En: Croft, William y Alain Cruise. 2004. Cognitive linguistics.

Cambridge: Cambridge
University Press.

9
From idioms to construction grammar

9.1 Introduction

The cognitive linguistic approach to syntax goes under the name of con-
struction grammar. It is not an exaggeration to say that construction grammar
grew out of a concern to find a place for idiomatic expressions in the speaker’s
knowledge of a grammar of their language. The study of idioms led to calls for a
rethinking of syntactic representation for many years before construction grammar
emerged, and some of this work will be referred to in this chapter. At least partly
independently of construction grammar, a number of researchers have emphasized
the need to represent linguistic knowledge in a construction-like fashion. But in
cognitive linguistics, these concerns led to a grammatical framework in which all
grammatical knowledge is represented in essentially the same way. This chapter
presents the arguments for a construction grammar.
Construction grammar, like any other scientific theory, did not arise in a theoreti-
cal vacuum. Construction grammar arose as a response to the model of grammatical
knowledge proposed by the various versions of generative grammar over the period
from the 1960s to at least the 1980s, and other syntactic theories that emerged as
direct offshoots of generative grammar. (These models in turn represented exten-
sions of the organization of a traditional descriptive grammar of a language, albeit
with significant changes in terminology.)
In most theories of generative grammar, a speaker’s grammatical knowledge is
organized into components. Each component describes one dimension of the prop-
erties of a sentence. The phonological component, for example, consists of the rules
and constraints governing the sound structure of a sentence of the language. The
syntactic component consists of the rules and constraints governing the syntax –
the combinations of words – of a sentence. The semantic component consists of
rules and constraints governing the meaning of a sentence. In other words, each
component separates out the specific type of linguistic information that is contained
in a sentence: phonological, syntactic and semantic. In addition, all versions of
Chomskyan generative grammar have broken down the syntactic component fur-
ther, as levels or strata (such as ‘deep structure,’ later ‘D-structure,’ and ‘surface

225
226 Cognitive approaches to grammatical form

structure,’ later ‘S-structure’; Chomsky 1981) and modules or theories (such as

Case theory, Binding theory etc.; Chomsky 1981).
Further components have been proposed by other linguists. Some have argued
that morphology, the internal formal structure of words, should occupy its own
component (e.g. Aronoff 1993). Others have suggested that information structure,
that is, certain aspects of discourse or pragmatic knowledge, should have its own
component (Vallduvı́ 1992). However many components are proposed, the general
principle remains: each component governs linguistic properties of a single type:
sound, word structure, syntax, meaning, use.
From our point of view, the number of different components is not as crucial as
the fact that each type of linguistic knowledge is separated into its own component.
We may describe this as a ‘horizontal’ model of the organization of grammatical
knowledge, following its typical diagrammatic representation:

(1) phonological component

syntactic component

semantic component

In addition to these components, there is the lexicon, which characterizes the

basic units of syntactic combination. The lexicon differs from these components
in that the lexicon gives, for each word, its sound structure, its syntactic category
(which determines how it behaves with respect to the rules of the syntactic com-
ponent) and its meaning. Thus, a lexical item combines information from the three
components in (1) (and can include information from other components, such as its
morphological structure and its stylistic pragmatic value). It represents a ‘vertical’
component as against the ‘horizontal’ components:

(2) phonological component

lexicon

syntactic component

semantic component

The components are intended to be highly general rules that apply to all struc-
tures of the relevant type. Thus the rules of the phonological component apply
to all word forms and all phonological phrases (for prosodic and other phrasal
 From idioms to construction grammar 227

phonology); the rules of the syntactic component apply to all sentences and
sentence types; and the same applies to rules for other components.
Of course, there must be some general way to map information from one compo-
nent onto another; for instance, there must be a way to map the syntactic structure
of a sentence onto the semantic structure of the meaning conveyed by the sentence.
These rules are called linking rules, and are also intended to be highly general,
applying to all sentences of the language. One might ask at this point, why are
the linking rules just a bunch of rules that link components, while the components
define the way that grammatical knowledge is divided up in the speaker’s mind?
As we will see, that is essentially the question that construction grammar asks.
The response of the generative grammarians is that the rules inside each compo-
nent are so highly intertwined and self-contained that they represent a cohesive
structure relative to the linking rules (and if they are not so highly intertwined, the
components are broken down further into levels, modules etc.).
In sum, the final model of the organization of grammatical knowledge in the sorts
of syntactic theories prevalent from the 1960s to the 1980s will look something
like the diagram in (3):

(3)
phonological component
linking rules
lexicon

syntactic component
linking rules
semantic component

One of the crucial characteristics of this model is that there are no idiosyncratic
properties of grammatical structures larger than a single word. Phrases and sen-
tences are governed by the highly general rules of the syntactic component and
their counterparts in the semantic and phonological components, and the equally
highly general linking rules. On the other hand, words represent an arbitrary and
idiosyncratic joining of form (phonological and syntactic) and meaning. The re-
striction of arbitrariness in grammar to the lexicon is a central principle of gener-
ative grammar, reiterated in recent versions of generative grammar (e.g. Chomsky
1993:3, 4).
One of the consequences of this model is the rejection of the concept of con-
struction in the traditional grammar sense of that word. In traditional grammar,
one describes a syntactic structure such as is found in the sentence in (4a) as ‘the
passive construction’:

(4) a. Janet was promoted by the company.

b. [Subject be Verb-PastParticiple by Oblique]
228 Cognitive approaches to grammatical form

The passive construction would be described as necessarily possessing the com-

bination of syntactic elements given in (4b), including the subject noun phrase,
the passive auxiliary verb be in some form, a verb in the past participle form, and
(optionally in the case of English) a prepositional phrase with the preposition by. In
addition, a traditional characterization of the passive construction would indicate
that the agent of the action is expressed by the object of the prepositional phrase,
and the undergoer is expressed by the subject.
In the generative model, as many of these properties of the passive construction
as possible would be described by the general rules of the various components, and
any idiosyncratic properties would be placed in the lexicon. For example, the fact
that the subject precedes the verb is true of a large class of constructions in English
(see [5]); the fact that the auxiliary follows and is in a finite form in contrast to
the verb is also true of a large class of constructions (see [6]), and the fact that
the prepositional phrase follows the verb (and also a noun) it modifies, and the
preposition governs the object form of the noun phrase, is also true of a large class
of constructions (see [7]):
(5) a. Active: John ate.
b. Relative Clause: the tart that John ate . . .
c. Adverbial Clause: before John ate the tart . . .
d. Conditional: If John ate a tart, then I will have a tart as well.
e. Comparative: John ate a bigger tart than I did.
(6) a. Perfect: John has eaten the tart.
b. Progressive: John is eating a tart.
c. Future: John will eat [infinitive] a tart.
d. Modal: John might eat [infinitive] a tart.
(7) a. Oblique Adjunct Phrase: John ate the tart with a fork and spoon.
b. Prepositional Complement Phrase: John put the tart in the refrigerator.
c. Circumstantial Phrase: John ate the tart in the living room.
d. Nominal Prepositional Phrase Modifier: the tart on the table . . .

The logical conclusion of this process of analysis is the hypothesis that all
properties of syntactic constructions – that is, a grammatical structure larger than
just a single word – can be captured with the general rules of the grammatical
components and their interfaces, and thus there is no need for constructions in
grammatical analysis. Chomsky makes this claim explicit:
a central element in the work discussed here, as in recent work from which it
evolves, is the effort to decompose such processes as “passive”, “relativization”,
etc., into more fundamental “abstract features”. (Chomsky 1981:121)
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 grammatical construction is eliminated, and with it,
construction-particular rules. (Chomsky 1993:4)
 From idioms to construction grammar 229

Chomsky’s position on the generality of syntax and the irrelevance of construc-

tions to the analysis of grammar is the complement of his view that all arbitrary
and idiosyncratic aspects of grammar should be restricted to the lexicon.
The componential model of grammatical organization given above is charac-
teristic of the generative grammar and its offshoots. As we noted, in Chomskyan
theory the syntactic component is internally complex, beginning with levels in the
earliest versions and a further subdivision into modules in Government and Bind-
ing theory (Chomsky 1981). The most recent version, Minimalist theory (Chomsky
1993, 1995), apparently ends the internal organization of the syntactic component
and recasts the phonological component as an ‘articulatory-perceptual interface,’
which links the language faculty to the perceptual-motor system and the seman-
tic component as a ‘conceptual-intentional interface,’ which links the language
faculty to other human conceptual activity. Nevertheless, it appears that the broad
division into the three components in (3) remains, even if two components are
now thought of in terms of their embedding in the cognitive system as a whole,
and the third (syntactic) component is now the whole language system in between.
Likewise, the notion of the lexicon as the repository of idiosyncratic information
remains, and as such provides information linking the three components together
(Chomsky 1993:3; 1995:235–36).
Other syntactic theories that diverge from Chomskyan theory also retain the
organization into components. Earlier offshoots such as Relational Grammar
(Perlmutter 1983) and its offshoot Arc-Pair Grammar (Johnson and Postal 1980)
included multiple levels in the syntactic component but retained the separation
of components. Later offshoots, such as Lexical-Functional Grammar (Bresnan
1982), Generalized Phrase Structure Grammar (Gazdar, Klein, Pullum and Sag
1985; see also Pollard and Sag 1993) and Categorial Grammar (Wood 1993), re-
ject the concept of multiple levels but still retain the separation of components. In
more recent theories of non-Chomskyan syntax, there is a shift in emphasis from
separation of components to their interaction. But the development of construction
grammar marks a more direct break from the componential view of grammatical
organization.

