Multi-Word Expressions and Morphology

Francesca Masini

doi:10.1093/acrefore/9780199384655.013.611

Summary

Multi-word expressions are linguistic objects formed by two or more words that behave like a ‘unit’ by displaying formal and/or functional idiosyncratic properties with respect to free word combinations. They include an extremely varied set of items (from idioms to collocations, from formulae to sayings) which have been the privileged subject matter of fields such as phraseology, lexicology, lexicography, and computational linguistics. Far from being a marginal phenomenon, multi-word expressions are ubiquitous and pervasive: some estimate that they are as numerous as words in some languages, which makes them as central an issue as words for the understanding of human language. However, their relation with words, and morphology, is by far less explored, not to say neglected, especially in terms of demarcation, competition, and cross-linguistic variation.

1. Introducing Multi-Word Expressions: A Bare-Bones History

Morphology is about words: word structure, word meaning, and the systematic correspondences between the two (Haspelmath & Sims, 2010, p. 2; Booij, 2012, p. 7). Words, or better lexemes, have the essential function of organizing and categorizing human experience into symbolic units, and word formation is a process that implements this function by creating new lexemes that serve to name new concepts. Concept-naming, however, is not a prerogative of morphology, specifically word formation: there are other items in natural languages that do not conform to what are normally called ‘words’ and that serve precisely the same function. These items are so-called multi-word (or multiword) expressions (henceforth, MWEs).

Generally speaking, one may operationally define MWEs as linguistic objects formed by two or more words that behave like a ‘unit’ or ‘chunk’ in that they display some formal or functional idiosyncratic properties with respect to free word combinations (i.e., normal phrases). MWEs comprise a large and heterogeneous set of objects, including, for instance, idioms (e.g., spill the beans, pull the wool over someone’s eyes), complex nominals (e.g., weapons of mass destruction, sense of humor), irreversible binomials (e.g., kith and kin, salt and pepper), verb-particle constructions (e.g., wrap up, give up), and other complex predicates (e.g., take advantage of, give rise to). Sometimes ‘MWE’ is employed as an umbrella term that refers to this vast domain (as in this article); sometimes other terms are used instead, for instance, ‘idiom’, ‘collocation’, ‘locution’, and ‘fixed expression’.

This lax terminology depends on both scholarly and national linguistic traditions.¹ It is not a product of recent fragmentation but traces back at least to early Structuralism, or even to the pre-Structuralist period, and has been increasing until very recent times. de Saussure (1916) talks about ‘locutions toutes faites’ that are fixed, hence not the product of free syntagmatic combinations (e.g., rompre une lance ‘speak in favor’, à force de ‘by dint of’), whereas Bally (1909) speaks of ‘locutions phraséologiques’ (e.g., sans coup férir ‘without any resistance’, remporter une victoire ‘achieve a victory’), thus introducing the term ‘phraseology’, which has been adopted especially in the Russian and British traditions (Cowie, 1998a).² Later, Benveniste (1966) tackles a specific MWE in French that he calls ‘synapsie’—formed by the combination of two nouns (or a noun and a verb) connected by a preposition (e.g., clair de lune ‘moonlight’, machine à écrire ‘typewriter’)—that he identifies as the true productive compounding pattern in the French language (the very same pattern is labelled ‘synthème’ by Martinet, 1967). The terminological heterogeneity increases in the following decades, when other labels emerge, among which: ‘lexical phrase’ (Becker, 1975), ‘syntactic compound’ (Lyons, 1977), ‘fixed expression’ (Gross, 1996; Jackendoff, 1997; Moon, 1998), ‘locution’ (Gross, 1996), ‘phraseme’ (Mel’čuk, 1998), ‘formulaic sequence’ (Wray, 2002). The privileged term in the US tradition is ‘idiom’ (Makkai, 1972), which has been adopted by studies within Generative Grammar (e.g., Chomsky, 1981; Fraser, 1970; Jackendoff, 1997; Nunberg, Sag, & Wasow, 1994; Schenk, 1995; but see Di Sciullo & Williams, 1987, who use the term ‘listeme’) and, later, in early Construction Grammar (see below, in this section). The term ‘MWE’ (and its variants) starts emerging in the 1990s primarily in more applied fields like Natural Language Processing (Baldwin & Kim, 2010; Sag, Baldwin, Bond, Copestake, & Flickinger, 2002)—but also lexicography (see ‘multiword lexical units’ in Zgusta, 1967, 1971; ‘multiword lexeme’ in Gates, 1988; ‘multi-word unit’ in Fellbaum, 2015)—and then rapidly spreads among more descriptive and theoretical domains (see, e.g., Hüning & Schlücker, 2015; Kay & Michaelis, 2018; Masini, 2005; Schlücker, 2019), becoming one of the most widely used terms in present days.³

This huge terminological variation reflects the lack of a unified tradition and of communication between different research areas, a basic uncertainty as to where in the language faculty MWEs should be found and treated, and a certain conceptual complexity regarding MWEs: as will become more evident in the following sections, there are inherent reasons why this domain is so difficult to grasp and categorize.

This aside, what emerges quite clearly is that MWEs have been mostly investigated separately from morphology and word formation. They have been at the center of interest of other fields, primarily: phraseology (cf., Burger, Dobrovol’skij, Kühn, & Norrick, 2007/2008; Cowie, 1998b), lexicology/lexicography/terminology (cf., Cruse, Hundsnurscher, Job, & Lutzeier, 2002/2005), corpus linguistics and Natural Language Processing (cf., e.g., Ramisch, 2015; Ramisch & Villavicencio, 2018; see also Section 2.3), but also psycholinguistics (cf., e.g., Cacciari & Tabossi, 1993; Everaert, van der Linden, Schenk, & Schreuder, 1995; Sprenger, Levelt, & Kempen, 2006), language acquisition (cf., e.g., Meunier & Granger, 2008; Wray, 2002) and syntactic theories (especially Construction Grammar). More rarely have morphologists addressed these multi-word phenomena. A notable exception regards complex predicates (cf., e.g., Ackerman & Webelhuth, 1998; Butt, 1995)—in particular verb-particle constructions in Germanic (cf., Dehé, Jackendoff, McIntyre, & Urban, 2002; Los, Blom, Booij, Eleenbaas, & Van Kemenade, 2012) but also Romance languages (cf., Iacobini & Masini, 2007).

