1.1 Some Terminological Distinctions

A rough list of the different forms of idioms and formulaic language mentioned in the literature includes the following categories, here illustrated with English examples for convenience (Gibbs, 1994, 2007; Sailer, 2013):

(1)

a.

Sayings (e.g., take it easy, come on!).

b.

Proverbs, which describe recurrent situations (e.g., a bird in the hand is worth two in the bush, a stitch in time saves nine).

c.

Phrasal verbs (e.g., come in, take off).

d.

Binominals, typically of the form X-Conjunction-Y (e.g., by and large, black and white, spick and span, more or less) or of the form Noun-Preposition-Noun, which conveys plurality (e.g., day by day, student after student, face to face).

e.

Phrasal compounds (e.g., red herring, dead-line, accident-prone, shopping center).

f.

Formulaic expressions (e.g., at first sight, once upon a time).

g.

Lexical bundles, identifiable statistically and by intuition as recurrent sequences (e.g., a little bit of, you don’t have to).

h.

Collocations, or conventionalized co-occurrences including those formed by light verbs followed by specific object nouns (e.g., infinite patience, a hard frost, do a favor, give a look, take a step).

i.

Idioms, formulaic constructions whose interpretations are unpredictable from individual lexical meanings under the effects of regular compositional rules. Different types of idioms have been also identified in the literature:

Quasi-idioms, whose meaning includes the meaning of the lexical components plus an additional non-compositional meaning (Mel'čuk, 1995) (e.g., give the breast, start a family).

Phrasal idioms (lexically headed) versus clausal idioms (headed by a sentential functional head: a fixed tense or mood, a modal, obligatory (or impossible) sentential negation, or CP-material such as a complementizer or a wh-phrase) (Horvath & Siloni, 2016, 2017) (e.g., land on one’s feet vs. can’t hold a candle to someone/something).

idiomatic phrases (which do not distribute their meanings to their components) versus Idiomatically combining expressions (whose meanings—while conventional—are distributed among their parts) (Nunberg, Sag, & Wasow, 1994) (e.g., saw logs ‘to sleep’ vs. pull strings ‘to exploit connections’).

Constructional idioms, which are not lexically fixed in full and whose meaning is associated with the whole construction (Jackendoff, 1997b, 2008) (e.g., ‘the way construction’, ‘the X-er, the Y-er’ comparative correlative construction, ‘the V time-away construction’, ‘the V one’s part-of-the-body out/off construction’, ‘the X after X construction’ (cf. formal or lexically open idioms, Fillmore et al., 1988).

These terms do in no way represent a systematic taxonomy of multi-word units, and in fact describe different sets of idiomatic expressions. Thus, semantic phrasemes (following Mel'čuk, 1995, p. 179) can on the one hand be classified as idioms, collocations, and quasi-idioms. On the other hand, while all idioms are collocations, insofar as they are conventionalized or fixed, most collocations have been claimed not to be idioms insofar as they are compositional (Larson, 2017, p. 400).¹ Additionally, some formulaic expressions have been analyzed as grammatical constructions (e.g., let alone, Fillmore et al., 1988), but not all grammatical constructions can be considered members of formulaic language (e.g., ‘the X-er, the Y-er’ comparative correlative construction, Borsley, 2004; Dikken, 2005; Abeillé, Borsley, & Espinal, 2006).

Furthermore, it should be noted that the mere existence of this typology derives from the various dimensions on which idiomaticity lies (Nunberg et al., 1994):

Finally, it is worth pointing out that, even though idioms are often claimed to be characteristic of informal language and popular speech, most characteristically they are very specific to particular communication systems or registers. Furthermore, they have been said to be “especially useful in terminating a topic because of their distinctive manner of characterizing abstract themes in concrete ways” (Gibbs, 2007, p. 703), which means that they have a significant role in a theory of communication and cognition (Sperber & Wilson, 1995; Vega-Moreno, 2003). As such, idioms also reflect the interaction of the system of language with the emotion system, a system that deals with the assignment of values, positive and negative, including emotions, attitudes, and opinions to some entity, event, or situation (Foolen, 1997).

2.1 The Categories of Idioms

Studies on idioms usually suffer from two main flaws. Most studies deal exclusively with idioms in English, and most of them deal only with verbal idioms (those whose head is a verb; cf. Marantz, 1984; Fillmore et al., 1988; Nunberg et al., 1994; Croft & Cruse, 2004; Svenonius, 2005; Evans & Green, 2006, a.o.). However, it has to be pointed out that some attempts exist to study idioms both from cross-linguistic and cross-dialectal perspectives (Harwood et al., 2016; Markantonatou & Sailer, 2018).

Unfortunately, from a cross-linguistic perspective one can find basically lexicographic studies that provide lists of form-meaning pairs for various sorts of lexicalized multi-word expressions. Some of these studies contain in addition to the collection of idioms and their conventionalized meaning information regarding the syntactic manipulations each idiom allows, based on single (or few) speakers’ judgements.

