Binding constructions to the notion of constituent (and thus constituent structure), quite naturally, is a practice associated with the era of Structuralism.

As stated by Jacobson (1996: 54), some of the first works that made the notion of constituent structure explicit are those by Bloomfield (1984 [1933]) [3], Harris (1946) and Wells (1947). Describing the structural layers of a sentence, constituent structure is equivalent to establishing its hierarchical structure: hence, one can assume that a sentence’s constituent structure corresponds to its hierarchical structure. This is commonly identified by employing the (largely intuitive) procedure of immediate-constituent-analysis (IC-analysis). [4] In order to determine the constituents more objectively, this method is accompanied by syntactic test procedures drawing on distributional (i.e. more “visible”, hence more “objective”) facts, such as permutation (movement) tests, substitution (replacement) tests, elimination (deletion) tests and coordination tests, in short, the most relevant linguistic techniques or methods applied to syntactic analyses in structural linguistics in general and American Structuralism or distributionalism in particular (cf. Harris 1951; Helbig 1973: 239-243; Jakobson 1996: 56; Dürscheid 2003: 48-54, for example).

The concepts of construction and constituent turn out to be closely intertwined, as is also made obvious by Bloomfield’s (1984) understanding of construction:

Whenever two (or, rarely, more) forms are spoken together, as constituents of a complex form, the grammatical features by which they are combined, make up a construction. Thus, the grammatical features by which duke and -ess combine in the form duchess, or the grammatical features by which poor John and ran away combine in the form poor John ran away, make up a construction. (Bloomfield 1984: 169)

From such a description, we can take the term “construction” to refer to the more abstract level of a structural template, the grammatical features (the type) rather than to the specific complex constituent, the respective expression (the token). As Bloomfield’s exemplification also suggests, grammatical constructions can consist of free and bound constituents. This leads to a further subdivision into morphological and syntactic constructions, with the former extending the notion of construction to morpheme combinations (i.e. morphologically complex words), and the latter – consisting of free constituents only – subsuming compound words, phrases and clauses/sentences. It is noteworthy that Bloomfield also draws on meaning when determining that “[a]ny meaningful, recurrent set of such taxemes [that is the features of grammatical arrangement] [5] is a syntactic construction.”(ibid: 184)

More generally, any recurrent (functional) group of constituents is conceived of as a construction, and constituent is defined as any linguistic unit or construction which enters into some larger construction. Hence, within such a (structuralist/distributionalist) model it can be said that all but the smallest constituents are (syntactic) constructions and all but the largest constructions are constituents. That means that here the notion of construction is not restricted to the level of clause, but also subsumes any smaller expression provided that it is still a complex form.

The identification of constituents has been shown to be feasible by means of the test procedures mentioned above. However, this may turn out to be problematic when these procedures do not lead to an unambiguous result, i.e. when one larger construction allows for at least two different layers of structure that usually go with different meanings or interpretations. Turning the perspective, we can say that it is a problem for the analyst when two or more constructions have identical ultimate elements, but different constructional patterns. The latter result in different constituent structures, as is the case in (2) and (3), for example.

Complex ambiguous constructions of this type can be considered constructionally homonymous, i.e. as two constructions the relationship between which is a purely formal one, and these types of “constructional homonymy” must be explicable within a grammatical description of the language in which they occur.

The concept of constituent structure, reflecting the distributional facts of the constituents available in a particular language, was taken up in Generative Grammar, where it surfaced as phrase-structure rules. These, also called rewrite rules, generate the form of a sentence and its constituent phrases, all of which would fall under the concept of construction as conceived of in American Structuralism. For example, we have

Arranged in a tree diagram, the constituents iconically reflect the structural layers of a sentence: the constituents further up in the tree dominate the ones further down in the tree. As pointed out by Radford (1988: 111), this notion of dominance can be used to define the term ‘constituent’: “A set of nodes form a constituent (of some sentence-structure) iff they are exhaustively dominated by a common node.” Basically, this definition is not better and not worse than Bloomfield’s (see footnote 3), who speaks of the containment or inclusion of constituents in a complex form. It must be added here that phrase structure rules alone could only describe the default type of a language’s syntactic structures (declarative, active), and transformational rules were drawn on to cover a language’s total repertoire of syntactic structures.

A more general format of phrase-structure rules, resulting from the recognition of cross-categorial structural symmetry, was later postulated as X-bar syntax or X-bar theory:

The transformational component was likewise generalized to a single rule: move α.