9.2 The problem of idioms

Certain exceptions to the principle of the generality of rules governing

larger grammatical structures have been made in the history of generative grammar.
For example, the different syntactic structures required by different verbs, such
as those illustrated in examples (8)–(9), must be represented somewhere in the
grammatical model in (3):
230 Cognitive approaches to grammatical form

(8) a. Tina slept.

b. *Tina slept bananas.
(9) a. David consumed the bananas.
b. *David consumed.

Until at least recently (e.g. Haegeman 1994:40–42), generative grammar has

accounted for this pattern of distribution by subcategorization frames associated
with lexical items in the lexicon, as in (10a–b):
(10) a. sleep: V, [ – ]
b. consume: V, [ – NP]

The effect of this device is to include phrasal syntactic information in the lexicon,
under the lexical entry for each verb. However, this way of handling the distribu-
tional patterns in (8)–(9) is consistent with the general principle that idiosyncratic
information is to be found in the lexicon.
There is another class of syntactic phenomena that poses a much greater problem
for the componential model of grammar and the principle that all grammar above
the word level can be explained by highly general rules. These are idioms. Idioms
are, by definition, grammatical units larger than a word which are idiosyncratic in
some respect. Some examples of idioms are given in (11):
(11) a. It takes one to know one.
b. pull a fast one
c. bring down the house
d. wide awake
e. sight unseen
f. all of a sudden
g. (X) blow X’s nose
h. Once upon a time . . .

It is difficult to give a precise definition of the category of idioms, for reasons

that will soon become clear. Nunberg, Sag and Wasow (1994:492–93) offer a
prototype definition of idioms with one necessary feature and a number of typical
features. The necessary feature is conventionality: ‘their meaning or use can’t
be predicted, or at least entirely predicted, on the basis of a knowledge of the
independent conventions that determine the use of their constituents when they
appear in isolation from one another’ (492). The other, typical, properties of idioms
they list are:
(12) a. Inflexibility: restricted syntax, as in shoot the breeze vs. *the breeze is hard to
shoot
b. Figuration: figurative meaning, as in take the bull by the horns, lend a hand
c. Proverbiality: description of social activity compared to a concrete activity, as
in climb the wall, chew the fat, spill the beans
 From idioms to construction grammar 231

d. Informality: typically associated with informal speech styles or registers

e. Affect: usually have an evaluation or affective stance towards what they de-
scribe

It is the necessary property of idioms that Nunberg et al. identify – their conven-
tionality – which is the relevant property of idioms with respect to the componential
model of grammar. If expressions such as those listed in (11a–h) are conventional,
then they must somehow be stored as such in a speaker’s mind. If so, then idioms
are part of a speaker’s grammatical knowledge. However, at least some aspects of
an idiom cannot be predicted by the general rules of the syntactic and semantic
components and their linking rules (we will leave out the phonological component
for the time being). Hence they pose a problem for the componential model. It
is possible to make certain sorts of stipulations to handle the conventionality of
idioms (cf. Nunberg et al. 1994:507). However, a more general treatment would
be preferable to such stipulations, if such a general treatment were available.
The linguists who ended up proposing the original construction grammar (in
Fillmore, Kay and O’Connor 1988) approached the problem of idioms from the
opposite direction. Instead of treating idioms as a problematic phenomenon from
the point of view of the componential model of grammar, they analyzed the wide
variety of idioms, and their analysis became the basis for a new model of grammat-
ical organization. The remainder of this chapter will take the reader from idioms
to construction grammar.1
Idioms can be characterized in many different ways. The description and clas-
sification that we will begin with is drawn from Fillmore et al. 1988, who used
their analysis to argue for a construction grammar. Fillmore et al. begin with three
features that can be used to classify idioms. The first feature they describe, drawn
from Makkai 1972, is the distinction between encoding and decoding idioms.
An encoding idiom is one that is interpretable by the standard rules for in-
terpreting sentences, but is arbitrary (i.e. conventional) for this expression with
this meaning. Examples given by Fillmore et al. are answer the door, wide awake
and bright red. These are all expressions that a hearer could figure out upon hear-
ing them. However, a speaker would not have guessed these expressions are the
natural-sounding English way to describe ‘open the door in response to some-
one knocking,’ ‘completely awake’ and ‘intense color.’ Another way of looking at

1 Fillmore et al. were not the first linguists to analyze idioms in a systematic way. There is a vast
literature on idioms, particularly in Europe where the study of idioms is called ‘phraseology.’ Another
important line of research on idiomatic expressions is the Firthian research on collocations. Nor were
Fillmore et al. the first to perceive the problem of idioms for a componential model of grammar, and
to propose an alternative; one such antecedent is Becker (1975). However, as in the case of many
scientific ideas, variants of the idea are proposed independently but commonly only one variant is
propagated through the scientific community (Hull 1988).
232 Cognitive approaches to grammatical form

encoding idioms is from the point of view of someone learning a foreign language.
For instance, an English learner of Spanish would not be able to give the correct
way of asking ‘how old are you?’, if s/he did not know it already; but if s/he heard
a Spanish speaker say Cuantos años tiene? (lit. ‘how many years do you have?’),
s/he would very likely figure out what the speaker meant.
A decoding idiom is one that cannot be decoded by the hearer: a hearer will
not be able to figure out the meaning of the whole at all from the meaning of
its parts. Fillmore et al. give the examples of kick the bucket and pull a fast one.
Any decoding idiom is also an encoding idiom: if as a hearer you cannot figure
out what it means, then you are also not going to be able to guess that it is a
conventional way to express that meaning in the language. One of the reasons that
a decoding idiom is a decoding idiom is because there are not any correspondences
between the literal and idiomatic meaning of the parts of the decoding idiom. For
example, kick the bucket is a transitive verb phrase, but its idiomatic meaning is
the intransitive ‘die,’ and there is nothing that corresponds even metaphorically to
a bucket.
The encoding/decoding idiom distinction corresponds rather closely to Nun-
berg, Sag and Wasow’s distinction between idiomatically combining expressions
and idiomatic phrases respectively (Nunberg, Sag and Wasow 1994:496–97).
Idiomatically combining expressions are idioms where parts of the idiomatic
meaning can be put in correspondence with parts of the literal meaning. For in-
stance, in answer the door, answer can be analyzed as corresponding to the action
of opening, and of course the door denotes the door. In an idiomatically combining
expression such as spill the beans, meaning to divulge information, spill can be
analyzed as corresponding to ‘divulge’ and the beans to ‘information.’ In contrast,
no such correspondences can be established for kick and the bucket in kick the
bucket (‘die’). Nunberg et al. call the latter idiomatic phrases.
The encoding/decoding distinction is not the same as the idiomatically combin-
ing expression/idiomatic phrase distinction, however. Some idioms, such as spill
the beans, are encoding idioms even though they are idiomatically combining ex-
pressions. The encoding/decoding distinction is rather vaguely defined: it refers to
how clever (or lucky) the hearer is in decoding an expression of the language. For
this reason Nunberg et al.’s distinction is preferable.
The encoding/decoding distinction, as well as the idiomatically combining ex-
pression/idiomatic phrase distinction, characterizes idioms in contrast to regular
syntactic expressions with respect to the interpretation rules linking the syntac-
tic component to the semantic component. With idiomatic phrases such as kick
the bucket, the interpretation rules cannot apply because the parts of the syntactic
phrase do not correspond to parts of the semantic phrase at all. With idiomatically
combining expressions such as spill the beans, the parts of the syntactic phrase
 From idioms to construction grammar 233

correspond to semantic elements, but only in an interpretation that is unique to the

idiomatically combining expression (spill does not mean ‘divulge’ except in the
idiom spill the beans). Thus, the idiomatic meaning of idiomatically combining
expressions cannot be determined from the general rules of semantic interpretation
for the words or for the syntactic structure.
The second distinction that Fillmore et al. offer for defining idioms is between
grammatical and extragrammatical idioms. Grammatical idioms are parsable by
the general syntactic rules for the language, but are semantically irregular (i.e.
they are encoding or decoding idioms). Examples include those discussed so far,
such as kick the bucket and spill the beans, and also examples such as (X) blows
X’s nose. All of these idioms follow the general English syntactic rule that di-
rect objects follow the verb, and that a possessive modifier precedes the noun it
modifies.
Extragrammatical idioms are idioms that cannot be parsed by the general syn-
tactic rules for the language. Fillmore et al. give as examples of extragrammatical
idioms first off, sight unseen, all of a sudden, by and large and so far so good. One
might think that extragrammatical idioms are rare, but Nunberg et al. suggest that
they may not be particularly rare. Nunberg et al. provide a sampling of extragram-
matical idioms, which are given here (Nunberg et al. 1994:515; note there is only
one idiom overlapping with Fillmore et al.’s list):