Much of recent work in this area has been framed within Construction Morphology (Booij, 2010, 2017), which is not surprising given that—as also observed by Hüning and Schlücker (2015)—Construction Morphology is linked to Construction Grammar (Hoffmann & Trousdale, 2013), a model whose foundations lie in studies on idioms, from Fillmore, Kay, and O’Connor (1988) onward. Indeed, Construction Morphology originated from work on phenomena in between morphology and syntax, in particular separable complex verbs in Dutch (Booij, 2002a, b). Later, investigation has extended to other types of so-called phrasal lexemes or lexical phrases (Booij, 2009, 2010; Masini, 2009), that is, multi-word items that serve as complex lexemes: they are not words in the proper sense, since they have a phrase-like structure, but at the same time present a unitary, often conventionalized semantics and display a certain degree of internal cohesion that sets them apart from free phrases. Under this approach, phrasal lexemes, which may belong to different parts of speech, are seen as part of our lexicon on a par with (morphologically complex) words, which leads to issues concerning the relation between words and MWEs, to be addressed in Section 3.

The rest of the article is organized as follows. Section 2 illustrates the main formal and functional properties of MWEs (2.1), the main classifications that have been proposed in the literature (2.2)—with special discussions on MWEs and parts of speech (2.2.1), and MWEs and periphrases (2.2.2)—and their significance for linguistic description and theory (2.3). Section 3 focuses on the word-MWE relationship in terms of both demarcation (3.1) and competition (3.2). Section 4 summarizes some of the issues regarding MWEs that have been at the center of theoretical debate, whereas Section 5 tackles the challenging and neglected question of the interaction between MWEs and linguistic typology, proposing some tentative reflections on MWEs and cross-linguistic variation. Section 6 provides some suggestions for further reading.

2. A Bird’s-Eye View of MWEs

As Section 1 makes clear, when talking about MWEs one may be referring to a highly heterogeneous set of linguistic units which are often named differently in different traditions and fields. Generally speaking, MWEs (whatever their exact nature) are recognized as such because they display one or more idiosyncratic properties at the formal, functional, or usage level. These idiosyncrasies must be specified as part of our competence, thus requiring MWEs to be stored at some level of representation. This sets them apart from fully predictable outputs of syntax.

It would be impossible to do justice to the wide variety of attested MWEs, each with its specific properties, and the many classifications proposed. Hence, this section summarizes some of the characteristics that are recurrent in MWEs, focusing on those that are more relevant for the subsequent discussion of the word-MWE relationship.

2.1 Formal and Functional Properties of MWEs

The property most readily associated with MWEs is semantic idiomaticity, that is, non-compositionality and figurativeness. See the examples in (1), from English and Italian.

(1)

MWEs may also be structurally idiosyncratic, that is, display syntactic anomalies. Some authors call the MWEs in question ‘extragrammatical’ or ‘ill-formed’ idioms (cf., Fillmore, Kay, & O’Connor, 1988; Moon, 1998; Nunberg, Sag, & Wasow, 1994)⁴:

(2)

One of the crucial formal properties of MWEs is fixedness, which may be expressed both syntagmatically (impossibility to interrupt the sequence, to change the order of the constituents, or to manipulate them syntactically, etc.; cf., the Italian examples in (3)) and paradigmatically (e.g., unsubstitutability, at different levels, cf., the English examples in (4)).

(3)

(4)

Idiosyncratic properties are not equally present in all MWEs, and, moreover, they are scalar rather than binary, which obviously results in great variability.

At the semantic level, for instance, both kick the bucket and pull the strings are non-compositional, but in kick the bucket non-compositionality is stronger, leading to complete opacity of meaning, contrary to pull the strings. In this respect, Nunberg (1978) claims that some idioms are partially analyzable, which is also testified by the possibility to apply some syntactic operations to their internal constituents (cf., also Schenk 1995; Wasow, Sag, & Nunberg, 1983). Along the same lines, Nunberg, Sag, and Wasow (1994, p. 491) distinguish between “idiomatically combining expressions (e.g., take advantage, pull strings), whose meanings—while conventional—are distributed among their parts” and “idiomatic phrases (e.g., kick the bucket, saw logs), which do not contribute their meanings to their components” (emphasis in the original) (cf., also Makkai, 1972). Some authors speak in favor of an ‘idiomaticity continuum’ (Wulff, 2008; cf., also Kay & Michaelis, 2012; Michaelis, 2017), along which MWEs are found with different degrees of idiomaticity (and hence productivity).

A similar reasoning holds at the formal level. It is often claimed that MWEs lie on a continuum between full variability in word combination and internal modification on the one hand and complete cohesion and rigidity on the other, somehow challenging the neat separation between syntax (intended as the repository of ‘rules’) and lexicon (intended as the repository of ‘irregularities’) (cf., Jackendoff, 1995). Within this continuum, there are MWEs that are completely rigid (often those that also display full idiomaticity of meaning), and MWEs that allow for some (morphosyntactic) variation of their components, displaying different degrees of fixedness which are not straightforwardly predictable (cf., Moon, 1998 on ‘creative’ idiom variability in English). Different degrees of internal cohesion/variability might correlate with parts of speech: De Mauro and Voghera (1996), for instance, claim that in Italian morphosyntactic rigidity is much stronger in multi-word nouns than multi-word verbs (cf., also Voghera, 2004).