In the X-bar format, we have X standing for any lexical category (V, N, P, A), the head, with complements and adjuncts being such phrases that are subcategorized by the head and optional additions to the head respectively. Specifiers are elements with various functions. If the heads to be specified are Ns or Vs, they act as deictic anchors, linking what is said to the hic et nunc of the speaker. X-bar structure was later also extended to the structure of the functional categories of I (inflection), C (complementizer) and D (determiner), where I has a VP as complement, C – an IP, and D – an NP. [6] Thus, in a generative model of language, this small number of rules – applicable to lexical and functional categories alike – licenses all of a language’s potential constituents. It follows from the understanding of construction as a self-contained unit or as a constituent of a larger unit, that these rules can also be assumed to generate all the constructions occurring in that language, namely heads potentially extended by complements (such as speak a second language) and/or adjuncts (such as the European League) and/or specifiers (a second language , the European League). The postulated rules would, however, only license such constructions as are consistent with this format and the constituents of which may have undergone (permitted) movement. They would, however, exclude from consideration all other expressions that are deviant in some further respect. Perhaps this is one reason why the notion of construction in the more traditional sense of a (clausal or phrasal) pattern that serves a certain, sometimes quite specific, function is notoriously absent in generative models of language. In such models (of syntactic competence), it is sentences that are the key syntactic units generated by phrase-structure rules and X-bar syntax respectively. The latter rules specify how sentences are structured hierarchically, i.e. how sentences are structured out of phrases and phrases out of words (cf. Radford 1988: 123). Within the principles and parameters model, for example, the structural description of an expression distinguishes between D-structure and S-structure. The former provides an arrangement of lexical items on the basis of what phrase structure (X-bar syntax) and predicate-argument structure (a head/predicate and its arguments, resulting from theta-role assignment and subcategorization) license, the latter emerges after potential transformations (“move α”) have been applied and is the structure to which the principles of binding theory apply and from which the phonetic form (PF) and the logical from (LF) are derived. [7] Structural information is thus concentrated in D- and S-structure, and phrase-structure rules as well as transformations represent all the information available and needed to account for or generate the syntactic configurations in a language.

Within the scope of the Minimalist Program, the derivation of a sentence starts from a set of lexical items which are put together in the computational system: words are assumed to be combined into phrases by means of an operation called ‘merge’, and clauses are assumed to be formed by essentially the same merge operation as phrases. The resulting structures commonly shown as tree diagrams provide only partial representations of the overall grammatical structure of the corresponding sentences, namely the words’ categorial status. Word-category information and the operation of merge are obviously not sufficient to specify a language’s potential sentence structures, since there are no rules to licence which word (in a sentence) can merge with which other word(s). The solution lies in suggesting that this information is contributed by further grammatical features carried by words. These are (i) head features (which describe the words’ intrinsic grammatical properties, such as tense for verbs), (ii) complement features (which describe the grammatical complements they take) and (iii) specifier-features (which describe the kinds of specifier/subject they can have). More generally, the complement-features of a word describe its complement-selection properties, and the specifier-features of a word specify the features of the head word of the expression it requires as a specifier/ subject. For a sentence to emerge, the features carried by the words needed/used in a sentence must be compatible with one another. The operation made responsible for this is “feature-checking” (cf. Radford 1997: chapter 3). Thus, quite an amount of structural information (in addition to – or perhaps even making up for – what is covered by X-bar syntax and such parameters as the head parameter in earlier versions of generative grammar) seems now to reside in the words and represents the information on the basis of which the operations of “merge” and “check features” do their work. In such a model, there is no need for constructions, indeed. As Marantz (1995: 380) puts it more radically, “the syntactic engine itself … has begun to fade into the background. Syntax reduces to a simple description of how constituents drawn from the lexicon can be combined and how movement is possible (…).”

Note that, in all versions of generative grammar, the structure of an expression is understood in terms of its syntactic make-up only, with functional considerations (such as the expression’s meaning and use) being absent at that stage. If at all, the word “construction” is used non-technically and refers to particular, predominantly formally determined, clause/sentence types, such as the passive construction [8], the depictive (cf. Zhang 2001, for example), or the resultative constructions (as e.g. in Müller 2002). In its more technical sense, that is construction understood in terms of a form associated with a particular function, the term is no longer needed, since any specific expression which does not follow from the very general rules of the grammatical component (the parameterized principles of Universal Grammar) – and is hence not predictable – is relegated to the lexicon or to the periphery of a language’s grammar, i.e. that part of a language’s grammar that is not part of UG. In contrast to this, a language’s (core) grammar (i.e. the “innate” grammar) is taken to be entirely compositional, so that complex structures can be described as being composed of their formal constituents, and the same procedure or principle is also assumed for their semantic interpretation. From this perspective, Chomsky (1995a: 4) states that “UG provides a fixed system of principles and a finite array of finitely valued parameters. The language-particular rules reduce to choice of values for these parameters. The notion of construction is eliminated, and with it, construction-particular rules”. However, in his 2000 publication, Chomsky concedes grammatical constructions at least the status of taxonomic artefacts that are useful in descriptions of a language, but have no theoretical standing (cf. Chomsky 2000: 8).

To sum up the story thus far, we have found readings of ‘construction’ in (earlier) structuralist models of language that link them to constituency on the basis of formal capacities (permutation etc.) and very general functional features (being head, complement etc.). That is they are seen as a product of applying the general syntactic rules of the language in question, thus drawing on word-categorial information, distributional criteria and very general functional specifications only, with meaning being neglected or even excluded from consideration. In Generative Grammar, a language’s structure found its first generalized expression in the form of phrase structure rules. These were later on sought to be eliminated by X-bar theory. Even later, in the Minimalist Program, X-bar theory was (and still is) subjected to critical analysis. Elaborating on the role of phrase structure theory in minimalist theory, Chomsky concludes that standard X-bar theory can largely be dispensed with: “The structures stipulated in earlier versions are either missing or reformulated in elementary terms satisfying minimalist conditions, with no objects beyond lexical features.” (Chomsky 1995b: 403). At the same time, and quite naturally so, constructions – as forms associated with a certain function – have been rejected on the grounds that the notion is not needed. For a language’s structural repertoire is sufficiently specified and captured by the constituent-forming operations of “Merge” and “Move” and such general (economy) principles as “Shortest Move”, “Greed” and “Procrastinate”, constraining the computational system in the derivation of an interpretable structure from the lexical choices made.