(13) by and large; No can do; trip the light fantastic; kingdom come; battle royal;
Handsome is as handsome does; Would that it were . . .; every which way; Easy
does it; be that as it may; Believe you me; in short; happy go lucky; make believe;
do away with; make certain

The grammatical/extragrammatical distinction characterizes idioms in contrast

to regular syntactic expressions with respect to the rules of the syntactic component.
Grammatical idioms conform to the syntactic rules, but are idiomatic in some other
fashion. Extragrammatical idioms do not conform to the syntactic rules, and for
that reason alone are idiomatic.
Fillmore et al.’s third distinction is between substantive and formal idioms. A
substantive, or lexically filled, idiom is one in which all elements of the idiom are
fixed. For example, the idiom It takes one to know one is completely fixed; one
cannot even alter the tense (*It took one to know one). A formal, or lexically open,
idiom is one in which at least part of the idiom can be filled by the usual range
of expressions that are syntactically and semantically appropriate for the slot. For
example, with the idiom (X) blows X’s nose, the expressions I have described as
X can be filled by a noun phrase (and a corresponding coreferential possessive
pronoun) that refers to a person possessing a nose: I blew my nose, Kim blew her
nose, They all blew their noses and so on. Fillmore et al.’s use of the term ‘formal’
234 Cognitive approaches to grammatical form

corresponds to Langacker’s term schematic to indicate a more general category

(see §3.2), and we will use the term ‘schematic’ here.
Fillmore et al. note that one potential confusion with respect to the substan-
tive/schematic distinction is that there may be a substantive idiom that fits the
pattern of a counterpart schematic idiom. For example, they posit a schematic
construction which they loosely describe as The X-er, the Y-er. Examples of this
schematic construction are given in (14a–c):

(14) a. The more you practice, the easier it will get.

b. The louder you shout, the sooner they will serve you.
c. The bigger the nail is, the more likely the board is to split.

There is also a substantive idiom that fits the pattern of the The X-er, the Y-er
schematic idiom:

(15) The bigger they come, the harder they fall.

The existence of the schematic idiom The X-er, the Y-er does not preclude the
existence of a substantive idiom like (15), just as the existence of a general syntactic
rule where the direct object follows the verb does not preclude the existence of an
idiom such as kick the bucket.
The substantive/schematic distinction characterizes idioms in contrast to regular
syntactic expressions on the one hand and the lexicon on the other. Both substantive
and schematic idioms have parts that are lexically completely specified, although
schematic idioms have parts that are specified in syntactic terms (that is, by a
syntactic category such as ‘noun phrase’ or ‘possessive pronoun’). In contrast,
syntactic rules make reference only to general syntactic categories such as V
(verb), NP (noun phrase) and so on, as in the phrase structure rules given in (16)
for simple active intransitive and transitive sentences:

(16) a. S → NP VP
b. VP → V
c. VP → V NP

The last distinction that Fillmore et al. give is for idioms with or without prag-
matic point (Fillmore et al. 1988:506). Idioms with pragmatic point are idioms
that, in addition to having a meaning in the usual sense of that term, also are
specifically used in certain pragmatic contexts. Obvious examples of idioms with
pragmatic point are idioms used for opening and closing conversations such as
Good morning or See you later, and for other specialized discourse contexts
such as telling a fairy tale (Once upon a time . . .; ibid.). Other idioms with
pragmatic point are those that have a certain conventional pragmatic content, as
with the schematic idiom illustrated by Him be a doctor?! On the other hand,
 From idioms to construction grammar 235

many other idioms such as all of a sudden do not have any specific pragmatic
point.
The with/without pragmatic point distinction characterizes idioms with respect
to the ‘information structure’ or ‘discourse’ component that some linguists have
argued for. They demonstrate that some idioms have conventional information-
structure or discourse-contextual properties associated with them, which again
cannot be predicted from general pragmatic or discourse-functional principles.
For example, it may be a general pragmatic principle that in taking leave, one
may make reference to a future meeting; but it cannot be predicted that the
specific phrase See you later is conventionally used in English for that pur-
pose, whereas in Spanish Hasta luego (lit. ‘until later’) is used for the same
function.
Fillmore et al.’s analysis demonstrates that idioms are quite varied in their
syntactic, semantic and pragmatic properties, ranging from completely fixed ex-
pressions to more general expressions, which may be semantically more or less
opaque and may not even correspond to the general syntactic rules of the language.
The distinctions discussed above are summarized in (17):

(17) a. encoding vs. decoding

b. idiomatically combining expressions vs. idiomatic phrases
c. grammatical vs. extragrammatical
d. substantive vs. schematic (formal)
e. with pragmatic point vs. without pragmatic point

Fillmore et al. use the features given above for a final, three-way categorization
of idioms. Their first category of idioms are unfamiliar pieces unfamiliarly ar-
ranged. The new aspect of this definition is the fact that certain words occur only
in a idiom. Examples of (substantive) idioms with unfamiliar pieces are kith and
kin ‘family and friends’ and with might and main ‘with a lot of strength.’ In other
words, such idioms are lexically irregular as well as syntactically and semantically
irregular. Unfamiliar words are by definition unfamiliarly arranged: if the words
do not exist outside the idiom, then they cannot be assigned to a syntactic category
in terms of a regular syntactic rule. Also, unfamiliar words unfamiliarly arranged
are by definition semantically irregular.
However, an idiom containing unfamiliar pieces unfamiliarly arranged does
not imply that it is an idiomatic phrase; such an idiom can be an idiomatically
combining expression. This point is made clearer by the schematic idiom of this
type given by Fillmore et al., the idiom The X-er, the Y-er illustrated in (14) above.
The unfamiliar pieces are the two occurrences of the, which are not definite articles
(in fact, they come from the Old English instrumental demonstrative þy). The
unfamiliar arrangement is the parallel syntactic structure, with a degree expression
236 Cognitive approaches to grammatical form

followed by a clause with a gap corresponding to the degree expression. In (14a) for
example, the parallel gapped structures are more . . . you practice and easier . . .
it will get ). Nevertheless, this idiom is an idiomatically combining expression,
in that the parts of the construction can be made to correspond with the parts of
its meaning (roughly, ‘the degree to which you practice determines the degree to
which it gets easy’).
Fillmore et al.’s second category of idioms are familiar pieces unfamiliarly
arranged. These idioms do not contain unique words but are extragrammatical. In
other words, such idioms are lexically regular, but syntactically and semantically
irregular. Fillmore et al. give all of a sudden and in point of fact as examples of
substantive idioms in this category. They give as an example of a schematic idiom of
this category the phrase Nth cousin (M times removed): this is a syntactically unique
construction. Again, idioms made up of familiar pieces unfamiliarly arranged may
be idiomatically combining expressions.
Fillmore et al.’s third and last category of idioms are familiar pieces familiarly
arranged. Such idioms are lexically and syntactically regular but semantically
irregular. Again, such idioms may be substantive or schematic; Fillmore et al. give
examples of both types. The substantive idioms they list are in fact not entirely
fixed expressions; they include pull X’s leg (which can have any person-denoting
noun phrase as X) and tickle the ivories ‘play the piano’ (which can be inflected
for tense/mood). Fillmore et al.’s schematic idioms are even more schematic; they
include what they call ‘fate tempting expressions’ such as Watch me (drop it, slip
etc.).
The types of idioms, and their comparison to regular syntactic expressions, are
given in Table 9.1.