A second common failure extends to the syntactic category of the idioms studied, usually verbal, and sometimes adverbial and/or prepositional.⁴ We hereby wish to remark that idiomatic expressions are multicategorial, depending on whether the head of the idiom is either a lexical or a functional category, with the result that idiomatic expressions can be associated with either VP, NP, AP, AdvP, and even with PP, DP, QP, ConjP, CoordP, and S/TP.⁵

2.2 Syntactic Properties

In this section four issues will be discussed: the internal versus external syntax of idioms (Harwood et al., 2016), the syntactic flexibility and referential availability of DPs (Nunberg et al., 1994; Harwood et al., 2016; Mateu, in press), the continuity constraint (O’Grady, 1998), and the selection constraint (Bruening, 2010, 2017).

It is believed that the internal syntax of idioms is built up through the same regular, compositional, structure-building mechanism that is at the basis of non-idiomatic structures (cf. Fellbaum, 1993; Nunberg et al., 1994; Svenonius, 2005, a.o.). In particular, it has been assumed that, if the object of study is a verbal idiom consisting of a V + a direct object DP (e.g., English spill the beans, kick the bucket), manipulations on the internal syntax of these idioms may consist on zooming in at the level of the DP structure, and exploring the need of the definite article as a contributing trigger for idiomaticity, as well as the possibility of using other determiners (indefinites, demonstratives, possessives, quantifiers, etc.) without altering this idiomaticity.

In relation to this issue, the distinction of Nunberg et al. (1994) between ICEs and IPs has become extremely relevant. An ICE such as spill the beans is characterized by being syntactically flexible and by having a referential DP, which means that the definite article can be replaced by other determiners and that the use of either a definite or an indefinite article constrains the referential properties of the DP while keeping its idiomaticity. Syntactic flexibility and referential availability allow not only for the change of the D, but also for modification, topicalization, and for the mapping of the complement noun into a figurative meaning. By contrast, an IP such as kick the bucket is characterized by not being syntactically flexible and by not having a referential DP, which means that in this case the definite D is not associated with a referential interpretation. Lack of syntactic flexibility and lack of reference go hand in hand, in such a way that only the idiom as a whole, but neither the full DP nor the complement noun, can be mapped onto a figurative meaning.

This distinction could be analyzed as postulating for an ICE such as spill the beans, which can be associated with both a literal, compositional interpretation, and with an idiomatic reading ‘divulge the secrets’, an internal syntax with a DP structure. By contrast, the internal syntax of an IP such as kick the bucket, which can be associated with both a literal, compositional interpretation, and the idiomatic meaning ‘die’, would project a DP layer only in the literal interpretation, whereas in the idiomatic reading the complement of the verb would be of the category N. Hence, the definite D in IPs has been said not to introduce a phase and to be semantically vacuous or expletive, so that the nominal expressions in object position have been analyzed as forming a complex predicate with the verb (Espinal, 2001).

The external syntax of idioms is relevant as far as it has been claimed that idioms represent opacity domains (i.e., phases) and, therefore, that there is a size limitation to idioms. In relation to this issue it has been claimed that all parts of a phrasal idiom must be minimally dominated by the same phrasal node (vP, CP) (Svenonius, 2005). Thus, it has been asserted that verbal idioms are composed of material from the vP domain and cannot be composed of material from the TP domain, the general idea being that while material in the TP domain (i.e., Tense, Mood, Aspect, Voice) can embed idioms, verbal idioms are never dependent on such material.⁶

According to this analysis of the external syntax of idioms, which stems from Chomsky (1980, 1981) and Marantz (1984, 1997), verbal idioms consist of verbal predicates and their arguments, as represented in (1) and (2) (cf. Harwood et al., 2016). Notice that idiomatic patterns consisting of a V + DP theme and a subject + VP are both analyzed as vP.⁷

(1)

(2)

Following this analysis, all verbal idioms, even the ones that look like sentential idioms, are generated within the vP domain: verbal idioms are constrained by the clausal-internal phase boundary, which is equivalent to saying that there are locality constraints on idiomatic constructions.

Let us now focus on a much discussed constraint that has been claimed to regulate the internal syntax of idioms, the so-called continuity constraint (O’Grady, 1998), which deals with idioms in terms of dependency relations.

O’Grady shows that, although it has frequently been suggested that idioms might form constituents at some level of representation, this hypothesis finds several counterexamples:

Therefore, the conclusion (against Fraser, 1970) is that idioms need not form constituents, but still they are subject to an important grammatical constraint: a head licenses its dependents (i.e., their arguments, modifiers, specifiers) in that its syntactic and semantic properties determine the number and/or type of other elements with which it can or must occur within a local syntactic environment. Consider the continuity constraint in (3) (from O’Grady, 1998, p. 284, (12)).

(3)

The proposed analysis reduces idioms to a continuous chain of head-to-head relationships, independently of the categorical status of the idiom (VP, S, etc.).⁸ This approach seems to have a number of advantages: (i) it straightforwardly accounts for the existence of non-constituent idioms (e.g., get one’s goat), (ii) it also accounts for the optional co-occurrence of some idioms with adjectives, quantifiers, and other types of modifiers (e.g., kick the (filthy) habit, pull (yet more) strings), and (iii) it allows handling discontinuous idioms in general (e.g., bring (something) to light, lead (someone) a merry chase).

