic text corpora. The term is fraught with difficulty, because there seems to be genuine confusion as to what precisely it refers to; and coupled with this is the inescapable fact that some lexical items seem not to have a semantic prosody at all, however one cares to define it.
My aim in this paper is to examine the reasons why semantic prosody is not always present (or detectable). Semantic prosodies seem to be inconstant friends, sometimes there, sometimes not, sometimes taken for granted, sometimes looked-for but vexingly evasive. The very fact of their inconstancy has given some, Whitsitt (2005) in particular, reason to conclude that the notion of prosody is a figment of corpus linguists’ imaginations, dreamed up from an eagerness to see something in the data, and that one believes (or not) in semantic prosody much in the same way as one believes (or not) in ghosts. There are, in fact, very good linguistic reasons why semantic prosodies are not always present, why they are not always required: semantic prosodies add meaning which goes beyond the meaning already expressed, suggested or connoted by word-semantics, but sometimes the words alone are enough, so the semantic prosody is redundant.
This paper examines the components of Sinclair’s extended unit of meaning in relation to formulaic phraseological sequences to demonstrate how semantic prosody (as defined by Louw and Sinclair) is inextricably linked to semantic analysability and how it appears to be a fundamental aspect of idiomatic language but not of all non-compositional language, as has previously been supposed.
2. Definition(s) of semantic prosody
One of the reasons why semantic prosody has been open to attack is that the uses to which the term is put vary considerably from author to author. This makes the concept appear vague and ill-defined and can lead to confusion and misunderstanding. For this reason it becomes necessary to establish what precisely semantic prosody does and does not refer to, and to this end this Section 2.1 revisits the origins of the term, while Section 2.2 investigates subsequent developments and modifications in its use over time. In this way the argument in the Section 4, in which the semantic prosodies of a range of linguistic examples are discussed, should be clearer, because the ambiguities surrounding the application of the term itself will have been tackled.
2.1 Back to basics
Semantic prosody as a concept underwent a degree of evolution before it was formally announced to the research community in Louw’s 1993 paper, Irony in the Text or Insincerity in the Writer?: The Diagnostic Potential of Semantic Prosodies. The most frequently-cited definition of semantic prosody comes from this paper: “a consistent aura of meaning with which a form is imbued by its collocates” (Louw 1993: 157). Yet despite its familiarity, this is more a description than a definition, and a partial one at that: Louw himself (personal communication) regrets its ubiquity, not least because it has bolstered the widespread interpretation of semantic prosody as being a form of connotation and primarily associated with positive or, more commonly, negative evaluation (Stubbs 1995, Louw 2000, Dilts and Newman 2006, Hunston 2007). Yet semantic prosody is more complex than mere positive/negative charge and, crucially, cannot be identified from introspection, as Louw explicitly states: “semantic prosodies are a collocational phenomenon and one which is preferably to be regarded as recoverable computationally from large language corpora rather than intuitively” (Louw 2000: 48). 
Louw and Sinclair collaborated closely on the phenomenon of semantic prosody from its origins in the early years of the COBUILD project to its crystallisation and terminological labelling in the early 1990s. Yet because the two scholars applied the concept to different kinds of language, their writings tend to emphasise different features of the whole. While Sinclair’s focus was on determining the baseline norms of semantic prosody, i.e. its role in typical phraseology, Louw’s preoccupation has been to investigate the semantic clashes which occur when the prosody is “violated” through atypical phraseolology. If one focuses too closely on Louw’s writings on semantic prosody, one is apt to overlook the most important feature of semantic prosody. In addition to providing an “aura of meaning” and being identifiable only by examining the repeated occurrences of a unit of meaning in corpus data, the “primary function [of semantic prosody] is the expression of the attitude of its speaker or writer towards some pragmatic situation” (Louw 2000: 57; my emphasis). In other words, “it expresses something close to the ‘function’ of the item – it shows how the rest of the item is to be interpreted functionally” (Sinclair 1996: 34). This attention to the functional aspect of language is a clear reference to the Firthian admonition (Firth 1968) to renew the connection between language and the context of situation, which comprises the interaction of:
A. The relevant features of participants: persons, personalities.
(i) The verbal action of the participants.
(ii) The non-verbal action of the participants.
B. The relevant objects.
C. The effect of the verbal action. (Firth 1957: 182)
Thus semantic prosody is not discernable from the words of a lexical item alone, but requires those words to be used by a particular set of participants to obtain a particular effect relative to particular objects.