In conclusion, MWE-hood seems to be characterizable as a gradient property: if at one end of the spectrum there are frozen, syntactically irregular fixed expressions such as by and large, on the other one finds compositional, transparent, but statistically marked expressions such as fresh air. Some expressions, therefore, embody the core properties of MWEs, being better representatives (e.g., uninterpretable, invariable idioms), whereas others have different degrees of inclusion into this set.

2.2 Types of MWEs

Many different types of MWEs can be identified according to their formal properties (e.g., more or less flexible) and their degree of idiomaticity. Indeed, the considerable variability of MWEs makes them hard to classify, both intra- and cross-linguistically (cf., among others, Moon, 1998). Various classifications have been proposed in the literature, each going with a specific terminology. Since it would be impossible to summarize them all here, a few selected examples are given which should illustrate the basic criteria that have been used to classify MWEs.

Overall, classifications of MWEs are of three types. Authors categorized them according to their (i) formal properties (degree of internal cohesion or fixity), (ii) idiomatic status (see Nunberg, Sag, & Wasow, 1994, Section 2.1), and (iii) function, or a combination of these.

As for (i), already Bally (1909), in his seminal work on phraseology, pointed out that locutions phraséologiques should be distinguished between unités phraséologiques on the one hand (e.g., manière d’agir ‘way of doing, attitude’, en guise de ‘by way of’), and séries phraséologiques or groupements usuels on the other (e.g., prendre une decision ‘make a decision’): whereas the former are characterized by total internal cohesion (and idiomatic meaning), the latter are only partially fixed since the internal constituents maintain some autonomy. In a later framework (Transformational-Generative Grammar), Fraser (1970) proposed to classify idioms according to a ‘Frozenness Hierarchy’ defined by two poles (from unrestricted to completely frozen), along which one finds idioms that allow or don’t allow certain syntactic operations, like insertion or extraction. In still another framework (Construction Grammar), Fillmore, Kay, and O’Connor (1988) differentiate between ‘grammatical idioms’, namely, idioms that comply with the general syntactic rules of the language (like spill the beans) and ‘extragrammatical idioms’, which are syntactically anomalous or ill-formed (cf., (2) in Section 2.1). The authors also suggest another distinction which will be of fundamental importance in the development of Construction Grammar, namely, the one between ‘substantive idioms’ and ‘formal idioms’: the former are idioms that are fully lexically specified (e.g., all of a sudden, going great guns); the latter are idiomatic patterns that are lexically open, since they contain variables, like the Xer the Yer (of which the expression the sooner the better is an instance). Within computational studies, Sag et al. (2002) distinguish between ‘lexicalized phrases’ (including fixed, semi-fixed and syntactically flexible expressions, see every which way, attorney general, and look up, respectively), namely, sequences which display at least some idiosyncratic syntactic or semantic property, and ‘institutionalized phrases’, namely, sequences that are compositional but particularly frequent, thus undergoing conventionalization (e.g., fresh air, kindle excitement).

As for (ii), a well-known classification (also adopted in subsequent works) is the one put forward by Makkai (1972) between ‘idioms of encoding’ (like answer the door), idiomatic to be encoded but less problematic to be interpreted, and ‘idioms of decoding’ (like red herring), which must be part of the speaker’s knowledge both to be encoded and to be decoded. Within the phraseological tradition, Vinogradov (1947)⁵ divides ‘phraseological units’ (frazeologičeskie edinicy) into: ‘phraseological fusions’, which are semantically unmotivated or opaque (like hot potato); ‘phraseological unities’, which are partially motivated, via metaphorical extension (like blow off steam); and ‘phraseological combinations’, where one constituent is used in its literal sense, whereas the other is used figuratively (like meet the demand). See also the already mentioned difference between ‘idiomatically combining expressions’ and ‘idiomatic phrases’ put forward by Nunberg, Sag, and Wasow (1994) (Section 2.1).

As for (iii), let us start again with Makkai (1972), who draws a distinction between ‘lexemic idioms’ and ‘sememic (or cultural-pragmemic) idioms’, which is basically functional in that the former identify idioms with a lexical function (like phrasal verbs), whereas the latter are expressions with a pragmatic function (formulae, sayings, clichés, etc.). Along similar lines, Fillmore, Kay, and O’Connor (1988) speak of ‘idioms without pragmatic point’ and ‘idioms with pragmatic point’ (e.g., formulae like how do you do?), whereas Mel’čuk (1998) speaks of ‘pragmatic phrasemes’ (or ‘pragmatemes’) and ‘semantic phrasemes’, which include idioms (shoot the breeze), collocations (launch an attack, strong coffee) and quasi-idioms, whose meaning is obtained by combining the meaning of the constituents plus an unpredictable factor (bacon and eggs, shopping center). Moon (1998), instead, elaborates a classification based on the discourse function of MWEs (‘fixed expressions and metaphors’, in her terminology): ‘informational’, that is, conveying information (rub shoulders); ‘evaluative’, that is, conveying the speaker’s attitude (kid’s stuff); ‘situational’, that is, relating to the extralinguistic context (excuse me!, talk of the devil); ‘modalizing’, that is, conveying truth values, advice, requests, etc. (I kid you not, you know what I mean); and ‘organizational’, that is, organizing discourse structure (by the way, for instance). In a similar fashion, Wray (2002) develops a classification of the lexicon (which puts morphemes, words, and formulaic sequences on a par with each other) into: ‘grammatical’ (in order to, on account of), ‘referential’ (pull one’s leg, highly unlikely), ‘interactional’ (great to see you, of course), ‘citations’ (like poems, songs, etc.), and ‘reflexive’ (goodness gracious!, what the hell).

