I

Idioms Idioms are a class of multi-word units ‘which pose a challenge to our understanding of grammar and lexis that has not yet been fully met’ (Fellbaum et al. 2006: 349). They are commonly believed to be qualitatively different from ‘normal’ language, but the precise nature of this difference can be elusive. Even amongst idiom scholars, it is difficult to find a consensus as to what precisely is, or is not, an idiom, because of the heterogeneity of the class. There is widespread agreement on one gen-

eral principle: an idiom is an institutionalised expression whose overall meaning does not correspond to the combined meanings of its component parts. However, this criterion can be said to apply to a wide range of phraseological structures, such as collocations, formulaic greetings, clichés and other conventionalised expressions which, although idiomatic to some extent, are not idioms in the strict sense. The challenge for idiom researchers is therefore to formulate a definition which is flexible enough to include all known idioms, yet exclude non-idioms such as those mentioned above. An idiom is composed of two or more con-

stituent parts, generally deemed to be words, although Hockett (1958: 177) admitted phonemes as constituents and Makkai (1972: 58) morphemes. Despite appearances to the contrary, each of these words does not contribute to the overall meaning of the phrase, which operates as if it were a lexical item in its own right and expresses a semantically complete idea which may be quite independent of the meanings of its

anomaly derive mainly from the fact that an idiom is not built up word by word, according to the grammar of the language, but is a noncompositional phrase which is learned, stored and recycled as a single chunk. Current psycholinguistic views support the

argument in favour of considering idiom as a type of ‘long word’ whose meaning is accessed directly and not through prior decomposition or analysis of the constituents (Gibbs 1994, 2002). However, when an idiom is encountered for the very first time, language-users have no choice but to decipher its meaning from the meaning of the constituents, usually doing so by taking into account the most salient meanings first (Giora 1997, 2002; Peleg and Giora 2001). That this tactic enjoys a limited success rate is due to the difficulty in identifying which meaning of polysemous components is relevant and the extent to which the idiom is semantically motivated or transparent. The ease with which an idiom can be inter-

preted is based on its level of semantic transparency as well as truth conditions and other contextual cues. A transparent idiom yields its meaning easily, because there is a straightforward connection between the phrase and the intended meaning. For example, not see the wood for the trees (‘to lose oneself in details and fail to see the larger picture’) requires little semantic re-elaboration; it is therefore located towards the transparent end of the scale. On the other hand, an expression which has a more arbitrary relationship with its meaning, such as to go cold turkey (‘suddenly stop taking a drug that you have become addicted to’), can be described as or

opacity of an idiom cannot be measured in absolute terms, as it is affected by the individual’s real-world knowledge, awareness of cultural norms, and general familiarity with the phrase. The more closely the wording of an idiom

reflects a real-world situation, the easier it is to interpret: make one’s blood boil reflects the heat felt in the body when enraged; to bite the hand that feeds you can easily be connected to ingratitude. In much the same way, an idiom which refers to a culturally familiar situation poses little difficulty to interpretation: knowledge of team sports reveals the principles of equality and inequality respectively encoded in a level playing field and move the goalposts. It is also true that an idiom which is familiar to the hearer is perceived as being more transparent than one which is not so familiar, regardless of its real-world or cultural relevance: like a red rag to a bull (‘a provocation’) is much less frequent than make sb see red (‘provoke or anger sb’) (Philip 2000), and therefore requires more effort in decoding. Finally, it is worth noting that, as with all figurative language, even transparent idioms pose problems for language learners who, lacking the necessary linguistic and cultural knowledge to decipher them, are apt to interpret them literally. While some idioms dovetail into our concep-

tual system, not all do, and one well-documented feature of idioms is their adherence to, or violation of truth conditions. When a phrase alludes to events or situations that cannot possibly occur in the real world, a literal interpretation is incongruous: human blood is always red (blue blooded), kitchen implements do not speak to each other (the pot calling the kettle black), and animals do not fall from the sky as precipitation (rain cats and dogs). In situations such as these, the only way to make sense of the meaning is to treat the expression as idiomatic. Not all idioms violate truth conditions, and many phrases can, at least theoretically, be read literally or figuratively depending on which interpretation best fits the context in which the phrase appears. A great deal of psycholinguistic literature

deals with the effects of context on the interpretation of phraseological homophones – idioms which can have both literal and idiomatic readings. Here context is textual, not pragmatic,

designed to sway the reader’s interpretation towards an idiomatic or a literal meaning (for an example of this, see Giora and Fein 1999: 1605). Outside experimental conditions, contextual cues are particularly important in determining the meaning of idioms whose literal and figurative meanings are either not well established or occur with relatively low frequency: the phrase cherry picking may be used literally or figuratively, but its location in a text on blue-chip business would be incongruous if read literally, thus triggering its idiomatic reading (‘being selective’). Recent corpus-based research into homonyms