Corpus texts facilitate the retrieval of recurrent patterns, but they do so at the expense of the context of situation in which the language under study was originally uttered. Semantic prosodies, therefore, have to be inferred by extracting information from the cotext which allows a picture of the context of situation to be built up. This is not dissimilar to the way that the semantic preference of a lexical item is identified, though in determining the semantic prosody, clues are as likely to lie in the colligational patterns as in the collocational ones, and may not even emerge from repeated forms, but from repeated nuances instead
As there is no single contextual feature which acts as a “key” to access the semantic prosody, the four worked examples found in Sinclair (1996: 83-93) are summarised here. The core naked eye is said to have a prosody of “difficulty” (ibid. 87), determined by the presence of modal operators (can, could), limiting adverbs (barely, rarely, just) and adjectives such as small, weak, faint and difficult combining with the semantic preference of “visibility” (ibid.). For true feelings, the identified prosody is “reluctance/inability” (ibid: 89), which emerges this time from the presence of a range of verbs which all express some form of impediment (prevent from, careful about, feel guilty about; try to, incapable of, unable to) combining with the semantic preference for “expression”. For the verb brook, “the prosody can be crudely expressed by ‘reported threat by authority’” (ibid. 91); crudely, because this example is an illustration of just how complex a semantic prosody can be. It involves an expression of intolerance combined with future modality, “the displacement by report of the threat, and the frequent naming of authority figures as subject” (ibid.) and the phrase is additionally “emotionally charged with the commitment of the threatener to carry out the threat” (ibid. 92). Finally, Sinclair decides that (my)place (i.e. the place where I live) has “informal invitation” as its prosody (ibid: 93), in patternings which differentiate one meaning of place (home) from others.
From this summary, it emerges that semantic prosodies are often difficult to describe clearly and succinctly, and this may well explain the widespread tendency to speak loosely of positive/negative prosodies rather than attempt to articulate the semantic prosody more precisely. Reducing a semantic prosody of the complexity observed above for brook to a mere positive/negative charge, in fact, is to ignore the primary function of semantic prosody as set out by Louw (2000: 57), and therefore fail to consider the contribution that the prosody makes to the overall meaning conveyed by the lexical unit. Positivity and negativity, evaluation and connotation are all important aspects of the semantics of a lexical unit, but as Sinclair (1998)  states, “[t]he semantic prosody of an item is the reason why it is chosen, over and above the semantic preferences that also characterize it” (1998: 19, my emphasis). In short, the semantic prosody associated with a lexical item communicates an attitudinal, evaluative or emotional stance with regard to a particular context or scenario and its outcome (anticipated or actual), not simply a vague and ill-defined “aura of meaning”.
Section 2.2 examines definitions of semantic prosody which have been put forward by other scholars and demonstrates in which ways these redefined meanings differ qualitatively from that which has emerged in this Section. Both “original” semantic prosody and what we might call “semantic prosody revisited” are important for the analysis to follow, though it will be seen that distinguishing the one form the other makes it possible to separate out two different levels of linguistic analysis.
2.2. Semantic prosody revisited
Because semantic prosody evolved over time before reaching maturity in Sinclair’s 1996 paper, the “aura of meaning” aspect, often involving “collocational inferences” (Hunston 2005: 260), has generally been the characteristic that other writers on semantic prosody have picked up on. Certainly, the earliest mention of semantic prosody, relating to set in (Sinclair 1987) and happen (Sinclair 1991) focus on this feature. However, as we have seen, this is not the primary aspect of semantic prosody, but just one part of it. More recent uses of the term, most notably those of Partington (1998, 2004), Morley and Partington (2009), Hunston (1995), and Hunston and Francis (2000) (see Stewart 2009 for a comprehensive overview) have tended to focus exclusively on the “aura of meaning” surrounding the node, paying little attention the functional meaning expressed by the unit of meaning as a whole.
In essence, then, there are two facets to semantic prosody, and although they both contribute to the finer shades of meaning of a lexical unit, they cannot satisfactorily be reduced to a single entity. Separating them out, and referring to each with its proper term, rather than persisting in the use of a single term to cover two distinct phenomena, is therefore desirable (see also Hunston 2007). This is especially true because the phenomena described by semantic prosody are all rather abstract and intangible aspects of meaning which are more difficult to describe effectively than is, say the allocation of collocations to lexical or semantic sets in order to identify the semantic preference. Hoey has already paved the way for this separation in terminology. In advancing the term semantic association (Hoey 2005: 16 ff.), apparently to subsume “two different concepts, sometimes confused with each other” (Hoey 2005: 22), i.e. semantic preference and semantic prosody, he has in fact opened up the middle ground between the two thus allowing linguists to discuss “auras of meaning” and other abstractions from the semantic preferences of a lexical item while reserving semantic prosody for reference to functions and pragmatic effects alone.
If we adopt the term semantic association to fill a gap rather than to pull different phenomena together as Hoey proposes (ibid. 22-23), Sinclair’s (1996) quadripartite extended unit of meaning can be split into five: collocation, colligation, semantic preference, semantic association, and semantic (pragmatic) prosody. Splitting up the abstract components of the unit of meaning is preferable to lumping them all together, and it can be noticed that a semantic preference sits very comfortably between semantic preference and semantic (or pragmatic) prosody. Semantic association does not describe the function (although it does contribute towards making a lexical item appropriate for expressing that function), so does not properly belong with semantic prosody; yet it is distinct from semantic preference in that it is describing secondary meanings – associations, evaluation and connotations – which are fundamentally psychological in nature and therefore not necessarily shared by all members of a speech community. Semantic preference, on the other hand, relates to received meanings which are shared and are attested in outside sources such as dictionary definitions. Thus the concrete instantiations of collocates and colligates determine which meaning of a lexical item is intended: what it means. The semantic preference of the lexical item for particular lexical/semantic sets, and the semantic associations that those may evoke, colour the selected meaning with attitudinal and evaluative nuances: how it means what it does, in relation to what. The semantic prosody renews the connection of this semantic information with the reality of language in use, taking meaning out of our heads and back into the real world: where, when, why, and to whom it means what it does.