In conclusion, MWEs enrich the lexicon in more than one way and serve different functions, from more referential to more pragmatic ones. Since this article deals with MWEs in relation to words and morphology, the next subsections deal with two relevant issues: types of MWEs by parts of speech (2.2.1) and the role of periphrases with respect to MWEs (2.2.2).

2.2.1 Multi-Word Expressions and Parts of Speech

MWEs may virtually feed many parts of speech, or word classes, from ‘major’ ones to ‘minor’ ones, with differences in productivity. This doesn’t amount to say that they actually do so in all languages.⁶ What we know is that they actually enrich many parts of speech in various European languages. For instance, there are multi-word items that belong to the class of nouns (5), verbs (6), adjectives (7), adverbs (8), prepositions (9), and conjunctions (10).⁷

(5)

(6)

(7)

(8)

(9)

(10)

Whereas MWEs belonging to ‘minor’ word classes seem to be the result of a diachronic process of lexicalization in these languages, those belonging to ‘major’ word classes can be productively created (see also Section 2.3). Exactly to which extent MWE creation is productive, and for which part of speech, depends on specific languages (see also Section 3.2).

2.2.2 What About Periphrases?

A special (and often overlooked) issue related to the typology of MWEs is the status and role of periphrases, namely, analytical structures that encode grammatical meanings, like TAM features (see, for instance, be going to + V or will + V in English). Several morphologists claim that periphrases can be regarded as exponents of inflectional cells (cf., e.g., Ackerman & Stump, 2004; Chumakina & Corbett, 2013; Sadler & Spencer, 2001), hence forms that are complementary to synthetic inflected forms in filling an inflectional paradigm. However, the notion of periphrasis is sometimes invoked also in relation to other types of structures, like causative constructions (see Italian fare uscire lit. make exit ‘to let out’), which are less clearly linked to inflectional paradigms.

Periphrastic constructions have many connections with MWEs: in both cases we are dealing with multi-word phenomena that differ from ordinary syntax by virtue of their special formal or functional properties. However, their function is different: whereas periphrases express grammatical meanings and thus correlate with inflection, MWEs encode more lexical meanings and correlate with word formation.

In line with this view, Bonami (2015) claims that periphrases are similar to syntactically flexible idioms and can be formally modeled as such. He also points out that “while idioms carry semantic content, and are thus multi-word equivalents of lexemes, periphrases carry morphosyntactic content, and are thus multi-word equivalents of inflected words” (Bonami, 2015, p. 86). See also Booij (2002a, b), who, in his studies on separable complex verbs in Dutch, notably speaks of “periphrastic word formation” (cf., also Booij, 2010).

In conclusion, inflectional periphrases and MWEs are strictly related phenomena that basically reproduce the inflection-derivation divide at the multi-word (analytical) level, and that might turn out to represent two sides of the same coin, that is, a larger common domain of investigation.

2.3 Why Do MWEs Matter?

One may wonder why one should care about MWEs, in general and in relation to morphology. There are at least three good reasons for doing so.

First, MWEs are anything but a marginal phenomenon. It is known that MWEs are numerous in many languages, although precise estimates are not easy to obtain, especially for less described languages. As Baldwin and Kim (2010, p. 268) note, the “number of MWEs is estimated to be of the same order of magnitude as the number of simplex words in a speaker’s lexicon (Jackendoff, 1997; Pauwels 2000; Tschichold 1998). At the type level, therefore, MWEs are as much of an issue as simple words.” Erman and Warren (2000) reveal that over 55% of the tokens in the texts they studied were instances of what they call ‘prefabs’. In WordNet 1.7 (Fellbaum, 1998), 41% of the entries are multi-word. De Mauro (1999) states that, out of approximately 360.000 lemmas in the GRADIT dictionary of Italian, there are 130.000 multi-word sublemmas (called ‘polirematiche’). Mel’čuk (1998, p. 24) goes as far as claiming that “in any language [. . .] phrasemes outnumber words roughly ten to one” (cf., also Mel’čuk, 1995). Moreover, “new (types of) MWEs are continuously created as languages evolve” (Baldwin & Kim, 2010, p. 268; cf., also Finkbeiner & Schlücker, 2019; Masini, 2019, in press).

Second, MWEs are a challenge for many applied domains of linguistics. For instance, they pose problems for language learners, who need to develop specific skills to master these expressions. It is not by chance that we frequently witness the emergence of new tools that serve precisely this purpose: see for instance the many available dictionaries of idioms and collocations available for the English language, such as the BBI Combinatory Dictionary of English (Benson, Benson, & Ilson, 2010), the REDES combinatory dictionary for Spanish (Bosque, 2004), or the Dizionario Combinatorio Italiano (Lo Cascio, 2013) for Italian. In addition, given their unpredictable properties, and their heterogeneous characteristics, it comes as no surprise that MWEs are a “pain in the neck” (also) for Natural Language Processing (NLP), to put it in the words of Sag et al. (2002). Whereas native speakers generally use and understand MWEs with no effort, NLP systems need to find a way to deal with them: “it is at the same time difficult and necessary to deal with MWEs in applications that involve some degree of semantic interpretation of natural language. As a consequence, the computational treatment of MWEs is considered as a major challenge in NLP” (Ramisch, 2015, p. 9; cf., also Villavicencio, Bond, Korhonen, & McCarthy, 2005, p. 370). Considerable efforts have been devoted in the past decades to develop techniques for the (semi-)automatic identification and extraction of MWEs from corpora, and, consequently, to build resources (lexicons, databases, dictionaries, annotated corpora, ontologies, etc.) to serve as gold standards for evaluation and for training statistical models, as well as for reference and language learning. The availability of resources for MWE is still quite limited compared with “the ubiquitous and pervasive nature of MWEs” (Ramisch, 2015, p. 9), but more and more research effort has been focusing on this in recent times.⁸

Third, some MWEs are used to cover one crucial function in language, that is, ‘concept naming’, which puts them in competition with word formation and its products. This leads us directly to the contents of Section 3.