First of all, corpus-linguistic research has revealed that in language, we do not only have structural (i.e. grammatical/syntactic) patterns, but patterns which go beyond a mere reflection of structural regularities. Numerous structures turn out to be repeatedly instantiated by the same lexical material, thus representing lexical patterns. These recurrent or habitual co-occurrences of words come to the fore, especially when larger amounts of texts are analysed. Their existence makes explicit that the language user does not simply insert words into the structural patterns he knows his language to have (conceivable as a slot-filling procedure), but that he has command of lexical chunks, i.e. lexically specified structural patterns. Therefore, it seems reasonable to extend the inventory of constructions (clausal and phrasal) so that it also includes such lexical templates as proverbs, verbal formulae, idioms, and the various types of collocations commonly told apart [9]. This is suggested not merely because they instantiate a particular structural pattern, but because they represent instantiations in which the choice of (all or some of) lexical material is pre-empted. In other words, they are prone to certain idiosyncracies, which make them irregular (or unpredictable) to a larger or smaller extent.

Secondly, if the frequency of co-occurrence of particular words is taken to generally be an indicator of their combinatorial relatedness, also other combinations would seem to qualify as constructions, such as and the, of the, a new (cf. the example quoted above). Biber et al. (1999: 990) call such strings of words “lexical bundles” and define them as “sequences of word forms that commonly go together in discourse”. Turning up in a merely statistical analysis of text, they are not necessarily complete structural units or they even cross structural boundaries. That is why they represent forms not usually described as constituents of a larger syntactic unit: do you want me to, going to be a, there was no significant (cf. ibid: 989), or the other examples just given.

From the current state of our discussion, we would have to reject such bundles as constructions, since they do not meet the requirements of either being a self-contained unit or representing constituents of larger constructs: they fail to pass the syntactic tests listed above and – what is more – they do not exhibit the feature of being functionally related.

Instigated by the data of corpus-based research, the second question raised and briefly touched upon in section 1 becomes a real issue. It is concerned with the actual form of the entities that are considered constructions: are they the “surface forms” of units detectable in verbal expressions, that is the concrete, lexically specified constituents (of larger expressions) or are they the more abstract configurations (be it formulae, rules or whatever) representing the commonalities found in a number of individual expressions. When deciding on their level of abstractness or specificity, we have to focus on the understanding of a construction as a pattern. It has already been said that the recognition of a phenomenon (here a construction) as a pattern makes it a representation of a class of individual entities. As is well-known from research into categorization (cf. Lakoff 1987; Taylor 1995, to give some prominent examples of discussions of categorization from the perspective of language), classification presupposes a number of exemplars to classify and abstraction to some degree, since otherwise we would not be able to recognize the pattern in the first place. From this it follows that individual entities can be said to be instances of a pattern (or exemplars/members of a class or category) rather than being patterns in themselves. For the case of construction this reasoning implies that the individual expression is not a pattern, but an instance thereof. A similar argument can also be taken from Bloomfield, who reserves the term “construction” for the grammatical features by which complex forms are combined (cf. the above quote). Yet, as the lexical patterns revealed by corpus-linguistic research have shown, also individual words occur in recurrent combinations of various forms and sizes. This is not only true of proverbs (birds of a feather flock together, once bitten twice shy), idioms ((straight) from the horse’s mouth, give someone the red carpet treatment, hit the hay, and verbal formulae (such as how do you do, good morning, sleep well, what can I get you etc, but also of collocations (make an appointment, peals of laughter, have lunch) and, of course, also of the “lexical bundles” (and the, a new, going to be a), which are (much) more frequent than the other recurrent forms. Table 1 is meant to illustrate this point. It lists examples, randomly chosen as instantiations of the types of expressions just mentioned and gives the raw frequency data of their occurrence in the British National Corpus (BNC): World Edition (December 2000, SARA Version 0.98), which is a corpus of approximately 100 million words of running text of various text types (written (books, periodicals, magazines etc) and spoken) of British English.

These recurrent lexically specified expressions (with the exception of the last type) also meet the criteria that have (so far) been put up for constructions. Firstly, they all represent functional wholes, and secondly, they are patterns. [10] More precisely, some of them (e.g. (particular types of) collocations) contain variable components, where the pattern would indeed be a first step of abstraction from the specific occurrences, others (such as proverbs and formulae) are often used without any variation, so that they can be said to represent the pattern and its instances at the same time.

We have found a construction in these (non-constructionist) grammars to be understood as (functional) groups of linguistic units. These are of various kinds, which is mainly due to the nature of the linguistic units making them up. This distinction motivates the differentiation between morphological constructions (made up of bound morphemes) and syntactic constructions (consisting of free morphemes) (cf. section 1.1). Moreover, the units’ specificity and schematicity respectively allows for the differentiation of further kinds or types of constructions. Units of a more abstract nature, such as form classes (representing word-categorial information), give us “grammatical constructions”. Units of a more specific nature, such as individual words, result in “lexically filled constructions”. A third kind is a combination of the two, producing “partially lexically filled constructions”. This is an extended version of a reading of construction that Goldberg and Casenhiser (in print) consider the one most commonly accepted in the linguistic community:

... in general, the term construction refers to classes of actual expressions, that is to grammatical patterns. ... [the term is] used to characterize, for example, relative clauses, passives, topicalization, and so on. On this view, a construction is any systematic phrasal pattern of form and function.

( http:www.princeton.edu/~adele/English%20Constructions.rtf accessed 28 February 2006).