Besides the continuity constraint, a principle of idiomatic interpretation has been proposed in the literature (Bruening, 2010, 2017), whose basic idea is that two syntactic constituents, X and Y, can be interpreted idiomatically only if one selects the other.⁹ In addition, Bruening (2010, p. 532, (25)) postulates the following constraint:

(4)

If X selects a lexical category Y, and X and Y are interpreted idiomatically, all of the selected arguments of Y must be interpreted as part of the idiom that includes X and Y.

This hypothesis starts from the assumption that non-selected elements, like adjectives, possessors, and the like, can appear in between pieces of idioms because they do not disrupt this selection. Adjectives and adverbs may be part of an idiom (e.g., beat a dead horse), because they select for the projection they adjoin to. Unlike the continuity constraint approach, according to which adjuncts are dependent constituents on nominal or verbal heads, the selection approach postulates that adjunts select the categories they adjoin to. Thus, in beat a dead horse the verb beat selects an NP headed by the N horse, and dead selects a nominal projection also headed by horse.

Lexical selection can specify a single lexical item, a list of lexical items, or even a semantic class of lexical items, a range of possibilities that is particularly important in the case of idioms that allow some degree of lexical substitution of the selected idiom item.

Furthermore, according to this hypothesis idioms can cross the vP boundary by involving modals, aspect, negation, imperatives, and other functional material. Hence, it is hypothesized that, in addition to lexical material inside the VP, idioms can also cross the CP phase boundary. Thus, an idiom such as the cat (has) got (someone)’s tongue ‘Someone is speechless’ is claimed to involve not only a verb get, a subject cat and an object tongue, but also a matrix Voice head that selects an external argument (the NP headed by cat) and a VP headed by get, as represented in (5) (Bruening, 2017, p. 188, (15)). Notice that in this representation, and coherent with the idea that the selection constraint applies only to lexical categories (see (4)), functional elements such as Determiners and Tense are excluded.

(5)

To sum up, in this section the main syntactic properties and constraints that apply to idioms have been synthesized. Still, a question remains: from a syntactic perspective what is an (im)possible idiom? This issue is addressed in section 2.3.

2.3 Possible and Impossible Idioms

The study of the syntax of idiomatic expressions has become relevant not only for identifying permissible idioms but also for predicting those sequences that are considered impossible idioms, because they violate some syntactic requirements. “Possible idiom types tell us about possible selectional configurations, non accidentally non existing idiom types tell us about impossible selectional configurations” Sportiche (2005, p. 79). Thus, there are no V + D idioms to the exclusion of the complement N (*kick the N), there are no V + N idioms to the exclusion of the P that selects the N (*fly P the handle), and there are no subject + object idioms to the exclusion of the lexical verb that connects the two (*the shit V the fan). Idioms have to meet this structural description (Sportiche, 2005, p. 81):

(6)

According to O’Grady (1998, pp. 287–289), the continuity constraint also makes specific predictions about the types of idioms that cannot occur:

According to a selection-based approach to the syntax of idioms (Bruening, 2010, 2017), even though no syntactic locality constraint is assumed to regulate the syntax of idioms, there still exist syntactic constraints on idiomatic phrases. The first such constraint is imposed by the Principle of Idiomatic Interpretation, which—as previously postulated—imposes selectional relations between all items of an idiom. This principle, which rules out discontinuities in idiomatic phrases: “where X selects Y and Y selects Z but only X and Z are part of the idiom” (Bruening, 2017, p. 189), predicts that an idiom such as *Someone has V-en an old shoe, with the main verb free, is impossible.

The second constraint that idioms are subject to is the Constraint on Idiomatic Interpretation, also previously formulated, which captures the fact that there are many verb-object idioms, with an open slot for the subject, but there are by far less subject-verb idioms with an open slot for the object (e.g., a little bird has told me that . . .).

According to Bruening, selection takes place exclusively between lexical categories (V, N, A, Adv). However, against this approach one might consider not only those idioms that have functional elements as part of idioms, but also those idioms whose head is a functional category (e.g., D, Conj, P, etc.). These idioms abound in many natural languages. Consider, for example, the Catalan la mare del Tano lit. the mother of.the Tano [expression of surprise], i ara! lit. and now [expression of disapproval], de gom a gom lit. from gom to gom ‘full’ (Espinal, 2004a); in spite of the fact that neither the definite article, nor the conjunction, nor the weak preposition are lexical categories, these functional categories together with their selected arguments (the noun mare ‘mother’, the adverb ara ‘now’, and the nonexistent word gom beyond that idiomatic expression) are part of the idioms, which leads us to conclude that functional categories should not be excluded from possible idioms.