3. Words and MWEs

As a premise, one should keep in mind that the study of the relationship between morphological words and MWEs is still quite in its infancy and definitely deserves further investigation. The previous sections called our attention to the huge variability and heterogeneity of MWEs. In this respect, one aspect that is particularly relevant to our present discussion is that MWEs may have different complexity and different functions. Some MWEs have the same distribution of sentences, like sayings or quotations; formulaic expressions may also serve as full utterances. Other MWEs, in particular those that have been called ‘phrasal lexemes’ (Booij, 2009, 2010; Masini, 2009), are closer to morphological words (especially compounds), since they have a word-like distribution (e.g., they could be substituted with an equivalent single word, in the same language or in another language) and they have the same concept-naming function of words, thus contributing to lexical enrichment, to the creation of new ‘lexemes’. In discussing demarcation (Section 3.1) and competition (Section 3.2) issues I focus precisely on these MWEs, leaving other types of MWEs aside, thus delimiting the scope of the interaction between morphology and the multi-word realm.

3.1 Demarcation Issues

Word-MWE demarcation is anything but a trivial issue. Since MWEs are made of two (or more) words, the morphological objects we need to delimit them from are mainly compounds. At the same time, however, it is often observed that compounds, being formed by two (or more) words/roots/stems, are the morphological structures closest to syntax, which is why drawing a line between compounds and phrases is often difficult and has inspired so much work. In fact, even defining what a ‘compound’ exactly is is far from easy (see Lieber & Štekauer, 2009, for a discussion). Introducing MWEs into the picture complicates the situation even more.

Given these shaky foundations, the question now is: is there a way to distinguish between MWEs and compounds? On the basis of which criteria? Are there criteria that would hold cross-linguistically? The expectation is that cross-linguistic validity is hardly achievable, not only because it is unclear if the notion of ‘MWE’ can be applied to any language (see Section 5) but also because of the behavioral variability of both MWEs and compounds within and across languages. Nevertheless, some attempts have been made to distinguish compounds and MWEs in specific languages.⁹

Booij (2009), for instance, argues that in Dutch AN compounds and AN phrasal lexemes can be formally distinguished since the latter display agreement inflection on the adjective (see the suffix -e in (11a,b)),¹⁰ whereas the former do not ((11c,d)).

(11)

According to Schücker and Hüning (2009), German works very similarly: in AN compounds the adjective is not inflected, bears the main stress and is generally monomorphemic (see (12a)), whereas in AN phrasal lexemes (see (12b,c)) the adjective is inflected, does not bear the main stress, and can be complex.

(12)

In Russian (and other Slavic languages), phrasal lexemes display regular agreement (see (13a)) or government (see (13b)) among the constituents (which are independent words), whereas most compounds don’t, since the first member is typically a root (hence a bound element) connected to the second constituent by a linking vowel (lv) (see (13c)), although this is not always the case (see, e.g., (13d)) (cf., Masini & Benigni, 2012).¹¹

(13)

In Italian, phrasal lexemes may display internal agreement (see (14a)), explicit relational markers such as conjunctions or prepositions (see (14b,c)), minor word classes such as articles (see (14c)), unlike ‘true’ compounds. Instead, bound elements such as stems (cava- in (14d)) can be said to characterize compounds, not phrasal lexemes (cf., Masini, 2019).

(14)

Some cases might be not clear-cut. Take for instance binomial expressions like va e vieni (lit. go and come) ‘coming and going’ in Italian (Masini, 2009): these items are formed by two verbal stems (hence bound elements) and are combined by the conjunction e ‘and’ (hence a relational marker), thus positioning in-between ‘true’ compounds and ‘true’ phrasal lexemes (cf., also Arkadiev & Klamer, 2019, for similar observations based on Russian). Therefore, one should be prepared to accept that, even when it is possible to identify useful criteria, the demarcation might be a matter of degree.

All in all, the demarcation between compounds and MWEs largely relies on criteria that typically distinguish words from phrases, with the complication that MWEs (viz. phrasal lexemes) are not free, full-fledged phrases, due to their idiosyncratic behavior (internal cohesion, fixedness, etc.; cf., Section 2.1). Quite expectedly, the criteria used vary from language to language, largely depending on language-specific properties. Hence, we expect the array of criteria to increase as we investigate more languages. Some criteria may be shared by more than one language (e.g., internal agreement), whereas others may not. It is also worth noting that the criteria discussed in this section are structural, not semantic.¹² Both these facts about distinguishing criteria—language-specificity and structural nature—are in line with the general belief in linguistic typology that wordhood can only be determined in structural (i.e., phonological and morphosyntactic) terms (Dixon & Aikhenvald, 2002), and with the claim that these criteria are largely language-specific (Haspelmath, 2011) (cf., Arkadiev & Klamer, 2019, for a recent discussion of the notion of ‘word’).

Given these premises, we might expect some variation in the actual applicability of a word-MWE distinction: there are languages in which the formal differences between compounds and MWEs are more evident and easily detectable (Russian seems to be a case in point, since compounds are mostly root-compounds); languages where compounding is mostly word-based (like Italian), which makes the distinction less clear; and languages like English where the justification for this discrimination is all the more feeble, given that even the compound versus phrase distinction is highly uncertain. As is known, English AN and NN combinations have long been ground for debate in this sense. Stress has often been invoked to be a discriminating criterion (in absence of other morphosyntactic criteria): compounds would be fore-stressed (blácktop, wátchmaker), phrases would be end-stressed (blue bírd, home vídeo) (cf., e.g., Bauer, 1998; Marchand, 1969). However, even this generalization has been challenged (see, e.g., Giegerich, 2005, 2009, 2015).