In generative models, constructions (as functional units) are banned from the structural component and, depending on the degree of abstraction, associated with other components of the language module. Particular functions linked with general grammatical constructions (such as phrase-structure rules) are considered a matter of interpretation, and (partially) lexically filled constructions do not figure in the structural component at all. They are relegated to the lexicon, the repository of everything idiosyncratic and unpredictable (cf. di Sciullo & Williams 1987: 3; Stowell & Wehrli 1992: 1). This is in agreement with the fairly common and widely accepted association of the notion of construction with forms that are somewhat peculiar with respect to form and/or meaning. Naturally, such constructions are poor candidates for being generated by X-bar syntax. (cf. also Goldberg & Casenhiser, in print, http:www.princeton.edu/~adele/English%20Constructions.rtf accessed 06/02/ 2006).

The remainder of this text will be concerned with the notion of construction as understood by construction grammarians.

Fillmore is one of the first linguists to have sketched out the grammatical framework of Construction Grammar (cf. Fillmore 1988, 1989, Fillmore et al. 1988). Though he situated his approach in the generative tradition, he defined its main tenets contradicting some important assumptions held there. The one central to his argument is the generativists’ rejection of the notion of construction as a necessary construct in their models (cf. section 1.2). It is made obvious that the elimination of the notion of construction in its (intuitive) understanding of structures with a particular use and meaning deprives a linguistic model of the possibility to account for important language data that cannot simply be relegated from the “core grammar” to a periphery neglected by or not included in the model:

... when constructions are interpreted as the products of maximally general rules, no place remains in the grammar for spelling out the non-predictable semantics and pragmatics that is frequently conventionally associated with particular constructions such as those we will describe [the the X-er the Y-er construction, or the let alone construction] (Fillmore et al. 1988: 507)

In other words, Fillmore puts the “old” notion of construction back to centre stage, and, as the quote elucidates, this move is motivated by the attempt to account for what is considered a language’s stock of idiomatic phrases and expressions, whose meaning and use cannot simply be derived from knowledge of their structural make-up (to avoid the ambiguous term “composition”) and the lexical material used in them. If such “exceptional” expressions, which are by no means rare, take the form of fixed expressions it would be conceivable to add them to the lexicon as multi-word expressions that need to be learned. However, as Fillmore et al. note, these exceptional phenomena are not restricted to this type of expressions. They also comprise expressions larger than words

which also have grammatical structure, structure of the kind that we ordinarily interpret by appealing to the operation of the general grammatical rules. This list is not merely a supplement to the lexicon: it contains information about fully productive grammatical patterns, including what have been variously referred to as ‘minor sentence types’, ‘special constructions’, and the like.” (ibid: 504)

Hence, they need to be kept as grammatical patterns, though as special ones which do not result from the application of a few general structural rules to a number of lexical items to be selected for the communication of the intended message or, from the perspective of comprehension, which cannot be understood on the basis of knowledge of the respective rules and words. Therefore, Fillmore (and Fillmore et al.) suggest/s that “a grammar of a language is in large part a repertory of holistic patterns, the language’s grammatical constructions.” (Fillmore 1989: 18, emphasis in the original). The latter are said to refer to expressions whose form, meaning and use is not predictable from knowledge of the general structural make-up of the language at issue and the words used (cf. Fillmore et al. 1988: 507). More technically, Fillmore specifies:

[b]y grammatical construction we mean any syntactic pattern which is assigned one or more conventional functions in a language, together with whatever is linguistically conventionalized about its contribution to the meaning or the use of structures containing it. (Fillmore 1988: 36)

This definition is less concerned with the question of (un)predictability, but more so with the emphasis that constructions are not merely a formal pattern, but formal patterns having particular meanings or uses associated with them. This characteristic, made criterial for the notion of construction, is what shapes the view commonly held in constructionist approaches, the view that constructions are symbolic, i.e. pairings of form and content. By ascribing this symbolic nature to any pattern – i.e. also to the ones that appear to be fully predictable from structural and word knowledge – Fillmore makes constructions the primary unit of a language’s grammar and bridges the gap between what generativists told apart as core and peripheral grammatical phenomena. In his view then “[t]he grammar of language can be seen as a repertory of constructions, plus a set of principles which govern the nesting and superimposition of constructions into or upon one another” (ibid: 37). This complex interplay is thought to be essentially guided by principles of unification and inheritance, and lexical items are linked to the constructions as attracted or required fillers of particular positions or slots therein (cf. ibid: 54). From this conception it follows that Fillmore’s notion of construction does not actually comprise lexically filled expressions. A similar conclusion can be drawn from his dicussion of concepts related to a language’s expressions that are not rule-governed in the generative sense of the term, expressions that are idiomatic (cf. Fillmore et al. 1988: 504-510). In a typology of such expressions, he differentiates between substantive and formal idioms, that is lexically filled idioms with all the elements fixed on the one hand, and lexically open ones with some elements fixed and others free to choose in accordance with a language’s structural principles on the other. [11] Of these, he states, only the latter are “absorbed into the category of grammatical constructions” (ibid: 506). This leaves the fully specified idioms to the lexicon – to be learned and used like other lexical items.