Close inspection of the definitions of semantic prosody that have been supplied by scholars other than Louw and Sinclair confirm that it is on the whole the semantic associations that they are describing. Here, the semantic prosody (i.e., association) is identified from the co-occurrence of the node with members of a particular semantic set (Hunston 1995: 137), and can be defined as “a sub-category, or a special case, of semantic preference” (Partington 2004: 149) which is “both shaped by and expressed in its semantic preferences” (Morley and Partington 2009: 142). Partington in particular has stressed the psychological nature of semantic prosody (i.e., association), equating it with evaluative meaning (Partington 2004:131; Morley and Partington 2009) and declaring it “an aspect of expressive connotation” (Partington 1998:66), as, too, does Whitsitt, who dismisses the phenomenon as “simply connotation spread over several words” (Whitsitt 2005: 285). These definitions give shape to an entity which merits its own classification, and for the purposes of this paper it will be referred to as semantic association, while the strictly pragmatic aspects of meaning will continue to be referred to as semantic prosody.
It has emerged from the discussion above that semantic association is closely linked to semantic preference and the co-occurrence of the node with words which share semantic traits. Semantic prosody, on the other hand, relies on collocates and colligates in roughly equal measure, if anything tending to favour the patterns and participants involved in verbal processes over lexical-semantic features per se. This leads to a further distinction, namely that semantic associations are polarised while semantic prosodies are essentially modal. Semantic association, being essentially psychological, prioritises affect and attitude, whereas semantic prosody is firmly grounded in real-world interaction. The part played by context is also relevant, influencing a number of factors. Firstly, as semantic association “exists when a word or word sequence is associated in the mind of a language user with a semantic set or class, some members of which are also collocates for that user” (Hoey 2005: 24, my emphasis), it can be severed from the cotext and context. In doing so it tends to prioritise salient meanings (see Section 3) which are word-centric. Semantic prosody connects the meanings expressed by particular wordings with the context of situation, and is inseparable from the phrasing. It therefore applies to textual, delexical meanings and spreads over the entire unit of meaning rather than being focused on the node word alone (which is why Sinclair referred to it as a “prosody” in the first place: see Sinclair 2003: 117). A second factor brought about by the psychological nature of semantic associations is that they can vary from one individual to another and from one set of circumstances to another. Semantic prosodies cannot be controlled, manipulated or changed by the individual language user, because they are an integral feature of the wording. Finally, and related to the previous point, semantic associations may or may not be activated, while a semantic prosody, “once it is identified with a phrasing … will be part of the meaning even if it has no clear expression” (Sinclair 1996: 92). These differences are summarised in Table 1.
Abstracted from semantic preference
Affective and attitudinal
Variable and individualised
Abstracted from various contextual features
Integral to context of situation
Integral to the phraseology
Table 1: Comparison of the features of semantic preference and semantic prosody.
3. Composition and analysis
Phrases all share one characteristic: they are non-compositional, meaning that the combination of words used in the sequence that they are used is not generated anew out of the grammar every time it is required, but is instead pre-constructed and thus available for use as a single lexical choice. This feature is the one that dominates in the terms used to describe conventional phraseological sequences. But while this is the trait common to all phrases, there is another, parallel, aspect: semantic analysability, or the ease with which the meaning conveyed by the phrase can be determined by the meanings of its component words. This trait is central to the identification of idioms – phrases par excellence – but also plays a role in the meaning expressed by collocations and compounds formed by fused collocations. Collocations seem to be analysable, but of all the meanings that a collocation might represent, taking into account the various senses and sub-senses of their components and their possible combinations, only one of these constitutes the meaning of that collocation.
Semantic analysability operates along a continuum. Collocations with their restricted meanings can be found at one end, and the most opaque of idioms lie at the other extreme. In between is a range of structures which often exploit underspecified, nonsalient, and/or metaphorical meanings, all of which influence the immediacy with which the meaning of a phrase can be grasped. Underspecified meanings are associated with polysemous lexical items and occur when the contextual patternings do not provide sufficient information to allow one precise meaning to be selected, with the result that the meaning remains fuzzy and ill-defined. Underspecification is slightly different from ambiguity, which involves (usually two) clearly defined meanings, the uncertainty lying in which of these is the one intended. Nonsalient meanings have earlier (Table 1) been called textual, or delexical. While salient meanings are “coded meanings foremost on our mind due to conventionality, frequency, familiarity, or prototypicality” (Giora 2003: 10), and are associated with orthographic words. Nonsalient meanings, on the other hand, are peripheral meanings which are typically hidden to introspection. Their semantic prominence is compromised because it depends on collocation and cannot be identified with a single word form. Metaphorical meanings may well be salient, and like underspecified and salient meanings, are usually associated with orthographic words. Section 4 examines five phrases positioned at different points along the semantic analysability continuum, to show how semantic prosody steps in to fill the communicative gap left when semantic analysis cannot be carried out completely.