In conclusion, the MWE-compound demarcation appears to be an element of variation among the languages of the world. One hypothesis that might be worth checking is that this demarcation correlates with the morphological type: with the limited data available so far, one may theorize that the demarcation is clearer in highly inflectional languages displaying root compounding, whereas in more isolating languages the boundary is definitely more blurred, if not absent. This should be checked against a wider typological investigation.

3.2 Competition Issues

Competition in morphology is generally viewed as a relation holding between different word-level strategies (words, affixes, or word formation processes) that compete to realize the same grammatical or lexico-conceptual meaning. However, recent work has claimed that morphological words also compete with MWEs (Booij, 2010, 2017; Hüning & Schlücker, 2015; Masini, in press), since both are lexical units used to achieve (stable) denotations.

The competition between morphological words and MWEs, however, is still underinvestigated. As already emerged in the previous section, compounds are those words that are more readily associated with MWEs, for structural reasons. As Jackendoff (1997, p. 164) notes, “the theory of fixed expressions is more or less coextensive with the theory of words. Toward that end, it is useful to compare fixed expressions with derivational morphology, especially compounds.” Indeed, the research in this area has been mostly focusing on them, so far. For instance:

•

Schlücker and Hüning (2009), Schlücker and Plag (2011), Hüning and Schlücker (2015), and Schlücker (2019) explore the competition between A+N compounds (Vollmond lit. full+moon ‘full moon’) versus phrases (krumme Sachen lit. bent things ‘criminal activities’) in German;

•

The same kind of study is carried out by Booij (2010, 2019) for Dutch (see the compound roodbaard lit. red+beard ‘person with read beard’ vs. the phrasal lexeme rode kaart lit. red card ‘red card (used in football)’);

•

Masini (2019, in press) explores the competition, in Italian, between: (i) N+Prep+N phrasal lexemes (macchina della verità lit. machine of.the truth ‘lie detector’) and N+N compounds (capostazione lit. head+station ‘stationmaster’); (ii) irreversible binomials (sano e salvo lit. healthy and safe ‘safe and sound’) and coordinate compounds (sordomuto lit. deaf+mute ‘deaf-mute’); and (iii) the simile construction with color adjectives (rosso come il fuoco lit. red as the fire ‘red as fire’) and the corresponding compound pattern (rosso fuoco lit. red fire ‘fire-like red’);

•

The competition between (multi-word) similes and N+A compounds (not restricted to color adjectives, though) is also investigated by Hüning and Schlücker (2015) and Schlücker (2019) for German (so weich wie Seide ‘as soft as silk’ vs. seidenweich ‘silky smooth, as soft as silk’); by Hoeksema (2012) and Booij (2019) for Dutch ((zo) recht als een kaars ‘(so) straight as a candle’ vs. hondstrouw lit. dog+faithful ‘faithful as a dog’); by Hyvärinen (2019) for Finnish (hidas kuin etana lit. slow as snail ‘slow as a snail’ vs. jääkylmä lit. ice+cold ‘ice-cold’).¹³

However, word-MWE competition can be conceived also in wider terms: Masini (in press), for instance, compares MWEs (viz. phrasal lexemes) with all kinds of morphological words (simple, derived, and compound). The study, which is based mainly on Italian data, shows that competition is at work between MWEs and words in general (not just compounds)—at different levels of abstraction (specific lexical items vs. patterns of formation) and along different dimensions (synchronic vs. diachronic)—and that it may lead to blocking effects operating in both directions (word ⇆ MWE),¹⁴ thus suggesting a view of the mental lexicon where both words and MWEs are stored on a par with each other.

The next big question is: how does this competition work? What are the principles governing it? Different factors seem to be involved: some have been identified, some others are still to be discovered and understood. From her case studies, Masini (2019, in press) concludes that the competing patterns tend to differentiate, by specializing for different functions, going in the direction argued for by Aronoff (2016, in press), where competition leads to either extinction of one of the competitors or to differentiation in terms of form, meaning, or distribution, as a result of a “struggle for existence” between linguistic expressions. In his above-mentioned study on similes versus compounds in Dutch, Hoeksema (2012) claims that the choice between the two structures is determined by the higher compactness and expressiveness of the compound, by the fact that the compound is not always available (whereas the simile/multi-word option would be), and by differentiation in meaning. Expressiveness, on the other hand, is taken to be more typical of MWEs than compounds by Masini (2019, p. 178) and Schlücker (2019, p. 75). Another factor that plays a role is paradigmatic analogy. Schlücker and Plag (2011) find that the selection between A+N compounds and A+N phrases in German is driven not by general preferences but rather by the individual lexemes involved (cf., also Schuster, 2016): the use of a given noun or adjective in other (already existent) compounds/MWEs seems to influence their use in similar structures when a new A+N formation is coined (e.g., since voll ‘full’ is often used in compounds, when a new voll+N combination is formed, it will tend to be coined as a compound rather than a phrase).