Approaching the development of a linguistic model from a usage-based perspective, Langacker (1987: 494) is mainly concerned with “the actual use of the linguistic system and a speaker’s knowledge of this; [i.e. with his] knowledge of the full range of linguistic conventions”. One of the general tenets of his theory, renamed from “space grammar” into “Cognitive Grammar”, is that grammar is inherently meaningful (cf. Langacker 1996: 51), that is “all valid grammatical constructs are attributed some kind of conceptual import” (ibid: 52). In other words, Langacker does not only consider symbolic the entities making up a language’s lexicon but also its morphological and syntactic structures themselves (Langacker 1987: 12). In his view, lexicon, morphology and syntax are not essentially different in that all of them are concerned with meaningful structures, i.e. with symbolic units that link a form (a signifier) and a meaning (a signified). With this assumption he goes one step further than Fillmore, who (implicitly) keeps grammar and lexicon apart by stating that constructions are the primary units of grammar and lexical entries – the variously preferred fillers for the slots made available by them. In Langacker’s model, the attribution of a particular structure to the one or the other domain is a matter of its specificity/schematicity, entrenchment and symbolic complexity, all of which are gradual phenomena. Hence, a language’s symbolic units comprise units that are different by degree rather than in kind. Lexical items are semantically and phonologically specific, highly entrenched, and of variable complexity, also including a language’s fixed phrases, formulae and (substantive) idioms. Grammatical structures relate more schematic forms to more schematic meanings, are less entrenched and tend to higher complexity (cf. Langacker 1991: 3; 44-46). Here is where we will have to look for constructions, provisionally and broadly defined as functional grammatical patterns.

Not surprisingly, constructions figure prominently in Cognitive Grammar. The term appears in Langacker’s elaboration on the nature of grammatical structure in general and the understanding of grammar as symbolization in particular. He claims that

[g]rammar resides in patterns for combining simpler symbolic structures to form progressively more complex ones. Any such combination is referred to as a construction. It consists of two or more component structures that are integrated to form a composite structure. (Langacker 1991: 5)

This definition licences constructions to comprise both morphological and syntactic composite structures, the criterion being symbolic complexity. In other words, any symbolically complex expression of a language is a construction in Langacker’s understanding, so that constructions subsume both complex lexical items and syntactic structures. With this reading of the term, Langacker goes beyond that given by Fillmore in that there is no need for a construction to exhibit a feature or aspect unpredictable from a speaker’s knowledge of his language’s general structures and words (cf. section 2.1). Technically speaking, one could here also see a link to Bloomfield’s notion of construction, who used the term “construction” to also refer to morpheme combinations, i.e. complex words, and who, moreover, also spoke of the meaningfulness of syntactic constructions (cf. section 1.1).

As regards the actual form assumed for constructions, their definition as any combination of symbolic structures leaves room for the incorporation of all sorts of complex expressions topicalized so far: fully and partially lexically specified idioms as well as complex lexical items and – given that they are symbolic/ meaningful – also the more schematic syntactic patterns or phrase structure rules. These rules figure in Cognitive Grammar as constructional schemas

which are both schematic and symbolically complex. ... a compositional “rule” assumes the form of a constructional schema: a template representing in schematic terms the common relationships among component and composite structures observable across the set of specific expressions that support its extraction. (Langacker 1992: 3 and 6)

It can be summarized for a language’s grammatical patterns that they represent schematic symbolic(ally complex) units and emerge at any level of abstraction. That means they form hierarchies of constructions of various degrees of schematicity, from fully schematic constructions or high-level schemas, such as the English ditransitive pattern [[V] [NP] [NP]], via more specific subschemas, such as [[GIVE/give] [NP] [NP]] or [[SEND/send] [NP] [NP]], to fully specific expressions instantiating the higher level schemas, such as send me a postcard. All of these forms relate to one another, representing a complex category of patterns, “a network whose nodes are constructional schemas linked by categorizing relationships of elaboration and extension.” (Langacker 1991: 7).

The complexity of the ideas associated with the notion of construction also reflects in Langacker’s own extensive definition given in his glossary (Langacker 1991: 548): “Grammatical construction

Langacker’s definition of a construction as any combination of simpler symbolic structures prompts a further question, namely that of whether a construction is always symbolic. Since symbolic structures are understood as relating a semantic and a phonological structure, it is not implausible to assume that also complex semantic and phonological structures per se can be considered constructions. Langacker (1987: 82ff) does not make this point explicit, but the assumption is in accordance with his discussion of composite semantic and phonological structures arising from an integration of the respective components, where he notes that “syntagmatic combination is thus bipolar. But it is also symbolic, for the integration of components at the semantic pole corresponds to, and is symbolized by, the integration of components at the phonological pole. This is true of grammatical constructions in general.” (ibid: 84). Mainly drawing on Langacker’s theory of Cognitive Grammar, Taylor (2002) supports this interpretation by binding the notion of construction to that of analysability: Since not only symbolic, but also semantic and phonological structures can be complex, hence analysable, all three are actually considered constructions (cf. ibid: 561f). Thus, we can assume constructions to exist in the form of complex symbolic, semantic and phonological structures, though the latter are always correlated and do not exist in isolation.