4. Degrees of idiomaticity
The focus on collocation and phraseology that characterises corpus-based semantics sometimes gives rise to the impression that word meaning does not exist, that all meaning is contextual, and that words severed from context are meaningless. In a sense, this is correct: words in isolation, devoid of cotext and context of situation, serve no communicative purpose and are therefore semantically empty. But word-meaning does exist – especially in our minds, as the vast psycholinguistic literature attests – and it comes to the fore in compositional language “whose only restraint is grammaticalness” (Sinclair 1991: 109). Teachers of language will notice that learners favour open choice in their production because they lack the necessary repertoire of prefabricated structures and collocations to do otherwise. Technical writing too favours open choice, as its fully compositional and fully analysable nature keeps ambiguity to a minimum. Yet open choice, by definition, does not admit regularities of patterning, except those imposed by the grammar of the language. It therefore resists forming units of meaning, and therefore no semantic prosody should be expected to be present. Nor does language constructed according to the open choice principle have any need for a semantic prosody – that aspect which connects language to action and interaction – because the words are used in their fully salient senses and express all the meaning that is required.
4.1 From bad to worse
While the very first members of the idiomaticity cline include collocations such as of course, which may well fuse into single words over time in the same way that maybe and anyway already have (Sinclair 1991:110), the argument which relates semantic prosody to idiomaticity will start from phrases which are not likely to fuse in the foreseeable future. One such item is the noncompositional sequence from bad to worse. The phrase has been examined in detail by Stewart (2008), from which the following summary is taken.
If a situation goes from bad to worse, it becomes even more unpleasant or unsatisfactory. (COBUILD)
As the definition reveals, from bad to worse is fully analysable, with the core meaning-bearing parts, bad and worse, being used in their most salient, basic senses. The definition also features the most commonly-found verb collocate, go, which Stewart (ibid.) notes is found in all its inflected forms. Fifty percent of the occurrences of the lemma go have things as their grammatical subject, with matters and situation also occurring, but much less often (3 and 2 times respectively, in 42 occurrences of the phrase in the BNC).
In terms of colligation, the phrase favours past tenses and eschews present simple and future tenses; but more interestingly, in 80% of occurrences it marks a clause boundary, being immediately followed by a full stop or comma. There is a semantic preference for movement, expressed by go, tumbled, trip and movement, which is conducted by things rather than people. Closer inspection reveals that things is a proxy form for situation or circumstance. This allows us to reinterpret movement as a metaphor for change, so the semantic preference relates to changing situations or circumstances.
Grouped by Stewart (2008) alongside the semantic preference, but here re-assigned to the category of semantic association, is a range of “unfavourable-sounding states of affairs”, including war, crisis, disaster, tensions, anxieties and several others. Stewart (ibid.) notes that what these terms have in common is a “melancholy attitude of unresolvability and hopelessness”, which is reinforced by the colligational features (past tenses and clause-finality, which lend a suggestion of foreclosure).
Having already pointed out that the phrase is fully analysable, we should expect nothing less than that the situation is bad to start with (from + bad + to + worse). However, what is important in terms of the semantic association is that the negativity is connected to lack of control, because it seems that “where people are at the mercy of forces beyond their control, the things which build up intransitively are negative and uniformly threatening” (Louw 2000: 52). This leads us to posit a semantic prosody of powerlessness, perhaps more accurately expressed as “inability to prevent a threatening situation from deteriorating further”. Yet although satisfactory as prosodies go, the only thing it reveals that the phrase’s compositional meaning does not is that there is a lack of control… and this is already accounted for in the semantic association with hopelessness. This observation is perplexing: the semantic prosody merely reinforces the information that the less abstract elements of the unit of meaning already convey.
Table 2 provides a summary of the analysis discussed. Subsections 4.2 to 4.5 investigate phrases whose semantic analysability becomes progressively more problematic, and where, as a result, semantic prosody can be seen to take on a more important role.
Circumstances or situations change (for the worse)
Inability to prevent a threatening situation from deteriorating further
Catch somebody in the act, although apparently as analysable as from bad to worse, is in fact a phrase in which underspecification of meaning plays a crucial role. While catch poses no difficulty to interpretation, as it is used in its salient sense here, the nature of the act is not encoded in the compositional word-meanings, making this an instance of a partially analysable idiom (catch + in what act?). Knowledge of this multi-word lexical item includes knowing that the act is not approved of, as the dictionary definition (below) confirms, so we must surmise that this knowledge comes from contextual information. The analysis will also show how underspecification interferes with the identification of the full unit of meaning, and the repercussions that this has on the identification of the semantic prosody.