Table 9.1 Types of idioms compared to regular syntactic expressions

Lexically Syntactically Semantically

Unfamiliar pieces unfamiliarly arranged irregular irregular irregular

Familiar pieces unfamiliarly arranged regular irregular irregular
Familiar pieces familiarly arranged regular regular irregular
Regular syntactic expressions regular regular regular

9.3 Idioms as constructions

Having presented their analysis and classification of idioms, Fillmore et

al. argue that the proper way to represent speaker’s knowledge of idioms is as
constructions. For Fillmore et al., a construction is a schematic idiom. That is,
 From idioms to construction grammar 237

some elements of the construction are lexically open on the one hand, and so the
idioms fitting the description cannot simply be listed as ‘phrasal lexical items.’ In
this respect, schematic idioms differ from substantive idioms. Fully substantive
idioms, such as It takes one to know one or The bigger they come the harder they
fall, can simply be listed as lexical items. Listing substantive idioms would require
the allowance for multiword lexical items in the lexicon. But this concession to
the linguistic facts does not conflict greatly with the principle of componential
grammar that arbitrary and idiosyncratic linguistic knowledge is found in the
lexicon (cf. the discussion of subcategorization frames in §9.2). Hence substantive
idioms do not require any drastic departure from the componential model of the
organization of grammar.
Schematic idioms, on the other hand, cannot simply be listed in the lexicon.
And schematic idioms are idioms; that is, they are semantically and possibly
also syntactically and lexically irregular. Syntactic, semantic and in some cases
pragmatic properties of schematic idioms cannot be predicted from the general
rules of the syntactic and semantic components (and the pragmatic component)
or the general rules linking these components together. Instead, the syntactic,
semantic (and in some cases pragmatic) properties must be directly associated
with the construction. Such a representation would cut across the components in
the componential model of grammatical knowledge, and hence represents a direct
challenge to that model, at least for idioms.
Fillmore et al. make the case for constructions as units of syntactic represen-
tation by examining one construction in great detail, the construction containing
the conjunction let alone, and demonstrating that it has syntactic, semantic and
pragmatic properties that cannot be described by the general rules of the language,
but is rule-governed within the context of the let alone construction and certain re-
lated constructions. The following discussion will present some of the more salient
unique properties of the let alone construction.
The syntax of the let alone construction is complex. Let alone can be described
as a coordinating conjunction; like other conjunctions, it conjoins a variety of
like constituents (Fillmore et al. 1988:514; the emphasized elements represent
prosody):

(18) a. Max won’t eat shrimp, let alone squid.

b. We’ll need shrimp and squid.
(19) a. Max won’t tou ch the shrimp, let alone clean the squid.
b. I want you to cook the shrimp and clean the squid.

However, let alone fails in some syntactic contexts where and is fine, and vice
versa (Fillmore et al. 1988:515–16; Fillmore et al. also discuss WH-extraction and
It-clefts):
238 Cognitive approaches to grammatical form

(20) a. Shrimp and squid Moishe won’t eat.

b. *Shrimp let alone squid Moishe won’t eat.
c. *Shrimp Moishe won’t eat and squid.
d. Shrimp Moishe won’t eat, let alone squid.

Example (20d) shows that let alone allows sentence fragments for the second
conjunct. In this respect let alone is like certain other conjunctions, including
comparative than (Fillmore et al. 1988:517, 516):

(21) a. John hardly speaks Russian let alone Bulgarian.

b. John speaks Russian, if not Bulgarian.
c. John speaks better Russian than Bulgarian.

However, unlike comparative than and ordinary conjunctions, let alone is impossi-
ble with VP ellipsis (deletion of the verb phrase excluding the auxiliary; Fillmore
et al. 1988:516):

(22) a. Max will eat shrimp more willingly than Minnie will.
b. Max won’t eat shrimp but Minnie will.
c. *Max won’t eat shrimp let alone Minnie will.

The let alone construction is a focus construction, like a number of other con-
structions of English (see Prince 1981b, discussed below), hence its characteristic
prosody. In fact, let alone is a paired focus construction, like those given in (23b–c)
(Fillmore et al. 1988:517):

(23) a. He doesn’t get up for lunch , let alone breakfast .

b. He doesn’t get up for lunch , much less breakfas t.
c. She didn’t eat a bi te, never mind a wh ole meal.

The let alone construction allows for multiple paired foci (see [19a]), and in such
sentences allows multiple let alones (Fillmore et al. 1988:520):

(24) a. You couldn’t get a poor man to wash your car for two dollars, let alone a rich
man to wax your truck for one dollar.
b. You couldn’t get a poor man, let alone a rich man, to wash, let alone wax, your
car, let alone your truck, for two dollars, let alone one dollar.

In this respect, let alone is similar to the construction not P but Q (illustrated
in [25]) and to the respectively construction (illustrated in [26]); but Fillmore et
al. argue that in other respects let alone differs from both of these constructions
(1988:521–22):

(25) Ivan sent not an album but a book, and not to Anna on her anniversary but to
Boris on his birthday.
(26) Fred and Louise hated their shrimp and squid respectively.
 From idioms to construction grammar 239

Let alone is a negative polarity item, not unlike any; it occurs in negative contexts
and certain other contexts (Fillmore et al. 1988:518):

(27) a. He didn’t reach Denver , let alone Chicago .

b. He didn’t reach any major city.
(28) a. I’m too tired to get up, let alone go running with you.
b. I’m too tired to do any chores.

However, unlike these polarity items, let alone is allowed in certain contexts where
other negative polarity items are disallowed (Fillmore et al 1988:519; example
[29a] is attested):

(29) a. You’ve got enough material there for a whole s emes ter, let alone a week.
b. *You’ve got enough material for any semester.

The semantics as well as the syntax of let alone is complex and not entirely pre-
dictable from more general rules of semantic interpretation from syntactic struc-
ture. As mentioned above, the let alone construction has at least one paired focus
(e.g. the pair semester and week in [29a]). The interpretation of a let alone sen-
tence requires the following steps. First the interpreter must recognize or construct
a semantic proposition in the fragmentary second conjunct that is parallel to the
proposition in the full first conjunct. Second, the interpreter must recognize or
construct a semantic scale underlying the elements in the propositions. This is not
always easy. For instance, the scale for (18a) may have to do with the assumed de-
gree of distastefulness of shrimp versus squid, or it may have to do with the relative
cost of shrimp versus squid (and Fred’s stinginess; Fillmore et al. 1988:524–25).
More specifically, the interpreter must perform the following semantic opera-
tions. The interpreter must construct a scalar model, which ranks propositions
on a scale – for example, the scale of distastefulness of eating seafood or the cost
thereof. The propositions in the two conjuncts must be from the same scalar model –
in this case, ‘Fred not eat shrimp’ and ‘Fred not eat squid.’ The two propositions
are of the same polarity (in this case, negative). Finally, the initial, full conjunct
denotes the proposition that is stronger or more informative on the scale – Fred not
eating shrimp is more informative than Fred not eating squid, on the assumption
that people who would eat squid would also eat shrimp but not vice versa. This
semantic analysis can be generalized to multiple paired focus versions of let alone
(see Fillmore et al. 1988 for details). This whole semantic apparatus is required for
the interpretation of the let alone construction, and is not necessary (as a whole)
for other constructions.
Finally, there is a specific pragmatic context in which the utterance of a let alone
construction is felicitous (Fillmore et al. 1988:532). First, the discourse context
is one such that the weaker (less informative) proposition, that is, the underlying
240 Cognitive approaches to grammatical form

proposition of the fragmentary second conjunct, is at issue – for example, the issue
of whether or not Fred eats squid. The weaker proposition accepts or rejects this
context – in this case, Fred doesn’t eat squid rejects it. But simply uttering the less
informative proposition is not cooperative since the speaker knows that the strong
proposition represented by the initial conjunct is true. So the speaker utters the let
alone sentence. Fillmore et al. note that let alone is similar pragmatically to other
conjunctions allowing sentence fragments, such as those illustrated in (21b–c)
above. However, some of these conjunctions present the stronger proposition in
the second, fragmentary conjunct, unlike let alone:

(30) a. He didn’t make colonel, let alone general.

b. He didn’t make general; in fact, he didn’t even make colonel.