The understanding of constructions (i.e. composite structures) to consist of component structures quite naturally leads to a further issue. How do we have to understand the composition to operate, especially with regard to the integration of the semantic units? Is it that everything of the composite structure can be deduced from its components, or, to put it more bluntly, does the composition of structures follow the principle of compositionality? This principle, usually attributed to Frege and presupposed by formal semantic theories, states that the meaning of a complex expression is fully determined by the meanings of its components and the way they are combined. Langacker, describing this commonly held conception of composition in terms of the “building-block” metaphor (cf. Langacker 1987: 452-460, for example), elaborates that constructions are at best partially compositional in that the component parts just give clues to the meaning of complex structure:

The meaning of a complex expression is typically either more specific than any value regularly derivable from its components, or else it conflicts in some way with such values – ie it consitutes either an elaboration or an extension vis-à-vis its expected ‘compositional’ value. ... Rather than constituting a composite structure, the component structures correspond to certain facets of it, offering some degree of motivation for expressing the composite conception in the manner chosen. (Langacker 1999: 15f)

Hence, also structures the forms of which do not provide straightforward access to their meanings represent composite symbolic structures. That is, all kinds of “peculiar” expressions, such as formal and substantive idioms (to use Fillmore’s terms again) fall under the notion of construction, and the same holds for complex lexical items, such as derivations and compounds, the meanings of which may also be quite opaque. Expressions such as: clean-all, lack-brain, stretch-neck (traditionally known as exocentric or bahuvrihi compounds) may serve as a particularly convincing illustration of the latter (further examples can be found in Tuggy 1987, 2005). Fully compositional expressions, i.e. fully regular expressions (if they exist at all) represent the special case of a “construction” in that they are formed and understood following conventional patterns of composition. In such cases, the composite structure “derives from its component structures solely by virtue of regular compositional principles.” (Langacker 1987: 487).

Goldberg’s name has been associated with construction grammar since the beginning of her career as a linguist in the early 1990ies. Her work is rooted in the framework developed by Lakoff (1987, especially case study 3 on the There-construction), and her 1995 monograph A Construction Grammar Approach to Argument Structure is one of the classics in the field. As she says in the introduction, the book mainly aims at two things: at re-introducing and re-establishing “constructions” as a theoretical concept in linguistic theorizing and at explaining the semantics of (English) clause patterns (cf. Goldberg 1995: 2). Inspired by preceding work on data that show differences in the meanings of slightly different structures despite the employment of the same lexical items, such as the ditransitive and the to-paraphrase patterns (I brought Pat a glass of water vs. I brought a glass of water to Pat), she sets out to elaborate on the meanings rendered by (some of the) English argument structures. That is, she analyses these structural patterns as meaningful units, as constructions. Her understanding and definition of the term ‘construction’ is the same as Fillmore’s (hence, a bit more specific than Langacker’s reading (cf. section 2.2). She also lists as criterial for a construction that at least one of its properties must not be predictable from its constituents/component parts and its formal make-up: “C is a construction iffdef C is a form-meaning pair <Fi,Si> such that some aspect of Fi or some aspect of Si is not strictly predictable from C’s component parts or from other previously established constructions.” (ibid: 4). From this, it follows that the unpredictable form-meaning associations need to be stored or listed to be available to a speaker. Furthermore, this restriction suggests that predictable form-meaning pairs are excised from the inventory of constructions, which is debatable from at least two perspectives: Firstly, such a view is open to implying that predictable pairs are not to be listed as units of a speaker’s linguistic knowledge, because he may process them, i.e. arrive at their interpretation or construct them, as truely compositional expressions. Secondly, if predictable pairs are different from constructions, what then is their place in the language model under construction? On the other hand, under this view, the coverage of the notion of construction is also extended in that the criterion of unpredictability allows for or even necessitates the incorporation of morphemes into the scope/domain of constructions. [12] Here, Goldberg’s notion is wider than Langacker’s, who has the feature of complexity as criterial instead. Still, both views imply that the lexicon can no longer be neatly told apart from a language’s syntax, since constructions comprise elements from both domains.

Also in later discussions of “constructions”, Goldberg maintains her definition that “any linguistic pattern is a construction as long as some aspect of its form or function is not strictly predictable from its component parts or from other constructions recognized to exist” (Goldberg 2003: 219; Goldberg & Casenhiser in print). This is slightly at odds with a more general and less technical description of the phenomenon, where she says: “Constructions are stored pairings of form and function, including morphemes, words, idioms, partially lexically filled and fully general linguistic patterns” (Goldberg 2003: 219). Here fully predictable patterns (the last type listed) are included in the repertory of constructions. She brings this fact into line with her definition by adding that other constructionists extend the notion of construction to frequently occurring fully predictable patterns (cf. ibid; Goldberg & Casenhiser in print). At another place, Goldberg and Jackendoff (2004) give a definition that remedies this debatable point:

Construction Grammar defines constructions to be any stored pairings of form and function; according to this definition, words and morphemes are technically constructions as well. In addition, stored (typically highly frequent) regularities between form and meaning are considered constructions even if they are fully compositional. (Goldberg & Jackendoff 2004: 533, my emphasis)

This understanding, taken up in a similar way in Goldberg (2006: 5), is now closer to Langacker’s in that it comprises any stored form-function pattern, but it remains different from his in that also simple units (morphemes and simple words) are subsumed. A further distinction is hinted at by Goldberg’s criterion of a stored pattern: Langacker uses the term for any composite symbolic structure, no matter whether it is a (stored) unit (in the sense of a ‘mastered routine’), or whether it is a novel, not (yet) entrenched expression categorized by its attribution to an established composite structure, that is an expression not (yet) stored as a linguistic unit of a speaker’s grammar (cf. Langacker 1987: 71-73, 1999: 110-113).