If you catch someone in the act, you discover them doing something wrong or committing a crime. (COBUILD)
The definition provided above, being corpus based, reveals some of the information that the present analysis will illustrate, in particular that the act is likely to be a misdemeanour of some kind (“something wrong or committing a crime”). The nature of the act is revealed in the cotext, either by of plus present participle immediately following the phrase (7), e.g. caught in the act of + adultery / stealing/ pinching, or by mention of the crime or perpetrators before the verb (After my apples, eh? Well, I’ve caught you in the act). Lexical collocations are rather thin on the ground: thieves (2), burglars (2) and policeman (2) are the only repeated forms to be found in the 37 examples (see Appendix), although these are complemented by a range of single occurrences of criminal types (criminals, intruder, vandal) and law-enforcement officials (sergeant, investigators, officers). The phrase colligates overwhelmingly with the past simple form caught (30, against 3 infinitive, 3 present simple), with only one instance of a variant verb found in this data set (disturbed). Caught has a distinct preference for the passive (in over 80% of instances). At N+2 position, following the collocate of, the specified acts are always expressed with the present participle, it also colligates with third person pronouns, while only one proper noun is present. In common with from bad to worse, it colligates significantly with clause-final position (9 full stops, 11 commas, 1 exclamation mark, 1 parenthesis, plus three unpunctuated examples).
The semantic preference of catch somebody in the act is, as the dictionary definition above indicates, for crime – typically theft – and other wrongdoing – typically involving the breaking of moral or professional codes of conduct (including adultery, drug-taking, forgery, and the unauthorised acquisition of nuclear materials). The semantic associations for catch somebody in the act which emerge link together the semantic preference for unacceptable behaviour and the connotations which being caught conjure up. When one engages in illicit activity, one does so in full knowledge that it is wrong, but one hopes not to be found out. Being discovered, therefore, while inevitable in the long run, is a source of shame and embarrassment especially because of the fateful consequences it may have on one’s legal position or moral standing. Yet individuals rarely own up. There are no occurrences in the data of people admitting that they were caught – only reported accounts of others being caught, or nearly being caught, or managing not to be caught. So the perpetrator’s point of view, his or her loss of control and the negativity that would be associated with it (cf. Louw 2000), is not reflected in the semantic associations. We have instead the point of view of those who approve of maintaining law and order and upholding rules and regulations, because these people are the ones who have chosen to use the expression. If somebody is caught in the act, s/he will soon be getting his/her just deserts, and this gives the morally-upright commentator reason to gloat. From this the semantic prosody emerges: there is reason to rejoice over the (reported) discovery of wrongdoing in progress, and anticipation of the perpetrator’s (self-inflicted) downfall. Table 3 summarises the points made in this subsection.
As a final comment, it should be stressed that the ominous overtones associated with catch somebody in the act seem to hinge on the semantic associations surrounding catch. This can be stated with some certainty after examining all concordances of in the act. This is a phraseological fragment, in that does not express meaning in isolation but must combine with meaning-bearing words – the verbs which precede it and the actions to which it refers. Common to all the occurrences of in the act [of + action] in the BNC is a delaying effect, because the action is not named immediately after the verb (compare caught stealing with caught in the act of stealing), and this delay also emphasises the fact that the act is in progress because it is drawn out in the syntax as well as in fact. In combination with the semantic associations generated by catchsomebody doing something, the delayed onset and drawing out of the action brings it to the fore, highlighting the “crime” which an individual has been caught “committing”. The clause- or sentence-final position reinforces the sense of closure and finality already suggested by catch.
Past simple passive;
3rd person pronouns
Law-breaking, esp. theft;
Breaking of moral/ professional codes of conduct
Gloating over reported discovery of wrongdoing in progress; Anticipation of the perpetrator’s downfall
Table 3. Summary of unit of meaning with caught in the act as node.
4.3 Caught red-handed
Caught red-handed is closely related to caught in the act, as the analysis here will demonstrate. In a sense, caught in the act, with its underspecified element (the act), is a sort of superordinate phrase which finds greater specification in caught red-handed or caught in flagrante delicto: the former is typically associated with crime, the latter with adultery and other morally censurable deeds (Philip forthcoming).
Caught red-handed is a metaphorical idiom. Like all idioms it is noncompositional; it is located toward the transparent end of the continuum described in Section 3. The transparency of the metaphor red-handed gives an impression of analysability, but it is not fully analysable in the way that catch left-handed or catch one-handed are, because nominally literal catch is used in conjunction with non-literal, noncompositional, non-analysable red-handed. The phrase does not mean “catch with a red-coloured hand”, or “catch somebody whose hands are red”, but catch “[i]n the very act of crime, having the evidences of guilt still upon the person” (OED). This meaning cannot be computed from the meanings of the words except by using metaphorical inferencing to ascertain that red refers to blood, and that blood on the hands must be connected with murder rather than e.g. butchering, cooking or surgery.