The preceding discussion has presented some of the evidence that the let alone
construction has its own syntactic, semantic and pragmatic properties that cannot be
predicted from more general rules of syntax, semantics and pragmatics. A number
of other studies done in the emerging framework of construction grammar demon-
strate that other constructions also have unique syntactic, semantic and pragmatic
properties. A reading of these studies gives rise to two general observations.
First, the construction on which attention is focused by the researcher(s) turns
out to be just one of a family of related constructions. For example, the let alone
construction turns out to be just one of a family of coordinate constructions that
allow certain kinds of sentence fragments in the second conjunct, two of which
were illustrated in (21b–c). The let alone construction also turns out to be just
one of a family of paired focus constructions, two of which were illustrated in
(23b–c). Paired focus constructions are in turn related to a family of single focus
constructions. The phrase let alone is itself related to other negative polarity items,
and let alone is also related to a number of items that require a scalar model for
their semantic interpretation, such as even, almost, few and merely (Fillmore et al.
1988:530).
Likewise, Lakoff’s seminal study of the There-construction, as in There’s a
fox in the garden, uncovered a large family of related constructions with slightly
different syntactic and semantic properties, which are illustrated in examples (31)–
(32) (see Lakoff 1987, Appendix 3 for the analysis of the differences among There-
constructions):

(31) Deictic There-Constructions:

a. Central: There’s Harry with the red jacket on.
b. Perceptual: There goes the bell now!
c. Discourse: There’s a nice point to bring up in class.
d. Existence: There goes our last hope.
e. Activity Start: There goes Harry, meditating again.
 From idioms to construction grammar 241

f. Delivery: Here’s your pizza, piping hot!

g. Paragon: Now there was a real ballplayer!
h. Exasperation: There goes Harry again, making a fool of himself.
i. Narrative Focus: There I was in the middle of the jungle . . .
j. New Enterprise: Here I go, off to Africa.
k. Presentational: There on that hill will be built by the alumni of this university
a ping-pong facility second to none.

(32) Existential There-Constructions

a. Central: There’s a fox in the garden.
b. Strange [Event]: There’s a man been shot.
c. Ontological: There is a Santa Claus.
d. Presentational: Suddenly there burst into the room an SS officer holding a
machine gun.

Michaelis and Lambrecht’s (1996) study of Nominal Extraposition, illustrated in

(33a), reveals a family of related exclamative constructions, illustrated in (33b–e):

(33) a. It’s amazing the amount I spent!

b. I can’t believe the amount I spent!
c. The amount I spent!
d. I can’t b eli eve how much I spent!
e. It’s inc redi ble how much I spent !

The Nominal Extraposition construction in (33a) is characterized by extraposition

of the NP and a metonymic interpretation of the extraposed NP as referring to a
scale, unique to exclamatives (the exclamative character of the sentence is due to
the assertion of an excessive degree on the scale). The constructions in (33b–c)
share the metonymic interpretation of (33a), but the NP is not extraposed (33b)
or is an independent utterance on its own (33c). The exclamative constructions in
(33d–e) directly express the scale (how much), unlike (33a–c). The construction in
(33e) extraposes the degree expression, not unlike the extraposed (33a), while the
construction in (33d) does not. All five constructions in this family are distinguished
by the fact that they are a distinct speech act (expressed by the simple present tense),
and they assert an affective stance, namely contravention of expectation (this fact
restricts the main clause predicate to a gradable, contrary to expectation assertion).
The second observation upon surveying these studies of particular construc-
tions follows from the first. The number and variety of constructions uncovered
in these studies imply that speakers possess an extraordinary range of specialized
syntactic knowledge that goes beyond general rules of syntax and semantic in-
terpretation on the one hand, and a list of substantive idioms on the other. The
detailed analysis of such constructions is not the exclusive preserve of construc-
tion grammarians. Linguists working in a variety of approaches to syntax and
semantics have examined schematic idioms/constructions and demonstrated that
242 Cognitive approaches to grammatical form

they represent rule-governed and productive linguistic behavior, albeit limited to

the family of constructions analyzed.
One of the linguistic schools that calls itself ‘functionalist,’ which we will
call ‘autonomous functionalist’ (cf. Croft 1995:496–99), identifies constructions
that possess a specific discourse-functional or information-structural value. For
instance, Prince (1978) argues that the constructions known as It-cleft (illustrated
in [34]) and WH-cleft (35) each have their own discourse-functional value (Prince
1978:885):

(34) It is against pardoning these that many protest (Philadelphia Inquirer, February
6, 1977)
(35) What you are saying is that the President was involved (Haldeman, Watergate
tapes)

Prince notes that WH-clefts and It-clefts differ syntactically, in that the former
allow clefted adverbs or prepositional phrases as well as clefted noun phrases,
and the latter commonly allow verb phrases or sentences as clefted items (Prince
1978:884). Discourse-functionally, WH-clefts can be used when the information
in the subordinate clause is in the hearer’s consciousness (Prince 1978:894). In
contrast, Prince identifies at least two distinct ‘sub-senses’ for It-clefts, illustrated
in (36)–(37) (1978:896, 898):

(36) So I learned to sew books. They’re really good books. It’s just the covers that
are rotten. (Bookbinder in S. Terkel, Working, 1974)
(37) It was just about 50 years ago that Henry Ford gave us the weekend. (Philadelphia
Bulletin, January 3, 1976)

Example (36) illustrates what Prince calls a stressed focus It-cleft. In the stressed
focus It-cleft, the subordinate that-clause is given but not assumed to be in the
hearer’s consciousness. The stressed focus It-cleft is interesting also in that it
has a phonological property associated with it: only the focused part (in small
capitals in [36]) has strong stress; the that-clause is weakly stressed. Example
(37) is an informative-presupposition It-cleft: the that-clause presents information
that is a general known fact, albeit not to the hearer and hence new to the hearer
(and therefore also not in the hearer’s consciousness). Informative-presupposition
It-clefts have a normal rather than weak stress on the that-clause. These examples
indicate that constructions may have phonological features associated with them
as well as syntactic, semantic and pragmatic/discourse features.
Birner and Ward (1998) analyze a wide range of preposing constructions, such
as Topicalization (illustrated in [38]; Birner and Ward 1998:51), postposing con-
structions, such as right-dislocation (as in [39]; Birner and Ward 1998:146) and
argument reversal constructions, such as inversion (as in [40]; Birner and Ward
 From idioms to construction grammar 243

1998:159):
(38) As members of a Gray Panthers committee, we went to Canada to learn, and learn
we did. (Philadelphia Inquirer, June 16, 1985)
(39) It’s very delicate, the lawn. You don’t want to over-water, really. (father in the
movie ‘Honey, I Shrunk the Kids’)
(40) Behind them, moving slowly and evenly, but keeping up, came Pa and Noah.
(J. Steinbeck, The Grapes of Wrath, 1939)

Birner and Ward argue that, although there are commonalities among the different
constructions with respect to the discourse status of the preposed element and
the status of the postposed elements for syntactically similar constructions, each
has its own unique discourse properties. In other words, the various preposing,
postposing and inversion structures they discuss must each be analyzed as distinct
grammatical constructions.
Wierzbicka has discussed the properties of several families of constructions in
various publications (see the papers collected in Wierzbicka 1980, 1988, as well as
the examples discussed here). For example, Wierzbicka argues that the schematic
idiom have a V and the related types give a V and take a V, illustrated in (41)–(43),
represent rule-governed constructions (Wierzbicka 1988:293, 338):
(41) a. have a drink
b. *have an eat
(42) a. give the rope a pull
b. *give the window an open
(43) a. take a look at
b. *take a look for

Wierzbicka argues that the item following the indefinite article is a verbal in-
finitive, not a noun, and hence differs from other have constructions that do take
a noun, or more generally a noun phrase. For example, the phrase have a cough
is nominal, in that one can also have a headache/have pneumonia and so on in
which the word is indubitably a noun (Wierzbicka 1988:295–96). In this respect,
the have a V construction is syntactically unique.
Semantically, Wierzbicka argues that have a V represents an action as limited
in time but not punctual, lacking an external goal, and repeatable, and is of benefit
to the agent/experiencer (1988:297–302). Wierzbicka argues that this semantic
characterization is still incomplete, since it provides necessary but not sufficient
conditions for the occurrence of verbs in this construction. Instead, she presents
ten subtypes of the have a V construction, just as Prince offers two subtypes of the
It-cleft. One of these types she describes as ‘aimless objectless action which could
cause one to feel good,’ exemplified by have a walk/swim/run/jog/lie-down and so
on. In this subclass, the verbs are intransitive but durative and atelic (Wierzbicka
244 Cognitive approaches to grammatical form

1988:303), hence the unacceptability of (44b–c):

(44) a. He had a walk.

b. *He had a walk to the post office.
c. *He had a get-up.

The verb cannot describe a purposeful action (one with an external goal), other
than a recreational activity, hence the unacceptability of (45b–c):

(45) a. He had a swim.

b. *He had a work.
c. *He had a pray.

One indicator of the conventional character of the interpretations of schematic

idioms/constructions, particularly pragmatic ones, is the lack of translatability of
the idiom. For example, it has been argued that the tautological statement in (46)
is inferred to mean something like ‘That’s the kind of unruly behavior you would
expect from boys’ on general pragmatic principles (Levinson 1983:125):

(46) Boys will be boys.