The last notion of construction to be presented here is Croft’s (2001, 2003). A major factor contributing to the development of his Radical Construction Grammar is the recognition of some inconsistencies between the widely-used method of syntactic (distributional) analyses and some basic assumptions held by many syntacticians. They become evident especially when syntactic phenomena are analysed that are thought to exist across languages. Croft (2001: 4) spells out that the method of distributional analysis (widely-used for the identification and definition of syntactic constituents or categories) is at odds with the assumption of syntactic categories as syntactic primitives. He argues that “[i]nstead, constructions are the basic units of syntactic representation, and categories are derived from the construction(s) in which they appear – as the distributional method implies.” This assumptions is the basis for his understanding of what constructions are.

Defining the term, Croft also draws on preceding work by Fillmore et al. (1988), Lakoff (1987) [13], Goldberg (1995) and others, some of which have already been touched upon in the above sections. In a first step, he emphasizes the similarity between constructions and lexical items, which both are considered symbolic, that is they are phenomena linking syntactic and possibly phonological (i.e. formal), as well as semantic, situation and discourse-related (i.e. functional) information. They are also different in that constructions can be schematic (at least partially so) and complex (cf. Croft 2001: 16). Thus, also here – as with Langacker and Goldberg – we find the continuum view of lexicon and grammar. Croft’s inventory of constructions comprises everything from simple words to fully schematic and regular patterns: “construction grammar has generalized the notion of construction to apply to any grammatical structure, including both its form and its meaning.” (ibid: 17).

In a second step, Croft demonstrates how construction grammar analyses the structure of grammatical constructions. Basically following Langacker (also in the schematic representations), he depicts constructions as symbolic units linking forms and functions and equates their internal structure with the syntactic structure exhibited by sentences instantiating them, that means, for example, that the intransitive construction is instantiated by Heather sings. Such a construction represents a part-whole structure, and this is the only structure Croft assumes constructions to have. Abstract syntactic relations, i.e. relations between the elements of a construction, such as its subject and its verb, he claims, can be dispensed with in Construction Grammar (cf. ibid: 18-25). Given the existence of syntactic roles in constructions (i.e. the roles of the constructional elements in the constructional whole, for example), the former turn out not to be necessary, for, in communication, the elements of an utterance’s syntactic structure can be identified via the syntactic roles of the construction, which, in their turn, are triggered by the components of the semantic structure identified for the instantiated construction at issue (for more details see ibid: 205f).

In a third step, Croft elaborates on the organization of constructions in a construction grammar. It has already been pointed out that, in such a framework, any grammatical unit is considered a construction. Starting out from Langacker’s (1987: 73) characterization of a language’s grammar as a “structured inventory of conventional linguistic units”, we should, naturally, assume (and this has already been done by Langacker himself) that constructions – as the basic units of syntactic representation – also represent a structured inventory (cf. also Croft 2001: 25). Hence, constructions can be assumed to form a taxonomic constructional network, with each construction constituting an independent node. Croft points out that the network does not form a strict taxonomic hierarchy, since most of the utterances made are simultaneously instantiations of several distinct constructions, or that usually, from the converse perspective, a construction only specifies some part of the structure of an utterance. Moreover, Croft assumes the organization of constructions into categories in a way also found for other categories, allowing for prototype organization, with a construction displaying a number of related senses (cf. ibid: 25-27). As already argued by Goldberg (cf. 1995: 31-39), this characteristic, i.e. polysemy, is the obvious thing to expect given that there is no strict division between lexicon and syntax. Croft (2003), however, argues that this characteristic may not be true polysemy, since it can be shown that particular constructional senses (demonstrated for the ditransitive construction) are closely associated with particular verb groups. As a result of the data available for the ditransitive construction, Croft postulates the existence of verb-class specific (and even verb-specific) constructions: “there is a distinct syntactic schema for each constructional ‘sense’ specifying the verb classes found with each meaning, with corresponding specific meaning” (ibid: 56). We find a similar, though less rigid, note on such a claim made by Langacker (2000: 3): “low-level schemas … may on balance be more essential to language structure than higher-level schemas representing the broadest generalizations.” There is also evidence from empirical studies for the fact that lower-level constructions play a much more important part in language use (and hence in speaker’s grammars) than commonly assumed (cf. Boas 2003; Hampe & Schönefeld, in print, for example).

A last issue raised in Croft’s discussion of constructions (cf. ibid: 27f) is where in such a taxonomic hierarchy grammatical information is represented. Is information redundantly specified, i.e. with the constructions at each level of schematicity, or is it specified only once with the constructions at the fully (or more) schematic level and inherited by all its subschemas? Both positions are held by proponents of construction grammar and the determination of which is the more plausible one is an empirical question.

In this brief section, the focus is on what may count as the smallest common denominator in the understanding of constructions in the construction grammar framework.

The decisive factor that unites all the notions presented above is the idea that constructions are symbolic units, i.e. units which link a particular form to a particular meaning, with meaning understood in its widest sense as the representation of a particular (semantic and/or discourse) function. Secondly, constructions within this framework are assumed to be related in networks, possibly via inheritance links, in which the indidvidual constructions are related to one another by abstraction/schematization and instantiation on the one hand side, and by extension on the other. They are, however, not assumed to be related via derivation, as is naturally the case in derivational models of language. A third factor relates to a criterion that is considered optional by some grammarians and definitional by others, namely the fact that constructions are complex (or composite) units consisting of component units, whereby the combination is rarely fully compositional with regard to the intended interpretation. This factor licences anything from a complex word up to phrases and clauses as constructions, whereby both lexically specified and more schematic forms may be(come) entrenched as conventional units. At the lexical end of conventional units, it is the specific forms that predominate, at the grammatical end – the partially and fully schematic forms, though, as the analyses of idioms have made obvious, there are also conventionalized lexically specified forms (i.e. substantive idioms) at that end. Grammarians who consider complexity as optional include into the inventory of constructions also all the bound and free morphemes, so that constructions subsume any conventional form-meaning pair, again at various levels of schematicity. This allows for the description of any grammatical structure in a common format, namely as pairs of form and meaning (cf. also Goldberg & Jackendoff 2004). According to Goldberg (2003, 2006), such a view is characteristic of a construction-based theory of language: “the network of constructions captures our knowledge of language in toto, i.e. it’s constructions all the way down.” (2006: 18, emphasis in the original).