Additional restrictions on possible idioms affect (i) the number of subordinate clauses, which must be restricted to one (Catalan No tenir on caure mort lit. not have where fall dead ‘poor’, no saber quin dia menja pa lit. not know what day eats bread ‘ignorant’), and the obligatory presence of a dative clitic (Catalan aixafar-li la guitarra (a algú) lit. crash.cl_dat the guitar to somebody ‘disrupt’).

Let us finally consider asymmetries in idioms built with ditransitive verbs (Bruening, 2010; Larson, 2017). Bruening observes that V + someone + NP (e.g., give (someone) the boot) are common, but there is no V + NP + someone (*throw the wolves someone), the reason being that the second object is what is selected by the V, while the first object (the experiencer/goal) is selected by an Applicative head. “If an idiom includes the first object, it necessarily includes Appl selecting that object and V, but then by [the Constraint on Idiomatic Interpretation] the argument of V must be included as well” (Bruening, 2017, p. 190).

In turn Larson (2017) argues that in English there are no dative idioms at all, of either the oblique or the double object variety. According to him structures of the sort give + someone + NP are not idioms because they appear to be fully compositional: give is always interpreted as conveying caused possession, and the figurative meaning is linked exclusively to the contribution of the final NP and is not a property of the construction as a whole. Similarly, expressions such as show + someone + the door should not be considered an idiom either, because the meaning of the verb is basically one of caused motion. “Putative double object idioms are in fact not idioms but collocations (. . .). Putative PP dative idioms either aren’t datives (rather, they are caused motion constructions) or aren’t ditransitives” (Larson, 2017, p. 424).

2.4 Idioms as Constructions

The previous discussion leads to some theoretical implications (Harwood et al., 2016): (i) idioms in general and verbal idioms in particular are larger than simply the verb and its arguments, as standardly assumed, and (ii) the size of verbal idioms is subject to cross-linguistic variation, sometimes including vP internal subjects (e.g., the shit hit the fan), sometimes including adjuncts (e.g., take the bull by the horns).

It has been claimed that idioms are a type of construction (Nunberg et al., 1994, p. 507; O’Grady, 1998, p. 290). “Constructions may be idiomatic in the sense that a large construction may specify a semantics (and/or pragmatics) that is distinct from what might be calculated from the associated semantics of the set of smaller constructions that could be used to build the same morphosyntactic object” (Fillmore et al., 1988, p. 501). Constructions, therefore, are form-meaning pairs that associate syntactic structures with semantic and pragmatic meanings not mediated by the pieces they are composed of.¹⁰

According to Jackendoff (2002, 2008) constructional idioms (constructions with a (partially) non-compositional meaning, of which not all terminal elements are fixed) can either have canonical syntax (e.g., Jim joked his way out of the meeting, That fool of a politician; Booij, 2002) or idiosyncratic non-canonical syntax (e.g., let alone, the more . . . the less, student after student; Fillmore et al., 1988). Still, what is more relevant from Jackendoff’s Parallel Architecture approach (Jackendoff, 1997a, 2007, a.o.) is that it can account for items that most characteristically do not have phonology, morphosyntax, or semantics, as in (7):

(7)

It can also account for items that show special links (interface connections) between phonological structures, syntactic structures, and conceptual structures. Thus, Jackendoff distinguishes among:

(8)

Finally, within a Parallel Architecture approach for a VP idiom like kick the bucket Jackendoff proposes a link, specified by means of subindex 4 between the three representations in (9).¹¹

(9)

Overall, the idea is that idiomatic phrases are associated with syntactic structures, but their meaning is not composed from parts of it. Rather, they are associated with clusters of information that include conventionalized meaning and specific pragmatic meanings that can be represented by means of conceptual structures.

3.1 Conventionality

According to their definition of conventionality, expressions can be defined as conventional when their meaning or use can’t be predicted, or at least entirely predicted, on the basis of a knowledge of the independent conventions that determine the use of their constituents when they appear in isolation from one another. So the conventionality of idioms, if understood in this narrow sense, refers to the discrepancy between the figurative or idiomatic reading and the predicted literal meaning of the expression. For example, idiomatic expressions like kick the bucket or spill the beans are both considered to be conventional because their meaning (‘to die suddenly’ and ‘divulge secret information’, respectively) cannot be predicted in the relevant sense previously defined.

To a certain extent conventionality is a semantic property that idioms can be said to share with so-called “collocations” (see Sinclair, 1991; Torner & Bernal, 2017, i.a.). Typically, collocations have been said to be expressions that can be interpreted, unlike idioms, more or less correctly out of context, but cannot be produced correctly if the conventional expression is not already known to the speech community: for example, compare AE thumb tack and BE drawing pin (see Croft & Cruse, 2004, pp. 249−250). To put it in the words of Fillmore et al. (1988), collocations are “encoding idioms,” that is, idioms that can be interpretable by the standard rules of interpretation but are conventional for these particular expressions with this meaning. In contrast, “decoding idioms” like kick the bucket are the ones that cannot be decoded by the hearer. Any decoding idiom is then an encoding idiom: if the hearer cannot figure out what it means, then she is also not going to be able to guess that it is a conventional way to express that meaning in the language.