Wierzbicka (1987) points out that in fact, the literal translation of (46) in various
languages (see examples [47]–[50]) does not have the same pragmatic meaning
as (46). Instead, different constructions are used to obtain approximately the same
pragmatic force (Wierzbicka 1987:96–97). In examples (47)–(50), the (a) sentence
is the closest literal translation to (46), the (b) sentence the one with the closest
pragmatic meaning to that of (46), and the (c) sentence is the literal translation of
the (b) sentence:

(47) French:
a. *Les garçons sont les (des?) garçons.
b. ?Les garçons seront toujours les (des) garçons. [still questionable]
c. ‘Boys will always be boys’
(48) German:
a. *Knaben werden Knaben sein.
b. ?Knaben bleiben (immer) Knaben.
c. ‘Boys remain (always) boys’
(49) Russian:
a. *Mal’čiki budut mal’čiki.
b. (Čego ty xočeš’?) oni že mal’čiki.
c. ‘(What do you expect?) They are boys’
(50) Polish:
a. *Chl
opcy be
da
chl
opcy.
b. (Jednak) co Paryż to Paryż.
c. ‘(However) what (is) Paris this (is) Paris’
 From idioms to construction grammar 245

These examples indicate that the pragmatic interpretation in (46) is in fact con-
ventionally associated with the equational tautology in which it occurs in English.
In fact, Wierzbicka argues for several different equational tautological con-
structions in English, which cannot be substituted for each other (Wierzbicka
1987:104):

(51) Nabstract is Nabstract .

a. War is war.
b. *Kid is kid.
(52) Nplural are Nplural .
a. Kids are kids.
b. *Wars are wars.
(53) Nplural will be Nplural .
a. Boys will be boys.
b. *Wars will be wars.
(54) An N is an N.
a. A party is a party.
b. *A war is a war.
(55) The N is the N.
a. The law is the law.
b. *The war is the war.

Wierzbicka argues that the semantic interpretations for the constructions exem-
plified in (51)–(55) can be characterized as follows: a ‘sober’ attitude toward com-
plex human activities (51); tolerance for human nature (52–53), the future subtype
indicating ‘the willful and uncontrollable spontaneity’ of the human type (1987:
107); obligation with respect to a human role, activity or institution ([54]–[55];
Wierzbicka argues that [54] has other readings as well). These semantic differences
cannot be inferred either from general rules of semantic interpretation in English
or general rules of the pragmatics of communication.
Even in the generative grammatical tradition, which is the theory most closely
identified with the componential model, there have been studies of schematic id-
ioms, in particular by Jackendoff (1990, 1997; see also Akmajian 1984 and Lam-
brecht’s 1990 reanalysis in construction grammar terms). For example, Jackendoff
(1997) analyzes the ‘time’-away construction, illustrated in (56):

(56) Bill slept the afternoon away.

Syntactically, the noun-phrase after the intransitive verb acts like a direct object
complement, and normally cannot occur with a transitive verb ([57a–b]). In some
cases the ‘normal’ direct object can appear in a with phrase, which it cannot do in
an ordinary active construction ([57c–d]; Jackendoff 1997:535):
246 Cognitive approaches to grammatical form

(57) a. Fred drank the night away.

b. *Fred drank scotch the night away.
c. Fred drank the night away with a bottle of Jack Daniels.
d. *Fred drank with a bottle of Jack Daniels.

Semantically, the ‘time’-away construction appears to have the same interpre-

tation as the durative adverbial with for (Jackendoff 1997:536):

(58) Bill slept for the (whole) afternoon.

However, unlike the durative adverbial, the ‘time’-away construction requires a

volitional subject ([59a–b]) and an activity rather than a state ([60a–b]; Jackendoff
1997:537):

(59) a. The light flashed for two hours.

b. *The light flashed two hours away.
(60) a. Celia sat for two hours.
b. *Celia sat two hours away.

The particle away as an aspectual particle seems to have a meaning and behavior
similar to away in the ‘time’-away construction:

(61) Bill drank away.

(62) *Celia sat away.

But the particle away is atelic (unbounded), while the ‘time’-away construction is
telic, as indicated by the It take NPtime construction:

(63) a. *It took a month for Lois and Clark to finally get to dance away.
b. It took a month for Lois and Clark to finally get to dance two blissful hours
away.

These properties indicate the uniqueness of the ‘time’-away construction with

respect to general syntactic and semantic rules. Jackendoff further argues that the
‘time’-away construction’s properties cannot be predicted from the properties of
other semantically related constructions, such as the resultative construction (64)
and the way construction ([65]; see Jackendoff 1997):

(64) The river froze solid.

(65) Dora drank her way down the street.

Jackendoff weighs two analyses of the ‘time’-away construction, the

construction-based account of the construction grammarians, and an account in
which a lexical rule derives the relevant verb that governs this type of construc-
tion. Jackendoff concludes that the only substantive difference between the two
 From idioms to construction grammar 247

accounts is that if one wants to ‘preserve the assumption that the lexical verb’s argu-
ment structure always determines the argument structure of the VP’ (Jackendoff
1997:557), then one must commit oneself to the lexical rule analysis. Jackend-
off himself inclines to the constructional analysis for the ‘time’-away construc-
tion, since he believes that constructions are necessary in other contexts anyway
(ibid.).
Jackendoff’s inclination in his 1997 paper is another step away from the com-
ponential model of generative grammar toward a construction grammar model.
Jackendoff’s inclination is also a step toward the construction grammarian’s bolder
hypothesis. Since one must posit constructions in order to account for a substantial
part of a speaker’s grammatical knowledge, is it possible to generalize the concept
of construction to account for all of a speaker’s grammatical knowledge? The next
section presents construction grammar’s arguments for the bolder hypothesis.

9.4 From constructions to construction grammar

The preceding section presented a number of case studies that argue for
the need to posit constructions as a unit of syntactic representation. A construction
is a syntactic configuration, sometimes with one or more substantive items (e.g.
the words let alone, have a . . . and away) and sometimes not (as with the focus
constructions, the exclamative constructions and the resultative construction). A
construction also has its own semantic interpretation and sometimes its own prag-
matic meaning (as with the tautological constructions). Hence a construction as a
unit cuts across the componential model of grammatical knowledge. The existence
of constructions would require a revision to the componential model in (3) that we
may represent as in (66):

(66)
phonological component
constructions

linking rules
lexicon

syntactic component
linking rules
semantic component

Constructions, like the lexical items in the lexicon, are ‘vertical’ structures that
combine syntactic, semantic and even phonological information (for the specific
words in a construction, as well as any unique prosodic features that may be
associated with a construction). As more and more constructions are discovered
and analyzed, construction grammarians came to argue that, in fact, grammatical
248 Cognitive approaches to grammatical form

organization is entirely ‘vertical’ (indeed, this approach is already suggested in

Fillmore et al. 1988).
We begin with the syntactic structure of constructions. In §9.3, constructional
analyses were proposed for schematic idioms. Schematic idioms were defined in
§9.2 as idioms in which some element or elements are lexically open, indicated by
a category label as in have a V. Schematic idioms were contrasted with substantive
idioms such as It takes one to know one, in which there are no lexically open ele-
ments. Hence, substantive idioms can be listed in the lexicon without substantially
altering the basic principles of the componential model.
But Fillmore et al. observe in a footnote that there is in fact a continuum from
substantive to schematic (1988:505, n. 3). Although we described all idioms with
any lexically open elements as schematic idioms in §9.2, in fact schematic id-
ioms vary considerably in their schematicity. Some schematic idioms such as the
verb-phrase idiom kick the bucket are fixed except for grammatical inflectional
categories:

(67) a. Jake kicked the bucket.

b. Jake’s gonna kick the bucket. [etc.]

Other schematic idioms have one or more open argument slots as well as inflec-
tional flexibility, such as give NP the lowdown ‘tell NP the news’:

(68) a. I gave/I’ll give him the lowdown.

b. He gave/He’ll give Janet the lowdown. [etc.]

Still other schematic idioms have open classes for all ‘content’ words, leaving
just a salient form such as the connective let alone as a substantive element:

(69) a. She gave me more candy than I could carry, let alone eat.
b. Only a linguist would buy that book, let alone read it.