Finally, parts of speech, or word classes, might have a role. Whereas compounding is normally believed to create items belonging to ‘open’ word classes that can be synchronically enriched with new members, MWEs may also belong to more functional word classes (at least in a number of European languages), as already noticed in Section 2.2.1. However, ‘functional’ MWEs are basically the result of diachronic lexicalization (and, in fact, they often undergo univerbation), whereas at least some of the MWEs belonging to open word classes result from a synchronic process of lexical creation. Therefore, synchronically speaking, both word formation and MWEs feed the same (open) classes (as is natural to expect). However, not all classes are equally fed by compounding and MWEs: languages differ in this respect. In Italian, compounds are mostly nouns, and secondarily adjectives, whereas compound verbs and adverbs are basically absent. MWEs, on the other hand, feed also verbs (to a great extent) and adverbs (Masini, 2019; Voghera, 2004). According to De Mauro (1999, p. 1177), in Italian MWEs (intended as phrasal lexemes here) are “the functional equivalent of verbal and nominal compounding in those languages where this process is more active” (my translation), referring to the common view that compounding in Romance is less productive than in Germanic. In this sense, MWEs would be compensating compounding, in a way, especially in specific areas where they are absent (like verbs; cf., Voghera 2004). Along similar lines, Schlücker (2019, p. 74) reports that, according to Fleischer (1996a, p. 152, 1997, pp. 17–20), in German “MWEs are most frequent in the verbal and least frequent in the adjectival domain, with the nominal and the adverbial domain in between” and that this distribution relates to the situation in word formation, where we have a very productive system to form new nouns, but very few pattern to form new verbs (cf., Barz, 2007, p. 28; Fleischer, 1996b, p. 336).

4. Theoretical Debate

The main controversial point regarding MWEs (and morphology) is that they look like the product of syntax but have properties that keep them apart from normal, free phrases and clauses. These properties, as illustrated in Section 2.1, may include fixedness (of some sort and level), idiomaticity of meaning (to different degrees), and conventionalization. If they depart from phrases but still look like phrases/clauses, what are they? Are they words or are they syntactic objects after all? This is the basic question that has characterized much of the theoretically oriented work on some specific MWEs: see, for instance, studies on verb-particle constructions in Germanic languages, which strived to decide on the more syntactic (cf., e.g., Aarts, 1989; den Dikken, 1995) or morphological/lexical (cf., e.g., Dehé, 2002; Los et al., 2012; Stiebels & Wunderlich, 1994; van Marle, 2002) nature of these expressions.

However, how important the answer to this question is largely depends on your theoretical framework, and on how it models the interaction between morphology and syntax. Theories that rely on a modular view of the grammar where morphology and syntax are separate and autonomous components need to find a clear answer. Models that do not presuppose a strict division between morphology and syntax can approach the question in a different (perhaps more relaxed) way: the task in this case wouldn’t be to assign MWEs to this or that component but to grasp the principles that govern their structure and behavior and find a way to account for them with the theoretical tools at their disposal. The latter is the view that is basically expressed by Booij (2002a, b) in two papers on separable complex verbs in Dutch—analyzed as ‘periphrastic word formation’ (cf., Section 2.2.2) —which paved the way to Construction Morphology (Booij, 2010), a morphological theory that assumes the same basic architecture for ‘morphology’ and ‘syntax’,¹⁵ thus allowing for phenomena straddling the boundaries between the two, like MWEs.

In this respect, it is not irrelevant to remind that, back in the Eighties, idioms set in motion Construction Grammar (cf., Hoffmann & Trousdale, 2013), which arose as a criticism of Generative Grammar’s treatment of idioms as a marginal phenomenon. Fillmore, Kay, and O’Connor (1988, p. 504), in what is considered a seminal paper, focus on the “repository of what is idiomatic in the language”, and conclude that this repository is too large just to be ignored or relegated to the periphery of grammar, and linguistic theory. Their solution is to reverse the perspective and to put idiomatic structures at the center of their theoretical formalization, with the hope that “the structure building principles of the so-called core and the machinery for building the phraseological units [. . .] may be of a uniform type, the former being a degenerate instance of the latter” (Fillmore, Kay, & O’Connor, 1988, p. 534). This view—which relies on the notion of ‘construction’ as a conventionalized form-meaning pairing (i.e., a sign)—challenges the largely adopted conception that compositionality is the (only) basic mechanism of language construction (although compositionality is by no means excluded by this model, cf., Michaelis, 2012), a conception which is at odds with the irregularity that characterizes MWEs (cf., among others, Moon, 1998).

More in general, a constructionist approach allows to put the theoretical problem of identifying a sharp divide between words and phrases in the background since all structures of language are regarded as ‘constructions’: what varies is their form (in terms of complexity and schematicity), their function, and their position within the network of constructions that constitutes our linguistic competence. This entails that there can be constructions that share formal properties but not function, and vice versa. In other words, by adopting this perspective, we might be able to account for the fact that a word and a MWE can share function (e.g., concept-naming) but not form, and that a MWE and a phrase can share some aspects of form but not function (see Masini, 2009).

A highly compatible view is put forward by Gaeta and Ricca (2009) in their discussion of compounds as either lexical units or morphological objects. The authors propose to view morphology and the lexicon as distinct entities and formalize this idea with the following classification, based on the two criteria [±morphological] (the property of being the output of morphological operations) and [±lexical] (the property of being a stored unit, by virtue of formal/semantic idiosyncrasies or high frequency) (2009, p. 38):

(15)

According to the authors, the case in (15a) represents prototypical compounds (morphological words that are also stored units with stable referents and unitary meaning), whereas (15d) represents prototypical phrases (free, descriptive sequences with compositional meaning). In the middle, we find (15b), namely, nonce compounds which are formed on the spur of the moment and that don’t necessarily (or are not likely to) get lexicalized, and (15c), which would represent MWEs.

5. MWEs and Cross-Linguistic Variation

Although contrastive, multilingual studies on MWEs are available, especially on European languages,¹⁶ not much has been said on these expressions from a more comprehensive, large-scale typological point of view. Within this domain, a number of questions are still waiting for an answer:

•

Do all languages have MWEs? Is ‘MWE’ a notion that can virtually apply to any language?

•

How diverse are MWEs cross-linguistically? How many types of MWEs are attested in the languages of the world?

•

What is the distribution of MWEs across languages? Does it correlate with morphological or syntactic type? Or with any other property?