suggests that context is less crucial than previously believed, and that one of the possible readings usually predominates. According to Hoey (2005: 82ff.), it can be argued that language users will avoid using a familiar idiom in a context where it could be interpreted literally, preferring instead to paraphrase or use an alternative expression. Thus, under normal communicative conditions, a person who is literally skating on ice which is thin would not be described as skating on thin ice; and if a person who hits a bucket with their foot is described as having kicked the bucket, humour automatically ensues because of the clear mismatch between the more familiar, idiomatic meaning and the literal description of events. Idioms are learned and reused as single lexical

items, yet they are not single words. While the canonical form of an idiom (the citation form used for dictionary definitions) is fixed for the purposes of language description, the reality of language in use is that most idioms can undergo a controlled amount of variation to their typical realisation. There is some divergence in opinion on this point between theoretical and descriptive studies on idioms. Pre-corpus scholars defined idioms as being fixed or frozen in form, in reference to the fact that they resist morphosyntactic change; now it is more common to find them described as stable (Cˇermák 1988) or of limited flexibility (Barkema 1996: 128). This difference in terminology is due to the fact that much pre-corpus literature on idioms deals only with what is theoretically possible, with the result that the categories and principles devised, while extremely detailed and rigorous, fail to reflect adequately the attested behaviour of idioms in data,

notably Moon (1998), have challenged the notion of fixity in light of the observation that most idioms do in fact allow variation to occur, so long as some vestige of the canonical form survives. Demonstrating the syntactic and semantic

stability of idioms has been one of the prime considerations of figurative language scholars, especially those working within the generativist tradition. Idioms are said to be transformationally deficient, and in order to prove the case that non-canonical realisations of idioms cause their meaning to revert to literal, they can be subjected to a series of tests. The tests adopted fall into two broad categories: lexical and grammatical. The lexical tests include the augmentation test (addition of lexical constituents), the elimination test (deletion of constituents), the substitution test (replacing a constituent by a semantically related word), the permutation test (rearranging constituents whose order is fixed). The grammatical tests include blocking of predication, blocking of the formation of comparative and superlative forms of adjectives, blocking of nominalisation and blocking of passivisation (Gläser 1988: 268-9). As Gläser explains, ‘[a]s soon as these practical procedures are followed, the resulting construction will be grammatically correct and empirically sensible, but it will cease to be an idiom’ (1988: 268). Transformation tests do not stand up well to

empirical scrutiny. Even before the widespread use of computer corpora, criticisms were levelled against this method of idiom classification, because it fails to look beyond the tested phrase and compare its behaviour to similar structures or semantically related language items. Chafe (1968: 122) argues that the blocking of passivisation can be explained by the underlying meaning of an idiom, not its idiomaticity. Citing kick the bucket, he points out that the literal equivalent die would similarly fail the passivisation test (*to be died). The other transformation tests do little better, and are of limited relevance to those idioms which have no literal homonym (hue and cry, in fine fettle, run amok). The availability of large, electronically search-

able linguistic corpora has allowed idiom scholars to put transformations and other theoretical considerations to the test. Corpus-based studies

widespread phenomenon, not one restricted to the creation of special linguistic effects such as punning, humour and irony. In Moon’s (1998) study of fixed expressions and idioms in an 18-millionword corpus, attested lexical andmorpho-syntactic variation is described in detail (1998: 75-174). Moon reports that approximately 40 per cent of the idioms and other fixed phrases studied occurred in a variant form (1998: 120). However, the larger the corpus is, the more variation occurs; in some cases the canonical form can be outnumbered by its variants (Philip 2008: 103). Even if idioms are not fixed, they do have a

stable form which is learned as a multi-word lexical item. This canonical form is subject to exploitation in the normal course of language use, and so idioms can appear with lexical and grammatical alterations, in truncated and augmented forms, and in phrases which merely allude to the original: ‘Talk about Mr Pot and Mr Kettle?’ (the pot calling the kettle black; Philip 2008: 103). The rules governing such exploitations have yet to be determined, but are believed to be predominantly conceptual and semantic in nature. It has been established that figurative expres-

sions are not merely colourful add-ons to the lexicon, but that they contribute to its evaluative inventory (Carter 1997: 159). Simply put, idioms have a literal counterpart in the language, but this counterpart is not a true synonym because it fails to express the evaluative meaning encoded in the idiom. Cˇermák (2001: 13) notes that ‘idioms are a primary means for the expression of positive and negative attitudes’, but goes on to lament the fact that little research has been carried out into the matter. Idioms resist pigeon-hole definitions because

they constitute a heterogeneous class of anomalous lexical items. In order to understand them fully, it is necessary to understand better the mechanisms at work in ‘normal’ language, and here, too, corpus analysis is challenging traditional descriptions. Idioms are less fixed than used to be believed, ‘normal’ language less free.