Finally, a constructional analysis has been proposed for some schematic id-
ioms in which all elements are lexically open, such as the resultative construction
(Goldberg 1995:181; attested examples):

(70) a. This nice man probably just wanted Mother to . . . kiss him unconscious.
(D. Shields, Dead Tongues, 1989)
b. I had brushed my hair very smooth. (C. Brontë, Jane Eyre, 1847)

Yet the resultative construction has no lexically specific element. It can be de-
scribed only by a syntactic structure, in this case [NP Verb NP XP], with a unique
specialized semantic interpretation.
It is a very short step from analyzing the resultative construction as a construc-
tion to analyzing all the syntactic rules of a language as constructions (Fillmore
 From idioms to construction grammar 249

et al. 1988:501, 534; Langacker 1999:19). After all, a syntactic rule such as
VP → V NP describes a completely schematic construction [V NP], and the seman-
tic interpretation rule that maps the syntactic structure to its corresponding semantic
structure is unique to that schematic construction. Indeed, Goldberg suggests that
there is a transitive construction just as there are more specialized schematic syn-
tactic constructions such as the resultative construction (Goldberg 1995:116–19).
Reanalyzing general syntactic rules as the broadest, most schematic constructions
of a language is just the other end of the substantive- schematic continuum for
idioms/constructions.
Turning to semantic interpretation, one can also argue that constructions and
compositional semantic rules differ only in degree, not in kind. As we noted in
§9.2, Nunberg et al. (1994) argue that most idioms are idiomatically combining
expressions. In an idiomatically combining expression, the syntactic parts of the
idiom (e.g. spill and beans) can be identified with parts of the idiom’s semantic
interpretation (‘divulge’ and ‘information’). Nunberg et al. argue that idiomatically
combining expressions are not only semantically analyzable, but also semantically
compositional.
Nunberg et al. observe that idiomatically combining expressions are only the
extreme end of a continuum of conventionality in semantic composition. The other
end of the continuum is represented by selectional restrictions. Selectional re-
strictions are restrictions on possible combinations of words which are determined
only by the semantics of the concepts denoted by the word. For example, the re-
strictions on the use of mud and car in (71)–(72) follow from the fact that mud is
a viscous substance and a car is a machine:

(71) a. Mud oozed onto the driveway.

b. ?*The car oozed onto the driveway.
(72) a. The car started.
b. ?*Mud started.

The restrictions on mud and car are not dependent on the conventional form in
which the concepts are expressed. If one used the word goo instead of mud or
automobile instead of car, the judgements in (71)–(72) would remain the same.
The combinations in (71a) and (72a) are semantically compositional: the meaning
of the whole can be predicted from the meaning of the parts.
An intermediate point on this continuum involves what are called collocations.
Collocations are combinations of words that are preferred over other combinations
that otherwise appear to be semantically equivalent. For example, Matthews argues
that toasted and roasted describe essentially the same process, but are restricted
in their acceptable combinations (Matthews 1981:5):
250 Cognitive approaches to grammatical form

(73) a. roasted meat

b. toasted bread
(74) a. ?*toasted meat
b. ?*roasted bread

Most linguists would analyze (73a–b) as semantically compositional as well. In

both cases, the meaning of the whole can be predicted from the meaning of the
parts. It is just that speakers of English conventionally use toasted with bread and
roasted with meat, but not the other way around. This convention does not affect
the semantic compositionality of the expressions in (73a–b).
Typically, collocations are expressions that can be interpreted more or less cor-
rectly out of context, but cannot be produced correctly if the conventional expres-
sion is not already known to the speech community (Nunberg, Sag and Wasow
1994:495). In other words, collocations are encoding idioms. For example, the
expressions in (75a) and (75b) are the American and British terms for the same
type of object; each is compositional to the speakers of that dialect, but a speaker
of the other dialect would not be able to know what conventional expression is
used to refer to that type of object:

(75) a. thumb tack (American English)

b. drawing pin (British English)

Nunberg et al. argue that exactly the same reasoning applies to idiomatically
combining expressions. Idiomatically combining expressions are largely fixed in
their words; any substitution leads to ungrammaticality, as in (76b–c) and (77b):

(76) a. Tom pulled strings to get the job.

b. *Tom pulled ropes to get the job.
c. *Tom grasped strings to get the job.
(77) a. She spilled the beans.
b. *She spilled the succotash.

However, given the meanings of the words in the idiomatically combining expres-
sion, the meaning of the whole expression is compositional:

By convention . . . strings [in pull strings] can be used metaphorically to refer to

personal connection when it is the object of pull, and pull can be used metaphori-
cally to refer to exploitation or exertion when its object is strings. (Nunberg et al.
1994:496)
When we hear spill the beans used to mean ‘divulge the information’, for ex-
ample, we can assume that spill denotes the relation of divulging and beans the
information that is divulged, even if we cannot say why beans should have been
used in this expression rather than succotash. This is not to say, of course, that
spill can have the meaning ‘divulge’ when it does not co-occur with the beans, or
 From idioms to construction grammar 251

that beans can have the meaning ‘information’ without spill. The availability of
these meanings for each constituent can be dependent on the presence of another
item without requiring that the meaning ‘divulge the information’ attach directly
to the entire VP. Rather it arises through a convention that assigns particular
meaning to its parts when they occur together. (Nunberg et al. 1994:497)

At first, Nunberg et al.’s analysis may look odd. To say that pull and strings
each have a meaning found only in pull strings, and that those meanings are
compositional in the idiomatically combining expression, seems ad hoc. The more
natural description is the traditional one, that the meaning of the idiomatically
combining expression is ‘noncompositional.’ In fact, it is sometimes said that
one of the strongest pieces of evidence for constructions as independent syntactic
objects is that there is some degree of ‘noncompositionality’ in the meaning of
the construction. But there is evidence that Nunberg et al.’s analysis is the right
one.
Some English words exist only in idiomatically combining expressions, such as
heed in pay heed. It makes sense to say that heed has a meaning, that is of course
found only in pay heed. It has been argued that heed is idiomatic, because it is
essentially synonymous with attention in pay attention, and yet does not behave
the same way (Radford 1988; see Nunberg et al. 1994:505):

(78) a. You can’t expect to have my attention/*heed all the time.

b. He’s a child who needs a lot of attention/*heed.

Nunberg et al. argue that heed does not in fact mean the same thing as attention
does, when attention is the object of pay (Nunberg et al. 1994:505):

(79) a. The children paid rapt attention/?*heed to the circus.

b. I pay close attention/?*heed to my clothes.
c. They paid attention/??*heed to my advice, but didn’t follow it.

The semantic differences are related to the difference between the verbs attend and
heed: ‘we clearly attend to much that we do not heed . . . one can take heed but not
attention, and . . . attention but not heed can wander’ (Nunberg et al. 1994:506).
In other words, heed in pay heed does have its own meaning even though it occurs
(as a noun) in only that combination. Hence, it is reasonable to assume that other
words have specialized meanings in idiomatically combining expressions, and that
those meanings are compositional.
Another important line of evidence for the compositionality of idiomatically
combining expressions is psycholinguistic. Speakers of English recognize the
meanings of words in idiomatically combining expressions, and recognize them
as figurative meanings, even though the figurative meanings are found only in the
idiomatically combining expressions (Gibbs 1990). These two pieces of evidence
252 Cognitive approaches to grammatical form

point to Nunberg et al.’s conclusion that ‘The dependency among the parts of
idiomatically combining expressions is thus fundamentally semantic in nature’
(Nunberg et al. 1994:505).
From a construction grammar perspective, Nunberg et al.’s analysis of idiomat-
ically combining expressions looks more natural. An idiomatically combining ex-
pression such as spill the beans is a construction. As a construction, it has unique
syntax: the verb must be spill and its object must be the beans. It also has a se-
mantic interpretation, namely ‘divulge information.’ All Nunberg et al. are saying
is that this construction has its own semantic interpretation rules, mapping spill
onto ‘divulge’ and the beans onto ‘information.’ The constructional analysis is
presented in the diagram in (80), using lowercase to describe form and uppercase
to describe meaning, boxes to represent the construction and its parts, and dotted
lines to indicate the syntax-semantics mapping (see §10.1):

(80)
spill the beans

DIVULGE INFORMATION

What Nunberg et al. have done is to dissociate conventionality from noncompo-

sitionality. Idiomatically combining expressions are not noncompositional. There
exist truly noncompositional expressions; these are idiomatic phrases such as saw
logs and kick the bucket. Idiomatically combining expressions differ from collo-
cations and ordinary expressions only in that the conventional way of expressing
the parts of its meaning are conventional and also relatively opaque, compared to
collocations and ordinary expressions.
Earlier analysts have assumed that an idiomatically combining expression is
‘noncompositional’ because the meaning of the whole is not predictable from
the meaning of the parts when those parts appear in other expressions than the
idiom. More precisely, idiomatically combining expressions have been treated
as ‘noncompositional’ because their meanings do not conform to the semantic
interpretation rules of regular syntactic expressions such as [Ve r b O b j e c t]VP
in the case of spill the beans. But spill the beans is compositional in the sense
that the parts of the syntactic expression can be mapped onto components of the
meaning of the idiom, as in (80). The way that spill the beans differs from regular
syntactic expressions is that there are rules of semantic interpretation associated
with just that construction that are not derivable from the [Ve r b O b j e c t]VP
pattern of which spill the beans is an instance.
 From idioms to construction grammar 253