That the hands might be blood-covered does not seem to be key to definitions of the phrase:
If someone is caught red-handed, they are caught while they are in the act of doing something wrong. (COBUILD)
But who is caught, and doing what? If we bear in mind that this phrase is closely linked to caught in the act, it comes as no surprise to find some overlapping features in the units of meaning. Similar collocates and colligates can be found, most notably groups of wrongdoers and groups of authority figures: thief (3), burglar (2), robbers,suspect, the accused; and police(man) (3), sergeant, the force, security services. Stealing (2) as well as lifting, pinching, and taking are more prominent in this data than in that for caught in the act, and there is an additional criminal type to be found: terrorist, and bomb (2). The phrase. Like its super-ordinate, has a marked colligational preference for clause-final position, and for the verb catch to be used in the past simple passive. Proper nouns are more common (six people are named), and personal pronouns are comparatively uncommon (5), general nouns (burglars, adults, an offender, the thief) being the preferred means of referring to the culprits. The semantic preference is markedly oriented towards crime, in particular theft and other financial irregularity, drug dealing and smuggling, and terrorism, and interestingly the perpetrators of these crimes are often “respectable” members of society (bosses, clients, shop steward, Hollywood stars) rather than belonging to the criminal classes.
The semantic associations of caught red-handed combine the meanings of caught already mentioned in Subsection 4.3, with the metaphorical extensions, or entailments, of hand and the finality and inevitability that clause-final position seems to lend. Hand is an incredibly productive metaphor, and can refer to anything that is held by or touched by the hand as well as recalling gestures which involve the hands; and it can also indicate control of self and others (MacArthur 2005). The presence of red in conjunction with hand may also trigger intertextually-derived meanings, which constitute a form of associative connotation (Kerbrat-Orecchioni 1977: 112 ff.). For example, previous study of Shakespeare might bring to mind the famous lines: “What, will these hands ne’er be clean? […] Here’s the smell of blood still: all the perfumes of Arabia will not sweeten this little hand” (Macbeth V/i), and so exploit intertextuality to create a mental picture, or “image schema” (Lakoff 1986) for the metaphor. Even though the phrase is not used in blood-related contexts, then, the activation of an image schema based on intertextually-derived connotations would reinforce the salient word meanings even if the phrasal meaning is delexicalised. This would have an influence on the semantic associations.
With all this semantic activity potentially going on around the core caught red-handed, what place does the semantic prosody have? It is very similar to that of caught in the act, which is unremarkable given the affinity between the phrases. But since the semantic prosody links the associations back into a real context of situation, we find that caught red-handed reports the discovery of others’ misdemeanours, usually when in progress, but also at the planning stages, or à fait accompli, as examples (1) and (2) illustrate respectively.
1. Police say they’ve caught a terrorist red handed on his way to plant a bomb.
2. They were caught red-handed as they carried part of a 200kg haul, worth £264,000 on the streets, into a warehouse near Madrid.
When people are reported as having been caught red-handed, their guilt is undeniable, and their punishment inevitable. The reason why this expression is used, therefore, is to communicate satisfaction that the criminal deeds have been stopped and that the person(s) responsible will not be able to avoid punishment. The data studied does not yield the same sense of triumphalism that was present in concordance lines for caught in the act, despite the fact that those who are caught red-handed are engaging in serious crime, not moral peccadilloes (this is the domain of caught in flagrante; see Philip forthcoming). What we uncover by comparing the semantic prosodies is that discovering and punishing criminal behaviour seems to be less emotive an issue than discovering and punishing those who break moral codes. Table 4 summarises the data described in this Subsection.
Satisfaction that a wrongdoer has been caught and brought to justice
Table 4. Summary of unit of meaning with caught red-handed as node.
4.4 The grass is greener
The grass is greener is another phrase that looks to be quite analysable, because of its semantic transparency, but is in fact non-analysable. Although the word string makes perfect sense when analysed into its literal, salient meanings, the phrase “means” differently depending on whether it is referring to grass which is green (example 3) or to the metaphor (example 4).
(3) South, where the grass is perceptibly greener, the River Derwent rushes past the glory of Chatsworth House and the charm of Haddon Hall.
(4) No matter what Mr Gorbachov tries to do to improve the consumer flow, the grass is always greener over there, and now that the fence is coming down the arrival of so many Russians in the West could provide an exciting new cultural mix in Europe.
It is the metaphorical value of the phrase which gives rise to these different meanings, metaphors being identified by the incongruity of their literal meaning within the context in which they appear. A gap opens up between the meanings suggested by the words and the meaning actually expressed by the phrase, and this gap has to be filled with the additional information which allows the phrase to be understood meaningfully in context. When the metaphor consists of a phrase rather than a single word, the compositional meaning is rejected in favour of a contextually-relevant meaning. This essentially pragmatic meaning involves the string as a whole, thus diminishing the influence that the distinct word forms can exert semantically. It has been recognised that units of meaning associated with metaphors – metaphoremes – must obligatorily have a pragmatic function (Cameron and Deignan 2006). In corpus linguistics terminology, that pragmatic element is the semantic prosody.