The distinction between idioms and collocations has been said to be a gradual one. For example, compare the reductive definition of collocations provided by Nunberg et al. (1994, pp. 493–494): “when we encounter a fixed expression that is missing several of the relevant properties—say one that involves no figuration, lacks a proverbial character, and has no strong association with popular speech—we become increasingly reluctant to call it an idiom. Examples might be collocations like tax and spend, resist temptation, or right to life.”

3.2 Compositionality

Compositionality refers to the degree to which the phrasal meaning, once known, can be analyzed in terms of how it is distributed among the individual parts of the expression. It is precisely the semantic property of compositionality that allows Nunberg et al. (1994) to make an important dual distinction between non-compositional idioms (aka “Idiomatic Phrases”; IPs) and compositional idiomatic expressions (aka “Idiomatically Combining Expressions”; ICEs). For example, the meaning of a prototypical idiom like kick the bucket is not distributed among the parts of the idiom: that is, the expression as a whole is mapped onto the meaning of the idiom. In contrast, the meaning of an idiomatic expression like spill the beans is distributed among their parts: the individual parts of the literal expression can be mapped onto individual parts of the figurative/idiomatic meaning (e.g., spill: ‘divulge’ // the beans: ‘the information/a secret/ . . .’). Importantly, Nunberg et al. (1994, p. 499) only intend the weaker claim that speakers are capable of recognizing the compositionality of a idiom like spill the beans “after the fact, having first divined its meaning on the basis of contextual cues” (emphasis ours). In short, what these authors have done is dissociate conventionality from non-compositionality. According to them, all idioms are conventional per definition but most of them are semantically compositional (although they do not provide any statistical information to show that the lexicon of English contains more ICEs than IPs).

Moreover, there are some authors (Mendívil, 2009; Mateu, in press) who prefer avoiding the contradictory label of “compositional idioms” and use the alternative label of “idiomatic collocations.” Indeed, a positive consequence of using this terminology is that it allows one to maintain the standard claim that the meaning of idioms is non-compositional, whereas the meaning of collocations is compositional. Accordingly, there appear to be two types of idiomatic expressions: proper idioms and idiomatic collocations. See also Larson (2017) for the relevance of this distinction in the realm of dative idioms.

The semantic property of (non-)compositionality has been shown to be related to syntactic (in-)flexibility (Gibbs & Nayak, 1989; Nunberg et al., 1994, i.a.): accordingly, those idioms that are semantically non-compositional are predicted to be syntactically inflexible, whereas idiomatic collocations (aka idiomatically combining expressions; ICEs) are expected to have a more flexible syntax. As argued by Nunberg et al. (1994, pp. 500–503), grammatical operations like the ones provided by modification, quantification, topicalization, ellipsis, and anaphora, among other things, provide important evidence that the pieces of compositional idiomatic expressions have identifiable meanings that interact semantically with each other: for example, pull strings can be safely claimed to be an ICE from the following tests, which show that parts of this idiomatic expression should be assigned independent meanings (e.g., pull: ‘exploit’ // strings: ‘personal connections’), contributing to the interpretation of the whole: compare Pat got the job by pulling strings that weren’t available to anyone else // We could pull yet more strings // Those strings, he wouldn’t pull for you // Kim’s family pulled some strings on her behalf, but they weren’t enough to get her the job (examples taken from Nunberg et al., 1994).¹²

3.3 Transparency

The third relevant semantic property is transparency/opacity, which should not be confused with compositionality/non-compositionality, respectively. An idiomatic expression can be claimed to be conceptually transparent when there is a metaphorical motivation for the meaning it involves. For example, consider the following examples that have to do with anger (Gibbs, 1995, 2007): blow your stack, flip your lid, hit the ceiling, get hot under the collar, and get steamed up, among others. Following Gibbs, these related idiomatic expressions can be said to be conceptually transparent because they can be claimed to be motivated by the conceptual metaphor ANGER IS HEATED FLUID IN A COINTAINER.

The existence of so-called families of idioms like the previous one goes against the traditional claims that idioms are dead metaphors and that their form-meaning mappings are arbitrary. Idioms are not isolated multi-word constructions stored in our mental lexicon but form networks thanks to the intervention of conceptual metaphors that motivate their related meanings (i.e., conceptual metaphors make their meanings [more] transparent).