Thus, the common perception that a particular construction must be represented

as an independent syntactic unit because it is ‘noncompositional,’ is technically
incorrect. Constructions other than idiomatic phrases are compositional, that is,
the meanings of the parts of the construction are combined to form the meaning of
the whole construction. The reason that they must be represented as independent
constructions is that semantic interpretation rules associated with the construction
are unique to that construction, and not derived from another more general syntactic
pattern, as construction grammarians carefully note (see, e.g., Goldberg 1995:13
and Michaelis and Lambrecht 1996:219).
Indeed, one can think of the general ‘compositional’ rules of semantic inter-
pretation as being semantic rules associated with general (schematic) syntactic
structures, just as specialized rules of semantic interpretation are associated with
syntactically specialized extragrammatical idioms. Nunberg et al.’s analysis of id-
iomatically combining expressions can easily be extended to the general rules of
semantic interpretation that link syntactic and semantic structures. Consider, for
example, the English predicate adjective construction, illustrated in (81), and its
semantic interpretation:
(81) Hannah is smart.

The English predicate adjective construction has the form [NP be Adj]. It differs
from the ordinary verbal construction in requiring the copula verb be. One can ana-
lyze the semantics of the predicate adjective construction as follows. The members
of the Adjective category have a meaning that requires them to be combined with
the copula be in order to be interpreted as ascribing a property to a referent (unlike
verbs). The copula be has a meaning that requires combination with a member
of the Adjective category in order to be interpreted as doing the job of ascribing
(a property) to the subject NP. This analysis is in fact essentially the semantic
analysis that Langacker argues for (Langacker 1987:214–22; 1991a:204–5). In
Langacker’s terminology (see §3.5), Adjective symbolizes an atemporal relation,
and the copula be symbolizes a process that Adjective meanings must be combined
with in order to be predicated.
In like fashion, semantic interpretation rules can be provided for any schematic
construction describing the most general syntactic structures of the language. In
other words, all syntactic expressions, whatever their degree of schematicity, have
rules of semantic interpretation associated with them, although some substantive
idioms appear to inherit their semantic interpretation rules from more schematic
syntactic expressions such as [Verb Object] (see §10.2.1). Hence, the difference
between regular syntactic expressions and idiomatically combining expressions
is not that the former are ‘compositional’ and the latter are ‘noncompositional.’
Instead, the former’s rules of semantic composition are more general and the
254 Cognitive approaches to grammatical form

latter’s rules of semantic composition are more specialized. In semantics as well

as syntax, the concept of a construction can be generalized to encompass the full
range of grammatical knowledge of a speaker.
If syntax and semantics as a whole can be represented as constructions, what
about morphology and the lexicon? Morphology, like syntax, represents complex
grammatical units, made up of morphemes. From a structural point of view, the
only difference between morphology and syntax is that morphemes are bound
within a word, while words are morphologically free within a phrase or sentence.
Interestingly, analogs to almost all of the peculiar phenomena of idioms can be
found in morphology.
There are unfamiliar morphemes that exist only in single combinations, such as
cran- in cranberry (cf. kith and kin, pay heed). Such morphemes caused problems
for American structuralist analysis, because one had to assign a meaning (if any) to
the unfamiliar morpheme only in that word. This is, of course exactly the analysis
advocated by Nunberg et al. for their syntactic analogs.
There is also ‘extragrammatical’ morphology, that is, morphological patterns
that do not obey the general morphological rules of the language. The general rule
for plural formation in English is suffixation of an allomorph of -s to the noun stem.
The ablaut plurals of English such as feet, geese and so on are outside the general
plural formation rule. Arguably, the plural of brother-in-law, either brothers-in-
law or brother-in-laws, is also outside the general rule. Such examples are common
across languages. For example, the general rule for the position of agreement affixes
in K’iche’ Mayan is as a prefix immediately following the aspect prefix: x-at-w-
il-oh ‘I saw you [familiar].’ However, the second person formal morpheme is a
free word following the verb form, and hence is an ‘extragrammatical’ morpheme:
x-w-il alaq ‘I saw you [formal]’ (Mondloch 1978:27).
Morphological expressions can be placed on a continuum of schematicity. A
maximally substantive morphological expression is fully specified, as in book-s.
Partially schematic morphological expressions include book-NUMBER and NOUN -s.
Fully schematic morphological expressions include NOUN -NUMBER .
Finally, many words are what one might call ‘idiomatically combining words,’
where the meaning of a morpheme is specific to the stem it combines with (or a
subclass of stems). For example, -en is the plural of brother only when it refers to
a member of a religious community, and brother refers to a member of a religious
community when it is combined with -en.2 The derivational suffix -er refers to
the agent of the event denoted by the verb stem when that verb stem is in a class
including write, run and so on, but refers to the instrument if the verb stem is clip,
staple and the like, or the patient if the verb stem is fry, broil and so on. All of

2 We ignore here the fact that the plural stem for the plural in -en is distinct from the singular stem.
 From idioms to construction grammar 255

these observations suggest that in fact morphology is very much like syntax, and
that a constructional representation is motivated for morphology as well.
Lastly, the lexicon differs only in degree from constructions. Words in the lexi-
con are pairings of syntactic form (and phonological form) and meaning, includ-
ing pragmatic meaning. Constructions are also pairings of syntactic form (and
phonological form, for the substantive elements) and meaning, including prag-
matic meaning. The only difference is that constructions are complex, made up of
words and phrases, while words are syntactically simple. Some words are morpho-
logically complex, of course. But we have just argued that construction grammar
would analyze morphologically complex words as constructions whose parts are
morphologically bound. Morphologically simple words are atomic, that is, they
cannot be further divided into meaningful parts. But a word is again just the limiting
case of a construction (Fillmore et al. 1988:501).
The end point of this argument is one of the fundamental hypotheses of construc-
tion grammar: there is a uniform representation of all grammatical knowledge
in the speaker’s mind, in the form of generalized constructions. Table 9.2 com-
pares the different types of grammatical entities found in the componential model
of grammar and their analysis as constructions in construction grammar.
Table 9.2 The syntax-lexicon continuum

Construction type Traditional name Examples

Complex and (mostly) syntax [Sbj be- tns Verb -en by O bl]
schematic
Complex, substantive verb subcategorization [Sbj consume O bj]
frame
Complex and (mostly) idiom [kick-tns the bucket]
substantive
Complex but bound morphology [Noun-s], [Verb-tns]
Atomic and schematic syntactic category [Dem], [A dj]
Atomic and substantive word/lexicon [this], [green]

Syntactic rules (and the accompanying rules of semantic interpretation) are

schematic, complex constructions. The subcategorization frames required to han-
dle verbal syntactic behavior are schematic constructions with a substantive verb.
Idioms are complex and (at least partly) substantive constructions. Morphology
describes complex constructions, but constructions of bound morphemes. Words
in the lexicon are atomic substantive constructions, while syntactic categories
are schematic atomic constructions. In other words, grammatical knowledge rep-
resents a continuum on two dimensions, from the substantive to the schematic
and from the atomic to the complex. This continuum is widely referred to as the
256 Cognitive approaches to grammatical form

syntax-lexicon continuum. Thus, construction grammar conforms to Langacker’s

content requirement for a grammar: the only grammatical entities that are posited
in the theory are grammatical units and schematizations of those units (Langacker
1987:53–54).
The notion of a construction in construction grammar is much more general than
the traditional notion of a construction. In construction grammar, a construction can
be atomic or complex; it can have parts that are morphologically bound as well as
free; and any or all of the parts may be substantive or schematic. All constructions
in construction grammar, though, are pairings of a syntactic and morphological
(and, where relevant, phonological) form with a meaning, including pragmatic
meaning.
The model of grammatical knowledge in construction grammar is represented
in (82):
(82)
phonological properties
construction1

construction3
construction2

[etcetera]
lexicon

syntactic properties

semantic properties

A construction grammar consists of a large number of constructions of all types,

from schematic syntactic constructions to substantive lexical items. All of the
constructions possess properties of form (syntactic and phonological) and meaning
(semantic and pragmatic). All of these constructions are organized in a particular
way in a speaker’s mind. The next chapter describes how these generalizations are
elaborated in various theories of construction grammar.