The grass is greener can indeed be read as a literal statement, but the metaphorical meaning, being more frequent and presumably more familiar as a result, is the more salient of the two (see Giora 1999). It is defined thus:
If you say the grass is greener somewhere else, you mean that other people’s situations always seem better or more attractive than your own, but may not really be so. (COBUILD)
For this analysis, a total of 21 relevant concordances were studied, with BNC data being supplemented by corpus data available online; the unambiguously literal meanings were not taken into consideration for this analysis and thus do not appear in the Appendix. Recurrent collocates fall into two groups – the optional adverbial always (present in one third of the examples), and the (again, optional) completion of the comparison with a locative, which has two standard manifestations (on the other side (4) and over there (1), the latter finding a range of alternative lexical realisations (over the hill, somewhere else, in Cheshire, outside the Wimbledon gates). The most noticeable colligational preference is for the phrase to occur in a projected that- clause (6, plus a further three where that is ellipsed), and the phrase, when completed by the locative, is clause- or sentence final.
The semantic preference is characterised by locations other than where one finds oneself, though sometimes the location is displaced in time (i.e. same place, different time). This gives rise to the unfavourable comparison between one’s present situation or circumstance and that of others (or time past). The phrase occurs within verbal and mental processes (say, said, tell, think, know, decided, find,) which may be found in nominalised form (the discovery that, a feeling that, living proof that). The semantic association follows seamlessly from these semantic preferences: the speaker feels disgruntled because others find themselves in more favourable circumstances, these circumstances being displaced in space or time. It should be borne in mind that the contrast may well be exaggerated. In addition to these lexically-derived associations, there is the possibility that the phrase will conjure up an image schema involving well-kept suburban lawns and invoking the pettiness of neighbourly rivalry which is often the fruit of misunderstanding and characterised by imaginary or unintended slights, keeping up appearances, and competitive one-upmanship.
The semantic prosody of the grass is greener hinges on the contrast between fact and opinion. It expresses disgruntlement, indeed, but disgruntlement which is perceived by others as unjustified because no objective motivation underlies it. Thus the sense of grievance which is being expressed is perceived by others as whinging and petty. Needless to say, given this prosody first-person use of the phrase is rare (one example found in the data).
Perceived inequality creates (unjustified) sense of grievance
Table 5. Summary of unit of meaning with the grass is greener as node.
4.5 Cold turkey
The final case study to appear here, illustrating the most opaque, semantically non-analysable of noncompositional phrases, is cold turkey. Unlike the previous case studies, where it has always been possible to analyse the meaning in part, cold turkey is non-analysable. Nothing in the salient or metaphorical meaning of either words would give the reader any reason to imagine that the expression can be glossed as “give up suddenly”. The word-semantics are thus redundant and yield entirely to the pragmatic reality of use, both textual and contextual. The context provides a vague, undefined sense of what the intended meaning is, as will be seen in the analysis to follow, and in the first three columns of Table 6, but the complete meaning relies on knowledge of the semantic associations and semantic prosody. Without these components, the phrase’s meaning is incomplete, the upshot of which is that it cannot be understood fully from the words alone. Similarly, it cannot be used effectively if the speaker is unaware or unsure of what these “additional” components of meaning are. At this end of the continuum, therefore, pragmatic meaning takes hold.
Cold turkey is the unpleasant physical reaction that people experience when they suddenly stop taking a drug that they have become addicted to. (COBUILD)
A total of 26 concordance lines for cold turkey were found, again after supplementing BNC data with the 56m Wordbanks Online sampler and eliminating the six lines which referred to the food sense. Idiomatic cold turkey collocates significantly with the verb go (go (3), went (3)), and with verbs relating to discontinuation (come off (2), quit (2), weaned off, stop), this latter group collocating with a range of drugs (drugs, heroin (2), barbs, sleeping pills), drug-taking (smoking (2), drinking) and drug users (tobacco users, smokers, alcohol, addict). The lemma addict (addiction (3), addict, addicted) is also prominent, as is withdrawal. The final recurring collocate is cold turkey: the idiom collocates twice with itself, accounting for four of the 26 occurrences.
This wide range of collocates is not paralleled by a similarly wide range of colligates. The idiom colligates with scare quotes (6 times) and with clause-final punctuation (7). When used as noun modifier, the preference (3 out of 5) is for the rather unusual post-positive position, as in detoxing cold turkey (line 17), but not frequent enough to be considered a strong colligate. The semantic preference is characterised by addictive drugs, both legal and illegal, which itself attracts a preference for illegality through the presence of inmates, cops and jail. The semantic preference also lists a range of unpleasant withdrawal symptoms, from Delerium tremens to stomach cramps, gastric flu, the shits and worse (line 19).
It is from these preferences that the semantic associations of suffering and anguish emerge, brought on by the suddenness with which the dependency on the drug has been terminated (soothing his cold turkey, feels like a bad bout of influenza, the worst three days of my whole life). Unlike catch red-handed and the grass is greener, it is difficult to envisage how any relevant image schema could be formed to reinforce the meaning or semantic associations connected with cold turkey; elaboration of memories of eating left-over chicken for days after a big feast such as Thanksgiving or Christmas fails to communicate the dramatic effects of going cold turkey, and it is only by backtracking that one could connect cold with chills or other side-effects.