Importantly, Lakoff (1987), Gibbs and O’Brien (1990), and Gibbs (1995, 2007, i.a.) argue that conceptual metaphors should not be conceived of as mere generalizations of linguistic meaning. For example, one could say that the mentioned idioms refer to the idea of getting very angry not because of conceptual metaphor but because the words stack, lid, ceiling, hot, and steam also have abstract meanings that convey the idea of anger. Against this possible objection Gibbs and his colleagues argue that one way of uncovering metaphorical knowledge in idiomaticity is through a detailed analysis of speakers’ conceptual images for idioms. Through different psycholinguistic analyses, which are reviewed in Gibbs (1995, 2007), this cognitive scientist has shown that, when imagining anger idioms, the subjects of his experiments know that pressure causes the action; that one has little control over the pressure; that its violent release is unintended; and that once the release has taken place (i.e., once the stack has been blown, the lid flipped, or the ceiling hit), it is difficult to reverse the action. Alternative lexical theories are said to be unable to explain the presence of these specific conceptual inference patterns, which leads Gibbs and his colleagues to conclude that the motivation for these particular folk conceptions comes from two conceptual metaphors (ANGER IS PRESSURIZED HEAT and THE MIND IS A CONTAINER) and that the mapping of information from a source domain (e.g., heated fluid in a container) to a target domain (e.g., the anger emotion) defines our conceptualization of anger and motivates the idiomatic expressions used to talk about anger. It is also worth noting that the first metaphor can be applied not only to idioms like the previous ones but also motivates the meaning of non-idiomatic expressions such as I exploded with anger.

Similarly, Espinal and Mateu (2010) claim that the meaning of Catalan idioms like treure el fetge per la boca ‘to expel the liver through the mouth’ (cf. Engl. to work one’s guts out); petar-se el cul ‘to explode the butt’ (cf. Engl. to laugh one’s butt off); sortir-li els ulls de les òrbites ‘(the eyes) to leave the orbits’ (cf. Engl. to cry one’s eyes out), and more can be said to be transparent by positing the underlying conceptual metaphor AN EXTREME INTENSITY IS AN EXCESSIVE DETACHMENT OF A BODY PART. Their claim is that the meaning of these idioms can be shown to be transparent to the extent that it is motivated by the relevant conceptual metaphor. This notwithstanding, it is important to realize that their meaning is not compositional: for example, the direct object el cul ‘the butt’ (cf. Ens vam petar el cul; We laughed our butts off) is not referential and the meaning of the idiom (i.e., ‘to laugh a lot’) is not distributed onto the individual parts of the expression (e.g., butts-off: ‘a lot’). Importantly, the relevance and recurrence of body parts in many idiomatic constructions has been claimed to reflect the embodiment nature of human thought (see Johnson, 1987; Gibbs, 2006, i.a.).

3.4 Semantic Compositionality Versus Conceptual Transparency

After having reviewed the three main semantic properties of idiomatic expressions (namely, conventionality, non-/compositionality, and transparency/opacity), it is worth noting that some formal linguists have taken pains to show that semantic compositionality and conceptual transparency must not be mixed when dealing with the analyzability of idioms. Indeed, it is important to point out that, although some correlation is expected to be found between these two properties (e.g., clearly transparent idioms tend to be compositional and, conversely, opaque idiomatic expressions tend to be non-compositional), it is relatively easy to find transparent idioms that are non-compositional.

For example, evidence for the distinction between semantic compositionality and conceptual transparency comes from the fact that there are idioms like hit the ceiling ‘to get very angry’, which have been claimed to be transparent (i.e., metaphorically motivated) but are not compositional (e.g., cf. #The ceiling was hit by Joe; notice that this passive example is only well-formed on the literal reading).¹³

Incidentally, a piece of evidence for this distinction between conceptual transparency and semantic compositionality also comes from the different effects of lexical versus syntactic flexibility of idiomatic expressions.

On the one hand, lexical flexibility can be related to conceptual transparency but not to semantic compositionality:for example, compare transparent ICEs such as throw one’s {hat/cap} into the ring with transparent IPs such as hit {the ceiling/the roof} (NB: of course, such a lexical flexibility is not generally at all due to the typical lexicalized nature of idiomatic expressions: e.g., cf. ICEs spill {the beans/#the peas} and IPs kick {the bucket/#the pail}).¹⁴

On the other, syntactic flexibility can only be found in semantically compositional idiomatic expressions (ICEs). Conceptual meaning, being it transparent or not, is not relevant for syntactic flexibility. Rather it is only semantic meaning that is relevant to syntactic processes. Following this thread, Mateu and Espinal (2007) argue that, when dealing with the meaning of idiomatic expressions, two different kinds of meanings can be distinguished: syntactically transparent meaning (i.e., the one relevant for semantic compositionality) and syntactically non-transparent meaning (e.g., the one relevant for conceptual/metaphorical motivation).

3.5 Aspectuality and Referentiality in Idioms

In relation to the important claim of Nunberg et al. (1994) that “Idiomatic phrases” (IPs) are non-compositional but “Idiomatically combining expressions” (ICEs) are compositional, it has been noted that the former do not necessarily preserve the aspectual class associated to the literal meaning, whereas the latter do preserve it (see Glasbey, 2003, 2007; Espinal & Mateu, 2010; contra McGinnis, 2002). For example, consider the figurative meanings of idioms such as to work one’s guts out ‘to work a lot’ and to paint the town red ‘to have a very enjoyable time’, which are non-compositional (e.g., notice that the direct object in these examples is not referential). It is interesting to point out that the aspectual reading of these idioms is different from the literal one: that is, the aspectual reading associated to the idiomatic meaning is atelic (cf. John worked his guts out {for hours/*in ten minutes}; They painted the town red {for hours/*in two hours}). In contrast, and as expected, the aspectual reading that corresponds to the literal resultative constructions is telic: for example, compare John worked his debts off in ten months; They painted the wall red in two hours. Glasbey (2003, 2007) claims that the activity reading of the eventuality in these idioms has to do with the absence of a gradual patient (in the sense of Krifka, 1992), which is indeed present in their literal resultative-like counterparts. For an alternative account, see Espinal and Mateu (2010), who claim that the telic-to-atelic event type-shifting involved in idioms like work one’s guts out could have a conceptual explanation related to the unbounded nature of the notion of intensity involved in the figurative meaning.