The view just quoted is also the one held by most of the authors having contributed to this special issue of Constructions. Linguistic phenomena of various kinds, from intonational contours via antonym pairs to more complex (sentential) constructions, are postulated and demonstrated to be constructions, i.e. (complex) symbolic units, and are analysed in the respective ways. In some of them, more theoretical or more applied perspectives are employed. Others can most easily be characterized as case studies of verbal phenomena from a construction-grammar perspective.

The arrangement in this volume is from the case studies via the more theory-oriented papers to those with an applied focus.

In the first contribution, Egan deals with a group of near-synonymous constructions employing the verb promise. Analysing data from the British National Corpus (BNC), he can isolate a number of distinctive factors which are likely to trigger or motivate the usage of one construction over the other. This is the nominal or pronominal character (associated with the concepts of given and new) of the promisee and promiser NPs, for example. Such distinctive features suggest that the constructions at issue are constructions in their own right.

Gilquin’s contribution is concerned with the 10 most common periphrastic causative constructions in English, such as X CAUSE Y Vto-inf or X MAKE Y Vinf, focussing on the post-object verb slot. In order to identify potential preferences of the respective constructions for particular verbs, Gilquin submits the constructions at issue to a distinctive collexeme analysis. The results reveal a strong correlation between individual constructions and attracted (second) verbs, thus providing a means to distinguish the 10 constructions in a principled way.

Working within a Langackerian framework, Broccias applies a constructionist analysis to a group of constructions known as the “resultative family” (Goldberg & Jackendoff 2004), suggesting the existence of a non-causal change construction as one reading of the “transitive change construction” (SVOC-construction). On the basis of a (conceptual-semantic) analysis of a few instantiations of the SVOC-pattern, he argues that these signal the mere temporal co-extension of the two subevents involved. The shift from causation to temporal co-extension is claimed to be motivated by a construal operation, namely “figure-ground segregation”.

In the next contribution, Gonzalvez applies a constructionist approach to the analysis of Spanish passives of the type X be said/rumoured Y verbless which have no active counterpart. Spelling out the details of their semantics and considering them from the perspective of grammaticalization, he argues for the treatment of passives of this kind as constituting a paradigm of their own rather than in relation to their “non-existent” active counterparts.

Finally in this section, Lemmens focusses on the use of “objectless transitives” (i.e. lexical causative verbs with omitted patient argument), also called “Deprofiled Object Construction”. In his analysis he fruitfully combines two approaches having been taken to this phenomenon: i) analyses taking the lexical properties of either the verbs or the objects as crucial to the allowance of object omission and ii) Goldberg’s constructional analysis of the omission of the patient argument. Also touching the question of what the status of alternations (correspondences between constructions sharing the same verb) is as opposed to the status of surface generalisations, Lemmens’ article can just as well be put in the next, more theory-oriented group of articles, together with Capelle’s and Murphy’s contributions.

Capelle transfers the notion of “allophone/allomorph” (as variants of phonemes and morphemes respectively) to variants at the phrasal level, i.e. to variants of constructions, which he terms “allostructions”. On the basis of an analysis of English verb-particle constructions, he elaborates the notion of allostruction for instantiations of the same partially underspecified more schematic construction and suggests a way to incorporate the more general phenomenon of constructional alternation into a construction-grammar framework by means of the postulation of a new type of link, “alternation”.

Murphy raises the question of how to represent in linguistic theory antonymic relations which exhibit both paradigmatic and syntagmatic properties. Drawing on such features as the observed co-occurrence in contrastive constructions, she argues for an analysis of pairs of canonical antonyms as constructions in their own right.

The final three contributions show what the application of construction-grammar concepts and views to other fields of linguistic analysis can reveal about the phenomena and processes assumed to be effective there.

Diewald fruitfully applies a constructionist approach to the explanation of grammaticalization phenomena. She can show that the semantic-pragmatic notion of context, in many grammaticalization models a central factor to account for grammaticalization processes, can be more comprehensively grasped by the (more complex) notion of construction in that the latter allows for the direct and explicit incorporation of formal factors.

Marandin applies a constructionist approach to the analysis of French intonation. He exploits the construction-grammar notion of inheritance hierarchy for the description and systematization of the French repertoire of intonational contours. This enables him to postulate a hierarchical lexicon of French contours with a grammaticalized meaning and represents a first step towards an integration of intonational contours into grammar.

Last, but not least, Tomasello demonstrates that child language acquisition data and observed and/or postulated acquisitional procedures can be accounted for by assuming a construction-based model of language. In doing so, he both gives a survey of his own usage-based model of language acquisition and provides evidence for the psychological plausibility of the invocation of constructions as the primary form of linguistic knowledge.