Finally we come to the all-important semantic prosody, which is expressed through hypothesis and conditionals, and can be stated as “reluctance”, or, more completely, “decision to regain control over addictive substance, but reluctance to suffer the side-effects”. Whenever there is a choice to be made between abruptly giving up an addictive substance and gradually weaning oneself off it, the choice made is usually the former, though it is not a choice that is made easily, as can be appreciated from examples 5 and 6.
(5) But instead of allowing himself to be weaned off the drug, he went “cold turkey”. It affected his mind, or so it was said.
(6) The biggest fear was detoxing cold turkey because the cops wouldn’t put me in a hospital, they’d just let me go cold turkey and I’d probably die in jail.
Perhaps it is the semantic prosody that connects the idiomatic and literal readings of cold turkey. After all, unlike its hot counterpart it consists of leftovers which have to be finished rather than a dish that is in itself prepared specially. It is not eaten with particular enthusiasm, but rather with reluctance and sense of duty, something that is done in the belief that, in the long run, it is for the best even if short-term suffering has to be endured. Table 6 summarises the analysis.
Come off, quit,
Scare quotes; clause-final punctuation
Noun modifier (post-positive position)
Decision to regain control over addictive substance, but reluctance to suffer the side-effects
Table 6. Summary of unit of meaning with cold turkey as node.
5. Conclusions: semantic prosody and the idiom principle
The case studies shown in Section 4 may have laboured the point somewhat, but semantic prosody is present, or tangible, in different concentrations depending on how great the need is for word meaning to be supplemented by pragmatic meaning in language-based communication. The meaning of compositional and semantically analysable language is wholly accounted for by word meaning. Compositional language by definition is not patterned, because it has no need of patternings to fix meaning which is “inherent” in the words themselves. As idiomaticity increases and semantic analysis becomes more difficult, the semantic prosody assumes greater importance.
Knowledge of how a given linguistic item is used, and thus what it means, stems from stored memories of previous encounters with the item in text and context. This knowledge is then exploited when using the same linguistic item, thus perpetuating the patternings and ensuring continued success in communication. Through such repetition, marginal nonsalient meanings can be fixed in stable wordings which reinforce those meanings with each subsequent occurrence. Yet the pragmatic component of a unit of meaning is abstract and complex, as can be seen from the descriptions of the semantic prosodies in Tables 2 to 6.
The subtleties of semantic prosody and the difficulties to be found in its identification and labelling are exacerbated by the fact that it is not expressed by reiterated word forms, but by the recurrent presence of particular nuances which emerge from a wide variety of wordings. The abstract nature of this repetition makes it difficult to pinpoint precisely where the nuances lie and how they are formed, but they can be identified in corpus data. It is easiest to identify those prosodies which express meanings that the words themselves do not suggest, because of the contrast which results. However, it is not the role of semantic prosody to contradict word meaning, but rather to integrate it into human interaction. In some cases, therefore, the semantic prosody seems absent (from bad to worse), though it may in fact be reinforcing and harmonising with the meaning already suggested by a compositional reading.
Although the view has been contested (especially by Whitsitt 2005), the emergence of semantic prosody is likely to be attributable to historical processes (Louw 1993). There are many possible reasons for this, but ultimately deliteralisation – the process by which literal meanings become less concrete and eventually figurative because their referent is no longer immediately retrievable – is responsible. As a result of deliteralisation, the meaning expressed by a wording eventually parts company with the meaning expressed by its component words and what they normally refer to. A communicative gap ensues, and this is filled by pragmatic information regarding when and where the language is used and for what purpose. This becomes encoded as part of the wording in its extended unit of meaning, thus cementing the relationship between a phrasal pattern and a functionally complete meaning.
The importance of semantic prosody in communicating meaning and in the “renewal of connection with the processes and patterns of life” (Firth 1968: 14) can be difficult to appreciate when the language is familiar, its contextual patterns and contexts of situation already stored in memory. When a phrase becomes obsolete, however, knowledge of its function in context is lost. Awareness of the meaning of the words, of who says them where and when, may not suffice to understand fully the meaning that they convey.
By way of illustration, this paper ends with an extract from a historical novel (O’Brien 1970). The author went to great lengths to remain faithful to the colloquialisms used by sailors in the late 1700s, not only researching the language but also reproducing verbatim extracts from letters written to family and friends. The idiom illustrated is wet the swab (example 7). It is not enough to understand the words nor the participants in the context of situation: the semantic prosody holds the two together, and when it is absent, so too is the “real”, complete meaning.
(7) We do not keep fashionable hours in the service, and I grow so devilish hungry and peevish by then that you will forgive me, I am sure. We will wet the swab, and when it is handsomely awash, why then perhaps we might try a little music, if that would not be disagreeable to you. (O’Brien 1970: 17)