Interestingly, McGinnis (2005) points out that the atelicity of these idioms could be due to their having a non-referential object, but she finally discards this proposal by noting that the direct object of kick the bucket is not referential but its idiomatic aspectual reading is still telic. However, one could rebut this observation by noting that kick the bucket does not involve a gradual patient in the literal reading either (as shown by Levin, 1993; and Levin & Rappaport Hovav, 2005, i.a., impact verbs like kick or hit behave differently from change of state and incremental theme verbs), whereby the qualification of McGinnis (2005) could be claimed to lose ground. That is to say, the generalization could be explored that those idioms that have a gradual patient in the literal reading become atelic when the object DP is non-referential.

Finally, in connection with these related issues (compositional aspectuality and referentiality of the object) Espinal (2001, 2004b, 2009) has argued for the proposal in which the object nominals and object clitics of certain idioms in Romance are property denoting expressions rather than entity-denoting expressions. For example, in her analysis of some verb plus object idioms, this author argues that the nominal expression in object position of unergative argument structures (e.g., Cat. fer via, lit. “to make way”, i.e., ‘to hurry up’) denotes a gradable property and conveys event modification over a degree scale: for example, compare fer {més/molta/força/massa} via, lit. “to make {more/much/a lot/too much} way,” that is, ‘to hurry up (to a certain degree)’. These objects, because of their denotation, have been claimed to be semantically incorporated into the verbal predicate. That is to say, Espinal’s proposal is that, at some level of representation, namely at the syntax–semantics interface, these object nouns denote properties and form complex predicate units with the target verb (cf. Farkas & De Swart, 2003; Dayal, 2011, i.a., for semantic incorporation).

3.6 Thematic Constraints in the Formation of Idioms

To conclude this section on the meaning of idioms, it will be important to make some remarks on thematic role constraints mentioned in the literature. Due to its major theoretical relevance, one of the most relevant constraints has even been transformed into a hypothesis: that is, the so-called No Agent Idioms hypothesis (e.g., see Harley & Stone, 2013). It is often noted that there are few idioms, if any, that can be claimed to include agents. However, some apparent counterexamples can be found, such as A little bird told me that . . . In any case it should be noted that the agent in this example does not form a non-compositional IP but a compositional ICE: indeed, the subject must have a referential status (cf. ‘And I replied to him/her that . . .’), unlike the direct object of non-compositional idioms like kick the bucket, where this nominal phrase does lack referentiality.

Furthermore, some authors have claimed that there is an important thematic distinction that separates the object of non-compositional idioms (IPs) like kick the bucket ‘to die’ from the object of compositional idiomatic expressions (ICEs) like bury the hatchet ‘to reconcile disagreement’: the former is said to lack a thematic role (the bucket is a non-meaningful object), whereas the latter does have one (i.e., the one that would correspond to “disagreement”). For example, see Jackendoff (1997a, p. 169): ‘‘In other words, the hatchet is linked to bury via its theta-role; but the bucket has to be linked to kick syntactically because it has no theta-role. Hence the hatchet is movable and the bucket is not.”

However, if thematic roles are conceived in terms of conceptual structures, there appears to be a non-trivial problem when comparing conceptually opaque IPs like kick the bucket (where the bucket can be said to be non-meaningful; cf. also kick off, which in slang also means ‘die’) with conceptually transparent IPs like saw logs or hit the ceiling (where it is less clear that the object lacks a conceptual thematic role). A similar qualm could be claimed to hold for transparent idioms like to laugh one’s head off (see the relevant metaphor: AN EXTREME INTENSITY IS AN EXCESSIVE DETACHMENT OF A BODY PART). Assuming that there is a displacement involved in our conceptualization of the source domain of this idiom (one’s head could be claimed to be the displaced object, i.e., the Theme, in this domain), it is indeed dubious that the meaning of this idiom can be reduced, as Jackendoff claims, to ‘laugh a lot’. Assuming the dual distinction of meaning pointed out between (syntactically relevant) semantic structure versus (syntactically irrelevant) conceptual structure (a distinction that Jackendoff does not accept; see Jackendoff, 1983, 2002), the relevant conclusion seems to be that the fact that the object lacks referentiality in semantic structure does not involve that it has no thematic role in conceptual structure: for example, although the object the bucket in kick the bucket lacks both referentiality and theta-role, it can be claimed that the object logs in saw logs, or the ceiling in to hit the ceiling, lacks referentiality (that’s why these idioms can be classified as IPs) but does have a conceptual thematic role (at least in the conceptually structured source domain).¹⁵