•

What is the relation between MWEs and word formation cross-linguistically?

Although this is not the place to give an answer to these questions, it is important to raise them, to start bridging the gap between MWEs and typological investigation.

Within the vast realm of typology, MWEs would arguably fall within Lexical Typology (Koptjevskaja-Tamm, 2008; Koptjevskaja-Tamm, Rakhilina, & Vanhove, 2015), which deals with words, with the meanings they encode, and with the formal processes found in the world’s languages to create new lexical units. Although MWEs are definitely one such process, their role and significance within Lexical Typology has not been explicitly addressed yet.

As a matter of fact, thanks to language description, we already know a lot about possible multi-word patterns across languages (beyond SAE languages), but so far this knowledge has not been gathered and systematized.¹⁷ In the set of possible multi-word patterns, one might include well-known phenomena in both the nominal and verbal domains.

As for the verbal domain, one might cite complex predicates, for instance light verb constructions in languages such as Urdu in (16) or Persian in (17) (cf., Butt, 2003, 2010, for an overview of the “light verb jungle”):

(16)

(17)

One might even ask if other structures such as serial verb constructions (Aikhenvald, 2006) may qualify as MWEs, or if they are rather closer to periphrastic structures as discussed in Section 2.2.2.

Turning to the nominal sphere, some kinds of construct state structures in Semitic might qualify as MWEs. In Modern Hebrew, construct state constructions (formed by a noun plus a second noun in the construct state) are normally classified into ‘compound constructs’ and ‘phrasal constructs’, which “exhibit surface similarities yet distinct syntactic and semantic properties, making it challenging to draw the line between them” (Doron & Meir, 2013; cf., also Borer, 2009). For instance, compound constructs (like (18a)) are normally regarded as semantically more opaque, whereas phrasal constructs (like (18b)) are transparent and productive.

(18)

However, both can be distinguished from other types of (perhaps more ‘prototypical’) compounding patterns, like yeled-sax̠q̱an (lit. child actor) ‘child actor’ or migdal-ʾor (lit. tower-light) ‘lighthouse’ (examples from Edzard, 2013), as well as from genitival phrases like sefer šel ha-more (lit. book of the-teacher) ‘teacher’s book’, where the two nouns are conjoined by the preposition šel ‘of’.

Kageyama (2009) observes that Japanese presents ‘word-level compounds’ and ‘W⁺-level compounds’. The latter look like genuine phrases in many respect (e.g., they are pronounced like two independent words, they are completely productive, and their internal structure is visible to syntax), but at the same time they are still subject to a number of constraints (they resist the insertion of phrasal/functional categories and syntactic manipulation, such as deletion under coordination). So they are not full-fledged phrases, but neither do they behave as word-level compounds, which reminds us of MWEs.

(19)

Rice (2009, p. 547) discusses cases of difficult compound-phrase demarcation in Slave (Athapaskan) that might qualify as lexicalized MWEs. These items look like phrases but have non-compositional semantics, like for instance (20):

(20)

Similarly, Zamponi (2009, pp. 590–591) describes ‘word clusters’ in Maipure-Yavitero (Arawakan), whose constituents (unlike compounds) have retained independent stress, which have become “stable lexicalized expressions with idiosyncratic meaning”. These clusters, which correspond to descriptive noun phrases—either possessive ((21a)) or adjectival ((21b))—are more frequent than compounds, which are scarcely productive in these languages.

(21)

This is just a brief, impressionistic illustration of what appear to be possible MWEs in some typologically and genetically distant languages. Needless to say, the actual MWE-hood of these expressions is purely speculative and would need language-specific case studies to be confirmed. Clearly enough, a systematic study of MWEs in a wide-range cross-linguistic perspective would require a huge effort. Moreover, ideally, it should go hand in hand with a systematic study of word formation processes. Indeed, despite their differences, both complex words and MWEs (especially what we called here phrasal lexemes) are used for concept-naming and for creating new complex lexical units: they may compete with each other, or rather compensate for each other in the expression of lexico-conceptual meanings. By virtue of this common function, it would be desirable to pursue a unified treatment of complex lexical units, not only in linguistic theory and description but also in typological studies, possibly under the auspices of Lexical Typology, which would seem to be the appropriate subfield for a joint investigation of word formation and MWEs.

The success of such an investigation obviously depends on clear definitions. If ‘word’ is too tricky as a cross-linguistically valid concept (Haspelmath, 2011), ‘MWE’ cannot be any easier. This leads us to think that the typological study in question should rather rely on a suitable definition of the common function(s) shared by these constructions, leaving their demarcation to language-specific criteria. In this sense, one possible line of research would be to regard the ‘lexeme’ as a “comparative concept,” namely, a universally applicable concept “defined on the basis of other universally applicable concepts: universal conceptual-semantic concepts, general formal concepts, and other comparative concepts” (Haspelmath, 2010, p. 665). This unified treatment, if actually viable, should eventually lead to find answers to the diversity-related questions on MWEs mentioned at the beginning of this section, but also to a better understanding of what MWEs actually are and why/how they emerge.

Browse All

Share Link

Article contents

Multi-Word Expressions and Morphology

Multi-Word Expressions and Morphology

Summary

Keywords

Subjects

1. Introducing Multi-Word Expressions: A Bare-Bones History

2. A Bird’s-Eye View of MWEs

2.1 Formal and Functional Properties of MWEs

2.2 Types of MWEs

2.2.1 Multi-Word Expressions and Parts of Speech

2.2.2 What About Periphrases?

2.3 Why Do MWEs Matter?

3. Words and MWEs

3.1 Demarcation Issues

3.2 Competition Issues

4. Theoretical Debate

5. MWEs and Cross-Linguistic Variation

Further Reading

References

Notes

Related Articles