4.1 Lexicographic Studies

Many words have a phraseological tendency (the idioms principle; Sinclair, 1991) in addition to a terminological tendency (the open-choice principle). According to the idioms principle “a language user has available to him a large number of semi-preconstructed phrases that constitute single choices, even though they might appear to be analyzable into segments” (Sinclair, 1991, p. 110). According to the open-choice principle practically each position in a clause offers a choice.¹⁶

The existence of a phraseological tendency in language has inspired an extensive number of studies concerned with different aspects of ‘chunking’, but the basic question that linguists and lexicographers aim at answering is how idioms should be recorded in monolingual and bilingual dictionaries. The first problem raised at the time of answering this question is that meaning is context-dependent, which means that the meaning of an idiom is highly dependent on strict co-occurrences, for which corpus analyses can provide highly relevant information and measure the statistical significance of existing collocations.

Additionally, given a significantly frequent co-occurring group of words, a second problem has to do with its form and, more specifically, with the identification of the word-lemma to whose entry the idiom will be sorted in a dictionary. Is it the first lexical item or rather the last one? Or perhaps the identification of this word-lemma does not depend on word order but rather on structural criteria; that is, in the case of a VP idiom, the head is the V, and in the case of an NP idiom, the head is the N. To illustrate this claim, consider the verb blow in English. This verb is associated in the online Cambridge Dictionary with the idiomatic expression blow a fuse ‘become very angry’, while the noun fuse is associated with the idiomatic expression have a short fuse ‘get angry very easily’. Notice that in the latter expression, even though it has the form of a VP, the verbal head is a light verb with little semantic content and the predicate relies on the complement object noun. This is probably the justification why the idiom have a short fuse is a subentry of the N, the head of the predicate.¹⁷

The contrast just exemplified illustrates divergent decisions made in lexicography regarding the codification of idioms and how they should be registered in monolingual dictionaries. Additional differences may appear regarding the sort of information (definition, category, example, semantic relations with other idioms, etc.) that each entry provides.¹⁸ Interestingly, however, a quick look at the Collins Cobuild English Language Dictionary, a monolingual dictionary for learners of English, provides a different solution, given that both blow a fuse and have a short fuse are subentries under the noun fuse. The conclusion is that there is not a unique and perhaps an optimal way of registering and accessing idioms in lexicography. Different solutions seem to be the response to diverse lexicographic needs and analyses of what the syntactic or semantic head of an idiom is.

Lexicographic research has addressed the question of accessibility of phraseological information in dictionaries. Some studies of dictionary use and dictionary making have investigated whether their potential users would be able to find different sorts of idioms and collocations depending on the specific goal each type of dictionary is designed for (Lew, 2011, 2015).

Overall, current research shows that the study of idioms and collocations (Fellbaum, 2011) has been very productive from corpus-based linguistic and lexicographic studies interested in describing what constitutes a multi-word expression in different natural languages (Stubbs, 2001; Fellbaum, 2016). Other areas on which lexicographic research is crucial are computational linguistics (see section 4.2), machine translation (Volk, 1998), contrastive idiom analyses (Dobrovol'skij, 2000), corpus-based analyses (Moon, 1998; Fellbaum, 2007; Stubbs, 2007), dictionary-making (Mel’cuk, 1995, 2012; Moon, 2015; i.a.), and foreign-language learning (Hausmann, 2004).

4.2 Computational Studies

Researchers working on aspects of computational models of idiomaticity or “computational phraseology” (Heid, 2007, 2008) typically speak of “multi-word expressions” (MWEs) in their studies of Natural Language Processing (NLP). Sag, Baldwin, Bond, Copestake, and Flickinger (2002), in a work that has become a classic, call MWEs a “pain in the neck for NLP,” and they describe some of the phenomena that contribute to this designation. Idioms are of course classified under a group of MWEs. Indeed, one of the main issues that computational approaches to idioms have to deal with is how to automatically tell them apart from compositional constructions (e.g., see Fazly & Stevenson, 2006).

Since the turn of the 21st century, there has been an increasing interest in the computational analyses of MWEs. A good example is the LinGO project (Copestake et al., 2004), which has developed a format for representing MWEs (Copestake et al., 2002). Similarly, large NLP dictionaries, such as FrameNet (Ruppenhofer, Baker, & Fillmore, 2002), include descriptions of, and representation formats for, MWEs. For the LinGO lexicon, see Villavicencio, Copestake, Waldron, and Lambeau (2004).

Knowing whether an expression receives a literal meaning or an idiomatic meaning is important for NLP applications that require some sort of semantic interpretation. It is clear that some applications that would benefit from knowing this distinction are machine translation, (multilingual) information retrieval, and the like. For example, drawing on previous works on these applied fields, Villada-Moirón and Tiedemann (2006) explore to what extent word-alignment in parallel corpora can be used to distinguish idiomatic multi-word expressions from more transparent multi-word expressions and fully productive expressions.

4.3 Psycholinguistic and Neurolinguistic Studies